Quick search damage positioning method based on full-focus algorithm
Through the time-domain and frequency-domain full-focus algorithm combined with evolutionary strategies and clustering algorithms, a fast search and damage positioning method based on full-focus algorithm was constructed, which solved the problem of inaccurate defect positioning in variable-thickness tubular structures, and achieved efficient and stable Damage testing.
Patent Information
- Application Number
- CN202410983409.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2024-11-15
AI Technical Summary
The existing full-focus fast search damage positioning method based on variable-thickness tubular structure has not yet been able to effectively construct the time-domain and frequency-domain scattering source point model, and design the screening and update strategies during the search process, resulting in insufficient accuracy and stability in defect positioning.
The scattering point model is constructed using the time-domain and frequency-domain full focus algorithm, combined with evolutionary strategies and clustering algorithms, and designed a rapid search and damage positioning method. Through scattering signal extraction, population screening, population update, cutoff criteria and defect positioning analysis, we can achieve the Efficient positioning of defects in variable thickness tubular structures.
The impact of frequency on imaging results is reduced, the accuracy and stability of defect positioning is improved, missed detection and misdetection are avoided, contrast is enhanced, and visual positioning of defects is ensured.
Smart Images

Figure CN118961899A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of ultrasonic nondestructive testing, and aims to develop a fast search damage location method based on a full focusing algorithm, which can be applied to variable thickness plate structures or tubular structures. Background Art
[0002] The boom is a variable thickness tubular structure, which is an important part of the lifting machinery. Its structural health status affects the service safety of the entire equipment. Since the boom is in a complex working environment such as high temperature, corrosion, overpressure, and overload all year round, it is very easy to produce defects such as cracks and hole damage. At the same time, it is very easy to produce processing defects during the production process due to the influence of processing accuracy and welding process. The existence of these defects will greatly reduce the structural stability and operation safety of the boom. During its service, it is very likely to cause serious accidents such as arm breakage due to the influence of the defect position. Therefore, the research on long-distance, high-efficiency, and visual fast search and damage location methods is extremely important for the structural health monitoring of lifting machinery. The fast search damage location method based on the full focusing algorithm analyzes the damage location of the variable thickness tubular structure by using the sensor array in the layout area. A fast search damage location method is developed by integrating the full focusing algorithm, evolutionary strategy, and clustering algorithm, which is of great significance to the promotion of intelligent non-destructive testing and structural health monitoring research in variable thickness tubular structures.
[0003] At present, the research on the full-focus rapid search damage location method based on variable thickness tubular structures is still in its preliminary stage, and many problems need to be solved, mainly including: 1. How to construct a scattering source point model based on time domain full focusing or frequency domain full focusing algorithm in variable thickness tubular structures; 2. How to design a search process based on evolutionary strategies and clustering algorithms; 3. How to design screening strategies and update strategies in the search process to maintain the diversity of retained individuals and ensure the stability of the search theoretical model. Summary of the invention
[0004] The purpose of the present invention is to propose a fast search damage location method based on a total focusing algorithm, which can be used to realize ultrasonic guided wave damage location detection of variable thickness tubular structures.
[0005] Based on the above issues, this patent constructs a scattering point model based on time domain full focusing and frequency domain full focusing, and proposes a fast search damage location method based on the full focusing algorithm in combination with evolutionary strategy and clustering algorithm to achieve effective location detection of defects in large variable thickness tubular structures.
[0006] The fast search damage location method of the full focusing algorithm includes a fast search algorithm based on time domain full focusing and a fast search algorithm based on frequency domain full focusing. The fast search algorithm based on time domain full focusing constructs a scatterer function model using equations (1) and (2) to calculate the pixel value of each discrete grid in the detection area.TFM and P SCF They are the pixel values calculated by the time domain total focusing algorithm TFM and the symbol coherence factor SCF algorithm respectively. In the formula, s ij ——time domain signal; b ij ——It is ij(i≠j) The polarity symbol of τ is the transit time of the wave packet from the exciter (i) to the receiver (j) through the discrete grid (x, y); N is the number of exciters and receivers; γ=1 / N 2 ——Scaling factor, set to 1 here.
[0007] Figure 1 is a schematic diagram of the frequency domain full focusing imaging algorithm. Figure 1 It can be seen that the imaging principle of the frequency domain full focusing algorithm is as follows: Assume that the wave field at any pixel point (x, z) at time t is p(x, m, z, t), where m represents the emission of the mth array element. Step 1: Transmit wave field p t (x,m,z=0,t) and the received wave field p(x,m,z=0,t) are subjected to two-dimensional FT changes in the x and t directions to obtain the spectrum p t (k x ,m,z=0,ω) and p(k x ,m,z=0,ω), where k x and ω are the wave number vector component and instantaneous angular frequency in the x direction, respectively. Step 2: Use the phase shift factor exp(-jk z z) and exp(jk z z) to perform wavefield extrapolation, and obtain the wavefield at depth z = 0 and extrapolate it to the wavefield p at depth z = z t (k x ,m,z,ω) and p(k x ,m,z,ω); realize the phase migration process of the wave field, where c1 is the wave velocity; Step 3: For the wave field p t (k x ,m,z,ω) and p(k x ,m,z,ω) to make a two-dimensional IFT change, and the imaging result of the scatterer can be obtained by conjugation. corr (x,m,z,ω). Step 4: Obtain the reconstruction result of the entire imaging area through formula (4).
[0008] The fast search algorithm based on frequency domain full focusing reconstructs the pixel value of each discrete grid in the imaging area using equation (4). Where, p(x,m,z,t) is obtained by synchronous extrapolation of the transmitted wavefield and the received wavefield; I' TFM ——Reconstruction results of the entire imaging area.
[0009] The fast search damage location method based on the full focusing algorithm includes five steps, namely, scattering signal extraction, population screening, population update, cutoff criterion and defect location analysis.
[0010] Step 1: Scattering signal extraction. Extract the scattering signals of defects at different frequencies and the corresponding group velocities at different wall thicknesses, and perform composite analysis on the scattering signals at different frequencies to improve the accuracy of defect positioning and reduce the impact of frequency on imaging results;
[0011] Step 2: Population screening. Extract the individual related pixel values p in the detection area v , and perform population screening on all pixel value points. When the fast search algorithm based on time domain full focusing is adopted, the pixel value p v By adding equation (1) and equation (2), when the fast search algorithm based on frequency domain full focusing is adopted, the pixel value p v It is obtained by formula (4). Only the grid where the individual is located is analyzed to reduce the execution time of the full-focus search algorithm. First, all pixel value points are clustered based on the K-means clustering algorithm. Before clustering, the elbow rule is used to evaluate the optimal number of clusters K. Then, according to the maximum value p in each cluster, v Calculate the adaptive individual screening threshold k p Considering the diversity characteristics in the detection, the screening threshold is set to p v 0.6 times of k. Values greater than 0.6 may cause defects to appear biased. However, smaller values may cause interfering individuals to be retained in the analysis, reducing image contrast. p It is expressed by formula (5): k p =floor(0.6×max(p v )) (5) In the formula, floor represents the function that rounds the data to the nearest integer.
[0012] Keep its related pixel value p v Greater than k p As the parent generation, extract the cluster center and add it to the parent generation for subsequent analysis. If the number of retained individuals is greater than n k , then randomly select n kFinally, the mean of the pixel values of the parent generation is calculated according to formula (6): And use it as the population fitness ε of the current generation g . Where n p ——The number of parents of the current generation; m – the subscript number of the parent number.
[0013] Step 3: Population update. Considering the diversity characteristics in detection, an adaptive population update strategy is used to balance the number of offspring around defects. The parents are clustered into K groups based on the K-means clustering algorithm, and the maximum number of released individuals n is set. f , detection pair correlation ratio r p and clustering coefficient r cluster Calculate the number of released offspring n r . r cluster × p ×n f The value of may be close to zero. Therefore, in order to generate enough offspring, the minimum number of released individuals is set to 4. r It is expressed by formula (7): r p =p v / max(p v ) (9) In the formula, r cluster Based on cluster size Calculated clustering coefficient; r p ——Detection pair correlation ratio.
[0014] Secondly, according to the calculated release quantity n r and the predefined maximum release distance r f Update the genes of all parents to generate offspring. It is expressed by formula (10): p g,c =round(p g,p +sign(rand(1,n r )-0.5)×r f ×rand(1,n r )) (10) In the formula, p g,c - offspring; p g,p ——Father generation.
[0015] According to formula (10), n r The difference between the offspring and the parent is rf ×rand(1,n r ), the release direction of the offspring is determined by sign(rand(1,n r ),-0.5)determined.
[0016] Finally, the next generation of analysis individuals is generated by combining offspring and parents. Return to population screening. In this process, individuals in larger clusters will release a small number of offspring; conversely, more offspring will be generated. This will help generate almost the same number of offspring around different defects to reduce the impact of diversity characteristics on detection.
[0017] Step 4: Cutoff criterion. In order to detect different frequency components of ultrasonic guided wave signals, three cutoff parameters are defined in the cutoff criterion, including population fitness ε g , the number of evolution generations g and the number of frequency components n c First, if the evolutionary number is equal to g m or g ≥3 and ε g The gradient change is less than ε t , then save the parent. If the scattered signals at all frequencies have been analyzed, perform defect location analysis; otherwise, update the frequency and return to population screening. Secondly, if ε g The gradient change is greater than ε t or g The gradient change is less than ε t If the number of generations is less than 3, the number of generations is updated and the population is updated. The third generation is defined as the critical generation, which ensures that enough individuals are retained in the test.
[0018] Step 5: Defect location analysis. Composite analysis technology is used to improve the imaging contrast of defects and reduce the impact of frequency on detection. All parent generations p stored in the cutoff criteria g,p The defect visualization analysis will be performed. Then, the defect location is determined by observing the full addition result of the individual related pixel values. The full addition formula is as follows: P(x,y)=∑p v,i (x,y) (11) Where, P is the pixel value at the discrete grid (x, y); p v,i ——The sign coherence factor of the parents, i is the number of parents.
[0019] The present invention has the following advantages: a fast search and damage location method based on a full-focus algorithm is proposed. Compared with the traditional full-focus algorithm, the proposed method weakens the lateral effect of the spatiotemporal mapping trajectory and avoids missed detection and false detection caused by pixel value differences. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 It is a schematic diagram of the frequency domain full focusing imaging algorithm.
[0021] Figure 2 It is a flow chart of a method for rapid search and damage location based on a total focusing algorithm of the present invention.
[0022] Figure 3 It is the cross-sectional dimension parameter of the U-shaped crane arm which is the application object of the present invention.
[0023] Figure 4 The finite element model of the planar dense array U-shaped crane arm and the corresponding unfolded plate model. (a) The finite element model of the double-defect U-shaped crane arm; (b) The double-defect unfolded plate model.
[0024] Figure 5 It is the full-addition positioning result based on the time-domain full-focus fast search algorithm.
[0025] Figure 6 It is the full-addition positioning result based on the frequency domain full-focus fast search algorithm. DETAILED DESCRIPTION
[0026] The present invention is described in detail below with reference to the accompanying drawings and embodiments.
[0027] Figure 4 The finite element model of the plane dense array U-shaped crane arm and the corresponding unfolded plate model. The structural dimensions of the finite element model of the U-shaped crane arm in the figure are Figure 3 The cross-sectional dimensions of the middle U-shaped boom are consistent, with an axial length of 1510 mm and a time step of 1×10 -8 s, the total duration is 8×10 -4 s. Except for the welds and defects, the mesh type is hexahedral mesh C3D8R (1mm×1mm×1mm). The mesh at the welds and defects is set to wedge-shaped tetrahedron C3D6 with a mesh size of 1mm. Figure (a) is the finite element model of a double-defect U-shaped crane arm. The lower right corner of the figure is the left view of the U-shaped crane arm. There is an axial black solid line at the top of the left view, which is the axisymmetry line of the U-shaped crane arm. The U-shaped crane arm is unfolded along the top axisymmetry line in the left view to obtain the corresponding unfolded plate figure (b). Figure (b) is a double-defect unfolded plate model. The red box in Figure (b) is a partial enlarged view of the planar dense detection array. The center point in the figure is the excitation point T with coordinates (755,896). The distance between two adjacent detection points is 10mm, and the detection points are R1 to R from top to bottom. 20 The markings do not include the center point T(755,896), from left to right, C1 to C 20Mark. The coordinates of the receiving points in the planar dense array expansion plate are shown in Table 1, and the coordinates of the excitation points and defects in the planar dense array expansion plate are shown in Table 2. The excitation signal is a Toneburst signal with a center frequency of 100kHz to 250kHz and a duration of 5 cycles modulated by a Hanning window. The load is loaded in the form of point excitation and acts at the center point T (755,896).
[0028] based on Figure 4 The finite element simulation model of the double-defect U-shaped crane arm is used, and the fast search damage location method based on the full focusing algorithm is used to locate and quantitatively analyze the defects H4 and H5 in the U-shaped crane arm.
[0029] Step 1: Extract the scattering signals of defects at different frequencies and the corresponding group velocities at different wall thicknesses. The group velocities of the A0 mode at 100kHz to 250kHz for a wall thickness of 4mm are v g4-100 、v g4-150 、v g4-200 and v g4-250 , the group velocity of A0 mode at 100kHz~250kHz under 5mm wall thickness is v g5-100 、v g5-150 、v g5-200 and v g5-250 .
[0030] Step 2: Extract the individual related pixel value p in the detection area v , and perform population screening on all pixel value points. When the fast search algorithm based on time domain full focusing is adopted, the pixel value p v By adding equation (1) and equation (2), when the fast search algorithm based on frequency domain full focusing is adopted, the pixel value p v Obtained by formula (4). According to the size of the U-shaped crane imaging area, the range of individual genes (x and y) is set to [1,1792]. The accuracy of individual genes is equal to the size of the discrete grid in the imaging area, that is, 1mm. The maximum evolutionary generation is 10, the maximum distance of released individuals is 5mm, the maximum number of released individuals is 5, the cutoff threshold is 0.2, and the number of clusters in population update is 4.
[0031] All pixel value points are clustered based on the K-means clustering algorithm. Before clustering, the elbow rule is used to evaluate the optimal number of clusters K, and then the maximum value p in each cluster is used to calculate the optimal number of clusters K. v Calculate the adaptive individual screening threshold k p . Keep its related pixel value p v Greater than k p As the parent generation, extract the cluster center and add it to the parent generation for subsequent analysis. If the number of retained individuals is greater than n k , then randomly select n kIndividuals are used as parents. And the population fitness ε of the current generation is calculated g .
[0032] Step 3: Population update. Considering the diversity characteristics in detection, an adaptive population update strategy is used to balance the number of offspring around defects. The parents are clustered into K groups based on the K-means clustering algorithm, and the maximum number of released individuals n is set. f , detection pair correlation ratio r p and clustering coefficient r cluster Calculate the number of released offspring n r . Secondly, according to the calculated release quantity n r and the predefined maximum release distance r f Update the genes of all parents and generate offspring based on formula (10). Finally, generate the next generation of analysis individuals by combining the offspring and the parents. Return to population screening. In this process, individuals in larger clusters will release a small number of offspring; conversely, more offspring will be generated. This will help generate almost the same number of offspring around different defects to reduce the impact of diversity characteristics on detection.
[0033] Step 4: Cutoff criterion. In order to detect different frequency components of ultrasonic guided wave signals, three cutoff parameters are defined in the cutoff criterion, including population fitness ε g , the number of evolution generations g and the number of frequency components n c First, if the evolutionary number is equal to g m or g ≥3 and ε g The gradient change is less than ε t , then save the parent. If the scattered signals at all frequencies have been analyzed, perform defect location analysis; otherwise, update the frequency and return to population screening. Secondly, if ε g The gradient change is greater than ε t or g The gradient change is less than ε t If the evolutionary generation is less than 3, update the generation and perform population update.
[0034] Step 5: Defect location analysis. Composite analysis technology is used to improve the imaging contrast of defects and reduce the impact of frequency on detection. All parent generations p stored in the cutoff criteria g,p The defect location is then determined by observing the full addition of the individual related pixel values.
[0035] The algorithm of the embodiment runs to the end, and obtains the full-addition positioning result based on the time domain full-focus fast search algorithm and the full-addition positioning result based on the frequency domain full-focus fast search algorithm. Figure 5 The full-addition positioning result based on the time-domain full-focus fast search algorithm is shown in Figure a) based on vg4-100 ~v g4-250 The full-addition positioning result shows that the relative error value of defect H4 is 3.84%, and the relative error value of defect H5 is 2.90%. The full-addition positioning result combines the high-precision retained individuals at different frequencies, making the positioning accuracy of the fusion result relatively high and reducing the influence of different frequencies on the defect positioning accuracy. g5-100 ~v g5-250 By comparing the relative error values at different frequencies, it can be seen that the positioning accuracy of the full-addition positioning result is relatively high, and the number of retained individuals distributed around the defect is similar to avoid false detection or missed detection. It is verified that the time domain full-focus search imaging algorithm can achieve high-precision positioning of double defects under different wall thicknesses. Figure 6 This is the full-addition positioning result based on the frequency domain full-focus fast search algorithm. The retained individuals in Figures a) and b) appear evenly around H4 and H5. Based on the relative error value, it can be seen that the defect is located with high precision. Through the full-addition fusion method of different group velocities, the individuals converge more accurately around the defect, improving the spatial resolution of defect imaging. The influence of group velocity changes on the positioning results is weakened. It is verified that the frequency domain full-focus search algorithm can achieve high-precision positioning for defects with different wall thicknesses. Table 1 Coordinates of receiving points in the planar dense array expansion board Table 2 Coordinates of excitation points and defects in the planar dense array unfolded plate
[0036] The above embodiments are only used to illustrate the use method of the present invention in damage location detection in a tubular U-shaped crane arm, rather than to limit it. It should be understood that the above embodiments are only specific implementations of the invention herein for the use method of rapid search for damage location based on a full focusing algorithm.
[0037] Finally, it should be noted that the present invention is not limited to the above-mentioned optional implementation modes. Any modifications, equivalent substitutions, improvements, etc. made by anyone within the design concept and implementation principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A fast search and damage location method based on a full focusing algorithm, characterized in that: It includes a fast search algorithm based on time domain full focusing and a fast search algorithm based on frequency domain full focusing. The fast search algorithm based on time domain full focusing constructs a scatterer function model with equations (1) and (2) to calculate the pixel value of each discrete grid in the detection area; P TFM and P SCF They are the pixel values calculated by the time domain total focusing algorithm TFM and the symbol coherence factor SCF algorithm respectively; In the formula, s ij is a time domain signal; b ij Yes ij(i≠j) The polarity symbol of τ is the transit time of the wave packet from the exciter (i) through the discrete grid (x, y) to the receiver (j); N is the number of exciters and receivers; γ=1 / N 2 is the scaling factor, set to 1; The fast search algorithm based on frequency domain full focusing reconstructs the pixel value of each discrete grid in the imaging area using equation (4); Where, p(x,m,z,t) is obtained by synchronous extrapolation of the transmitted wavefield and the received wavefield; I' TFM ——Reconstruction results of the entire imaging area; The fast search damage location method based on the full focusing algorithm includes five steps, namely, scattering signal extraction, population screening, population update, cutoff criterion and defect location analysis; Step 1: Scattering signal extraction: extract the scattering signals of defects at different frequencies and the corresponding group velocities at different wall thicknesses, and perform composite analysis on the scattering signals at different frequencies to improve the accuracy of defect positioning and reduce the influence of frequency on imaging results; Step 2: Population screening; extracting the individual related pixel values p in the detection area v , perform population screening on all pixel value points; Step 3: Population update: Considering the diversity characteristics in detection, an adaptive population update strategy is used to balance the number of offspring around the defect; Step 4: Cutoff criterion: To detect different frequency components of ultrasonic guided wave signals, three cutoff parameters are defined in the cutoff criterion, including population fitness ε g , the number of evolution generations g and the number of frequency components n c ; First, if the evolutionary number is equal to g m or g ≥ 3 and ε g The gradient change is less than ε t , then save the parent generation; if the scattered signals at all frequencies have been analyzed, perform defect location analysis; otherwise, update the frequency and return to population screening; secondly, if ε g The gradient change is greater than ε t or g The gradient change is less than ε t If the number of generations is less than 3, the number of generations is updated and the population is updated. The third generation is defined as the key generation, which ensures that enough individuals are retained in the test. Step 5: Defect location analysis; use composite analysis technology to improve the imaging contrast of defects and reduce the impact of frequency on detection; all parent generations p stored in the cutoff criterion g,p The defect visualization analysis will be performed; then, the defect location is determined by observing the full addition result of the individual related pixel values; the full addition formula is as follows: P(x,y)=∑p v,i (x,y) (11) Where, P is the pixel value at the discrete grid (x, y); p v,i ——The sign coherence factor of the parents, i is the number of parents.
2. According to claim 1, a fast search damage location method based on a total focusing algorithm is characterized in that: In step 2, when the fast search algorithm based on time domain full focusing is used, the pixel value p v By adding equation (1) and equation (2), when the fast search algorithm based on frequency domain full focusing is adopted, the pixel value p v It is obtained by formula (4); Analyze the grid where the individual is located, first cluster all pixel value points based on the K-means clustering algorithm, and use the elbow rule to evaluate the optimal number of clusters K before clustering, and then according to the maximum value p in each cluster v Calculate the adaptive individual screening threshold k p ; The screening threshold is set to p v 0.6 times of k p It is expressed by formula (5): k p =floor(0.6×max(p v )) (5) In the formula, floor——represents the function that rounds the data to the nearest integer; Keep its related pixel value p v Greater than k p As the parent generation, extract the cluster center and add it to the parent generation for subsequent analysis; if the number of retained individuals is greater than n k , then randomly select n k individuals as parents; finally, the mean of the pixel values of the parents is calculated according to formula (6): And use it as the population fitness ε of the current generation g ; Where n p ——The number of parents of the current generation; m – the subscript number of the parent number.
3. According to claim 1, a method for fast search and damage location based on a total focusing algorithm is characterized in that: In step 3, the parents are clustered into K groups based on the K-means clustering algorithm, and the maximum number of released individuals n is f , detection pair correlation ratio r p and clustering coefficient r cluster Calculate the number of released offspring n r ; r cluster × p ×n f The value of may be close to zero; therefore, in order to generate enough offspring, the minimum number of released individuals is set to 4; n r It is expressed by formula (7): r p =p v / max(p v ) (9) In the formula, r cluster Based on the cluster size Calculated clustering coefficient; r p is the detection pair correlation ratio; Secondly, according to the calculated release quantity n r and the predefined maximum release distance r f Update the genes of all parents to generate offspring; expressed by formula (10): p g,c =round(p g,p +sign(rand(1,n r )-0.5)×r f ×rand(1,n r )) (10) In the formula, p g,c - offspring; p g,p ——parent generation; According to formula (10), n r The difference between the offspring and the parent is r f ×rand(1,n r ), the release direction of the offspring is determined by sign(rand(1,n r ),-0.5) decision; generate the next generation of analysis individuals by combining offspring and parents; return population screening.
Citation Information
Cited By
Ultrasonic measurement method for winding angle and layering thickness of carbon fiber fully-wound gas cylinder
CN119756254A
