Interface identification method of titanium-steel clad plate based on physical reinforcement and dynamic programming

Abstract

This invention belongs to the field of ultrasonic nondestructive testing, specifically relating to an interface recognition method for titanium-steel composite plates based on physical enhancement and dynamic programming. The method first acquires a fully focused imaging image of the internal structure of the titanium-steel composite plate, and then generates an ultrasonic image of the interface after locating the interface region based on signal intensity. Next, it establishes interface reflection energy weights based on material parameters and local incidence conditions corresponding to candidate paths. Within the candidate search area of ​​the interface ultrasonic image, using candidate pixels as path states, it constructs a dynamic programming cumulative cost function containing data cost terms and path continuity constraint penalty terms, and solves it recursively column by column. Finally, it obtains the final interface contour through path backtracking and smoothing. This invention effectively overcomes the problems of interface breakage, jumps, and misjudgments that easily occur in existing methods under complex wavy interface conditions such as small wave distances and large amplitudes, significantly improving the continuity, stability, and accuracy of interface recognition results.

Description

Technical Field

[0001] This invention belongs to the field of ultrasonic nondestructive testing, specifically relating to an interface recognition method for titanium-steel composite plates based on physical reinforcement and dynamic programming, as well as the corresponding computer program product, storage medium, and ultrasonic nondestructive testing equipment. Background Technology

[0002] Titanium-steel composite plates are widely used in chemical engineering, marine engineering, and other fields. Their explosively welded interfaces typically exhibit periodic wavy undulations, and this interface morphology is crucial for structural quality evaluation. Currently, high-resolution ultrasonic phased array total focusing imaging (TFM) is widely used for non-destructive testing of the interior of such composite plates. However, accurately extracting complex wavy interfaces from imaging results remains a significant challenge. Traditional extraction methods typically search column-by-column for the location of the maximum ultrasonic response to determine the interface profile. This local extremum extraction method is suitable for interfaces with gentle slopes; however, under complex conditions such as small wavelengths and large amplitudes, large-angle interfaces lead to drastic physical attenuation of ultrasonic reflection energy, and this method lacks constraints on the continuity of the global path.

[0003] In summary, traditional methods are highly susceptible to interference from non-interface echoes, frequently resulting in problems such as interface transitions, breaks, and misjudgments, making it difficult to meet the requirements of high-precision detection. Summary of the Invention

[0004] To address the issues of existing interface identification methods for titanium-steel composite plates being susceptible to interference from non-interface echoes, resulting in interface jumps, fractures, and misjudgments, this invention provides an interface identification method for titanium-steel composite plates based on physical reinforcement and dynamic programming, along with its corresponding computer program product, storage medium, and ultrasonic non-destructive testing equipment.

[0005] This invention is achieved using the following technical solution: An interface recognition method for titanium-steel composite plates based on physical reinforcement and dynamic programming, comprising: Obtain full-focus imaging images of the internal structure of the titanium-steel composite plate.

[0006] High-energy response bands in the full-focus imaging image are identified based on a preset intensity threshold, and corresponding candidate search regions are constructed. The interface depth is set in conjunction with the specimen structure, and imaging data within the corresponding depth range is extracted from the candidate search regions to form the interface ultrasound image.

[0007] Solving the interface of titanium-steel composite plate at different incident angles based on the Zoeppritz equation. Longitudinal wave reflection coefficient Set the interface reflection energy weight for: ; and then in and Establish a mapping between them. Utilize The image response is weighted to form a data cost term, and a cumulative cost function is constructed that includes the data cost term and the path continuity constraint penalty term.

[0008] In the ultrasound image of the interface, each candidate pixel in the candidate search region is determined according to a preset sampling interval, with rows in the depth direction and columns in the horizontal direction. Dynamic planning is then performed on the paths between candidate pixels within the candidate search region, updating the cumulative cost and recording the paths for all columns of pixels from left to right through a column-by-column recursive solution.

[0009] The pixel with the lowest cumulative cost in the last column is selected as the endpoint of the interface path. Starting from the endpoint, the path is traced back column by column. The path with the lowest cumulative cost is smoothed and used as the identified interface outline.

[0010] As a further improvement of the present invention, the expression for the cumulative cost function D(r,c) is as follows: ; In the above formula, r and k represent row indices; c represents column indices; D(r,c) represents the cumulative cost of the r-th candidate pixel in the c-th column of the candidate search region; D(k,c-1) represents the cumulative cost of the k-th candidate pixel in the (c-1)-th column of the candidate search region; Ω(r) represents the set of candidate pixels in the previous column connected to the current candidate pixel r; C(r,c,k) represents the transfer cost from the previous candidate pixel k to the r-th candidate pixel in the c-th column; C d (r,c,k) represents the data cost of moving from candidate pixel k in the previous column to candidate pixel r in the c-th column; The penalty coefficient represents the continuity constraint; |rk| represents the depth index difference between two candidate pixels r and k in adjacent columns.

[0011] As a further improvement of the present invention, the depth jump variable between the candidate pixel points included in Ω(r) and the current candidate pixel point is limited to a preset maximum jump range. The corresponding discriminant function is: ; In the above formula, Take 8 pixels; nRows represents the total number of rows in the candidate search area.

[0012] As a further improvement to the present invention, the data cost C d The formula for calculating (r,c,k) is as follows: ; In the above formula, A(r,c) represents the normalized image intensity of the r-th candidate pixel in the c-th column of the candidate search region; Indicates the local angle of incidence Interface reflection energy weight, This represents the correction parameter; Δz is the image depth sampling interval, and Δx is the image horizontal sampling interval.

[0013] As a further improvement of the present invention, and Methods for establishing inter-mapping include: Within a preset angle range, the incident angle Perform discrete sampling. Solve for each discrete angle value. Corresponding longitudinal wave reflection coefficient Then calculate the corresponding interface reflection energy weights. Creating representations and A parameter lookup table for one-to-one mapping relationships.

[0014] As a further improvement of the present invention, in the recursive solution process of the path with the minimum cumulative cost, based on any local incident angle... The corresponding interface reflection energy weight can be obtained by querying the parameter lookup table. Then according to Calculate the data cost C for transferring from candidate pixel k in the previous column to the r-th candidate pixel in the c-th column. d (r,c,k); Wherein, when the parameter lookup table does not contain At that time, based on the two closest adjacent angle values and Calculated using interpolation Corresponding interface reflection energy weight .

[0015] As a further improvement of the present invention, the smoothing methods used to generate the interface contour include the rloess smoothing method based on local regression, the Savitzky-Golay smoothing method, the Gaussian filtering smoothing method, and the wavelet denoising smoothing method.

[0016] The present invention also includes a computer program product, which includes a computer program that, when executed by a processor, implements the interface recognition method for titanium-steel composite plates based on physical enhancement and dynamic programming as described above, performs interface recognition on the full-focus imaging image of the titanium-steel composite plate obtained by ultrasonic testing, and then extracts the interface contour from the interface ultrasonic image.

[0017] The present invention also includes a storage medium storing a computer program. When the computer program is executed by a processor, it implements the interface recognition method for titanium-steel composite plates based on physical enhancement and dynamic programming as described above, performs interface recognition on the full-focus imaging image of the titanium-steel composite plate obtained by ultrasonic testing, and then extracts the interface contour from the interface ultrasonic image.

[0018] The present invention also includes an ultrasonic nondestructive testing device, comprising an ultrasonic probe, a data processing module, and a display module. The data processing module includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it generates an interface ultrasonic image of the test piece based on the detection signal of the ultrasonic probe and displays it through the display module; it also implements the aforementioned interface recognition method for titanium-steel composite plates based on physical enhancement and dynamic programming, thereby extracting the interface contour from the interface ultrasonic image and visualizing it within the interface ultrasonic image.

[0019] The technical solution provided by this invention has the following beneficial effects: This invention introduces an interface reflection energy weight based on the Zoeppritz equation during the interface identification process of titanium-steel composite plates. This weight dynamically adapts to the changing patterns of reflected longitudinal wave response under different local incident conditions, thereby correcting the physical response attenuation of ultrasonic signals in large tilt regions of complex wavy interfaces and achieving higher-precision interface identification.

[0020] This invention transforms the task of identifying interface contours into a dynamic programming problem of ultrasonic propagation paths. By constructing a cumulative cost function that includes image response terms, physical weights, and path continuity constraint penalties, the interface recognition process simultaneously possesses local physical adaptability and global path optimization capabilities. This effectively suppresses the interface jumps, breaks, and misjudgments that are prone to occur in traditional column-by-column extreme value extraction methods.

[0021] Before implementing interface contour recognition based on dynamic programming strategy, this invention also constructs candidate search regions within the approximate location of the interface and constrains the depth jump variables between adjacent column path points. This can reduce the search space, reduce non-interface echo interference, and improve the stability and computational efficiency of interface extraction.

[0022] The interface recognition method for titanium-steel composite plates based on physical enhancement and dynamic programming provided by this invention combines adaptive weights characterizing interface reflection energy established based on the Zoeppritz equation with a global path optimization strategy of dynamic programming, which can significantly improve the stability and accuracy of the solution. This method is particularly suitable for interface contour extraction under complex wavy interface conditions such as small wave distance and large wave amplitude in titanium-steel composite plates, and has good engineering applicability and promotion value. Attached Figure Description

[0023] Figure 1 This is a flowchart of the interface recognition method for titanium-steel composite plates based on physical reinforcement and dynamic programming provided in Embodiment 1 of the present invention.

[0024] Figure 2 This is a simulation image of the interface contour in a medium-complexity wavy interface scenario.

[0025] Figure 3 For the traditional maximum value method Figure 2 A visualization of the interface contour recognition results in a given scene.

[0026] Figure 4 For the present invention solution in Figure 2 A visualization of the interface contour recognition results in a given scene.

[0027] Figure 5 This is a comparison chart of the interface contour identified by the traditional maximum value method and the actual interface contour.

[0028] Figure 6 This is a comparison diagram of the interface contour identified by the present invention and the actual interface contour.

[0029] Figure 7 This is a visualization of the interface contour recognition results of the present invention in a large-amplitude interface scenario.

[0030] Figure 8 This is a comparison diagram of the interface contour recognition results of the present invention and the traditional maximum value scheme in a large amplitude interface scenario.

[0031] Figure 9 This is a visualization of the interface contour recognition results of the present invention in a wavelet-distance interface scenario.

[0032] Figure 10 This is a comparison diagram of the interface contour recognition results of the present invention and the traditional maximum value scheme in a wavelet-small interface scenario.

[0033] Figure 11 This is an image of the interface morphology of the sample under a metallographic microscope during the testing experiment.

[0034] Figure 12 This is a visualization of the interface contour recognition results of the present invention for a real sample.

[0035] Figure 13 This is a comparison diagram of the interface contour recognition results of the present invention and the traditional maximum value scheme in a real sample scenario. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0037] Example 1

[0038] In ultrasonic imaging of titanium-steel composite panels, large-angle interfaces cause severe physical attenuation of ultrasonic reflection energy, and interface echoes generate interference, leading to identification errors such as interface jumps and breaks in traditional methods. This embodiment provides an interface identification method for titanium-steel composite panels based on physical enhancement and dynamic programming. This method first identifies regions near interfaces with high-energy responses as regions of interest (ROIs) from the original full-focus ultrasonic imaging data. Then, an evaluation function is designed to assess the continuity and physical rationality of candidate interface paths, denoted as the cumulative cost function. To address the ultrasonic reflection energy attenuation phenomenon, this embodiment introduces an adaptive weight that dynamically adjusts with the incident angle into the designed cumulative cost function, constructed using the Zoeppritz equation. To avoid identification trajectory breaks, this embodiment also sets continuity constraints in the path cost function to penalize jumps. Finally, combining the constructed path cost function, this embodiment dynamically plans the connected paths between pixels within the ROI, and then obtains the path with the minimum path cost through recursion; this path is the interface of the titanium-steel composite panel.

[0039] This embodiment combines the adaptive weights characterizing the interface reflection energy established based on the Zoeppritz equation with a global path optimization strategy of dynamic programming, so that the interface extraction process has both local physical response adaptability and global continuity constraint capability, thereby improving the stability and accuracy of interface contour extraction under complex wavy interface conditions.

[0040] Specifically, such as Figure 1 As shown, the interface recognition method for titanium-steel composite plates based on physical reinforcement and dynamic programming provided in this embodiment includes the following steps: I. Obtaining full-focus imaging images of the internal structure of titanium-steel composite plates In this embodiment, an ultrasonic testing device using a one-dimensional phased array ultrasonic transducer with N elements as the probe is taken as an example. The probe performs full matrix data acquisition on the exploded composite titanium-steel plate in a "single-transmit, full-receive" manner. The transducer element width and element spacing are fixed, the center frequency is preset according to the equipment parameters, and the system sampling frequency is obtained according to the testing equipment settings. Based on this phased array transducer, each time one element is excited to emit an ultrasonic pulse, all elements simultaneously receive the echo signal. After N excitations, a complete full matrix dataset (FMC) can be obtained, denoted as G: G = {U ( i,j,t)}; Where i represents the transmitting element number, j represents the receiving element number, and t represents time; U(i,j,t) represents the amplitude of the received signal; and .

[0041] To facilitate subsequent envelope extraction and imaging processing, the full matrix dataset G is transformed using a Hilbert transform to obtain an analytic signal dataset. : ; in: This represents the metadata after the Hilbert transformation, which satisfies: ; In the above formula, Represents the Hilbert transform operator. The imaginary unit is used to recover the envelope and phase information of a signal.

[0042] Next, an imaging plane is constructed within the area to be detected, establishing a two-dimensional Cartesian coordinate system P(x,z), where x is the lateral coordinate and z is the depth coordinate. The lateral range of the imaging plane is set according to the sample width, for example, uniform sampling within the range of [-20mm, 20mm], with a sampling interval Δx of 40 μm; the depth range is set according to the sample thickness and interface position, for example, sampling within the range of [2mm, 15mm], with a sampling interval Δz of 40 μm.

[0043] For any focal point poi (x,z) on the imaging plane, the ultrasonic propagation delay t is calculated based on its geometric distances to the i-th transmitting element and the j-th receiving element. ij (x,z): ; Where, d i (x,z) represents the distance from the focal point to the i-th transmitting element, d j (x,z) represents the distance from the focal point to the j-th receiving element, and c represents the equivalent sound velocity within the monitored object, which can be taken as 5300 m / s in the titanium-steel composite plate of this embodiment.

[0044] Based on the above time delay, the delay correction signal S of the focal point poi (x,z) is obtained. ij (x,z): ; The intensity value I of traditional all-focus imaging is obtained by coherently superimposing the delay correction signals from all channels. TFM (x,z): ; By traversing all focal points on the imaging plane, a fully focused imaging image of the internal structure of the titanium-steel composite plate can be obtained.

[0045] In this embodiment, to facilitate subsequent interface contour extraction, the absolute value of the imaging result can be taken and normalized to obtain the imaging intensity matrix I. abs (x,z); then the normalized imaging intensity matrix is ​​used as the full-focus imaging image and subsequently used for contour extraction.

[0046] II. Interface Region Localization and Interface Ultrasound Image Generation

[0047] In the full-focus imaging image of the specimen, the reflected echo from the metal interface exhibits a high-energy distribution. Observing along the depth direction, a high-energy response band can be seen near the titanium-steel interface. Therefore, this embodiment identifies the high-energy response band in the full-focus imaging image based on a preset intensity threshold and constructs a corresponding candidate search region. Furthermore, by setting the interface depth in conjunction with the specimen structure, imaging data within the corresponding depth range is extracted from the candidate search region to construct an ultrasonic image of the interface.

[0048] Specifically, in this embodiment, the depth range of the high-energy response band can be determined as the approximate location range of the interface by combining the sample's geometric dimensions and structural parameters; and candidate search areas of corresponding shapes (such as rectangular strips) can be selected. Then, candidate interface regions at different depths are selected as regions of interest (ROIs) based on different types of specimens. The typical depth range is 11~14mm, and the specific value can be determined according to the structural parameters of the titanium-steel composite plate. Imaging data of the corresponding row is extracted within this depth range, and the resulting ultrasonic image of the interface region is denoted as I. roi (x,z).

[0049] III. Design the interface reflection energy weights and construct the cumulative cost function

[0050] In this embodiment, the design of the cumulative cost function includes the following process: First, in order to incorporate the physical reflection properties of the material interface into the interface extraction process, this embodiment establishes the titanium-steel interface at different incident angles based on the Zoeppritz equation. Physical weight of reflected energy under certain conditions The Zoeppritz equation is used to solve for the interface of titanium-steel composite plates at different incident angles. Longitudinal wave reflection coefficient Then according to Set a random Dynamically changing interface reflection energy weight .

[0051] Specifically, for the interface of the titanium-steel composite plate, the steel layer density is set as follows: kg / m³, longitudinal wave velocity is m / s, shear wave velocity is m / s; titanium layer density is kg / m³, longitudinal wave velocity is m / s, shear wave velocity is m / s. Angle of incidence exist Uniform sampling is performed within the range, for example, taking 1000 angle points. For any sampling incident angle... First, the ray parameters of the longitudinal and transverse waves on both sides of the interface are calculated according to Snell's law, and the reflected transverse wave angle, transmitted longitudinal wave angle, and transmitted transverse wave angle are further determined. Then, the reflection coefficient vector under the longitudinal wave incident condition is obtained by numerically solving the Zoeppritz equation, where the first component is the longitudinal wave reflection coefficient. Based on the longitudinal wave reflection coefficient It can be seen that the interface reflection energy weight With the angle of incidence The following relationship is satisfied between them: .

[0052] In practical applications, this embodiment can also use the calculated interface reflection energy weight. The energy values ​​corresponding to the normal incidence conditions are normalized to obtain a parameter lookup table for the normalized interface reflection energy weights. This calculation process is repeated for all sampling angles to obtain a one-to-one correspondence between the incidence angle and the normalized interface reflection energy weights, which is then stored. In the subsequent dynamic programming path search process, the local incidence angle is calculated based on the local path geometry, and the corresponding local interface reflection energy physical weights are obtained from the parameter lookup table through interpolation.

[0053] In practical applications, to facilitate rapid data processing, this embodiment... and Establish a mapping between them so that it is possible to determine the incident angle based on any given angle. Get the corresponding value quickly The value of . Where, and Methods for establishing inter-mapping include: (1) Within the preset angle range, the incident angle Perform discrete sampling.

[0054] (2) Solve for each discrete angle value Corresponding longitudinal wave reflection coefficient Then calculate the corresponding interface reflection energy weights. .

[0055] (3) Creating a representation and A parameter lookup table for one-to-one mapping relationships.

[0056] Among them, the angle of incidence The interval for discrete sampling can be flexibly set according to the data density requirements of the parameter lookup table.

[0057] Secondly, this embodiment also utilizes adaptive interface reflection energy weights. The image response is weighted to form a data cost term, and a cumulative cost function is constructed that includes the data cost term and the path continuity constraint penalty term.

[0058] In detail, in this embodiment, the cumulative cost function reflects the process cost from the starting point to the previous relay node and the transfer cost from the previous relay node to the current point in the path. Specifically, the expression of the cumulative cost function D(r,c) constructed in this embodiment is as follows: ; In the above formula, r and k represent row indices; c represents column indices; D(r,c) represents the cumulative cost of the r-th candidate pixel in the c-th column of the candidate search region; D(k,c-1) represents the cumulative cost of the k-th candidate pixel in the (c-1)-th column of the candidate search region; Ω(r) represents the set of candidate pixels in the previous column connected to the current candidate pixel r; C(r,c,k) represents the transfer cost from the previous candidate pixel k to the r-th candidate pixel in the c-th column; C d (r,c,k) represents the data cost of moving from candidate pixel k in the previous column to candidate pixel r in the c-th column; The penalty coefficient represents the continuity constraint; |rk| represents the depth index difference between two candidate pixels r and k in adjacent columns.

[0059] In practical applications, the depth jump variables between the candidate pixels included in Ω(r) and the current candidate pixel are limited to a preset maximum jump range. The corresponding discriminant function is: ; In the above formula, Take 8 pixels; nRows represents the total number of rows in the candidate search area.

[0060] Data cost C d The formula for calculating (r,c,k) is as follows: ; In the above formula, A(r,c) represents the normalized image intensity of the r-th candidate pixel in the c-th column of the candidate search region; Indicates the local angle of incidence Interface reflection energy weight, This represents the correction parameter; Δz is the image depth sampling interval, and Δx is the image horizontal sampling interval.

[0061] IV. Dynamic Programming for the Path with Minimum Cumulative Cost

[0062] In interface ultrasound images, if each pixel is considered a path node, there are numerous feasible paths from the starting point on one side to the ending point on the other. Based on the design principle of the cumulative cost function, the cumulative cost of a path along the interface is always the lowest compared to other paths. Therefore, using the designed cumulative cost function, this embodiment transforms the interface trajectory recognition problem in interface ultrasound images into a dynamic programming problem for path nodes.

[0063] In this embodiment, the path planning process includes: determining each candidate pixel in the candidate search area according to a preset sampling interval, with rows in the depth direction and columns in the horizontal direction in the interface ultrasound image; dynamically planning the path between candidate pixels in the candidate search area, and updating the cumulative cost and recording the path for all columns of pixels from left to right by recursively solving column by column.

[0064] In practical applications, based on the established parameter lookup table, in the recursive solution process of the path with the minimum cumulative cost, this embodiment first determines the path based on any local incident angle. The corresponding interface reflection energy weight can be obtained by querying the parameter lookup table. Then according to Calculate the data cost C for transferring from candidate pixel k in the previous column to the r-th candidate pixel in the c-th column. d (r,c,k). Where the parameter lookup table does not contain... At that time, based on the two closest adjacent angle values and Calculated using interpolation Corresponding interface reflection energy weight .

[0065] In detail, this embodiment also includes interface ultrasound image I roi (x,z) is Gaussian smoothed to reduce random noise and enhance the continuity of the interface energy band. Then it is normalized to obtain the normalized interface ultrasound image I. nor (x, z) has a maximum value of 1. In the interface ultrasound image, a candidate search region of size nRows × nCols is constructed with the pixel index in the depth direction as the row and the pixel index in the horizontal direction as the column, where each pixel point corresponds to a candidate interface position. At the same time, a cumulative cost matrix D and a path record matrix Path are constructed for this region using dynamic programming.

[0066] Initially, set the D element to infinity and the Path element to 0.

[0067] For the first column of pixels, the initial cumulative cost is set as follows: ; Here, the row index r corresponds to the position of the candidate point in the depth direction.

[0068] For each column from column 2 to column nCols, the cumulative cost is updated recursively column by column. For any column c and the current row r, a set of candidate predecessors connected to it is selected from the previous column. This set consists of a preset maximum jump range. Confirmed. That is: ; In this embodiment, Take 8 pixels.

[0069] For candidate predecessor index Calculate the longitudinal step size of adjacent paths. The local incident angle corresponding to the local tilt angle is obtained by using the lateral step size Δx. : ; according to The corresponding physical weights are obtained from the parameter lookup table using interpolation. .

[0070] Define the data cost term C d (r,c,k) is: ; in, The normalized image intensity of the current pixel. To correct the parameters and avoid a denominator of zero, in this embodiment, a value of 100% can be used. .

[0071] To constrain the interface depth variation between adjacent columns, a path continuity constraint penalty term is introduced. ,in The continuity penalty coefficient is taken as [value missing] in this embodiment. Therefore, the total transfer cost is: ; For the current pixel (r,c), its cumulative cost D(r,c) is determined by the minimum sum of the cumulative cost of the previous column and the current transfer cost, that is: ; Next, we will implement the predecessor index of the minimum value. Recorded in the path matrix middle: ; By repeating the above recursive process, this embodiment can complete the cumulative cost update and path recording of all column pixels from left to right.

[0072] V. Path backtracking and smoothing to generate interface outline

[0073] The pixel with the lowest cumulative cost in the last column is selected as the endpoint of the interface path. Starting from the endpoint, the path is traced back column by column. The path with the lowest cumulative cost is smoothed and used as the identified interface outline.

[0074] Specifically, in this embodiment, after the recursive calculation of all columns is completed, the pixel with the minimum cumulative cost in the last column is selected as the endpoint of the interface path, and its row index is denoted as r. end ,satisfy: ; Use this index as the depth index of the path endpoint. Starting from the path endpoint, backtrack column by column according to the path record matrix Path. For any column c (from nRows-1 to 1), use the corresponding index r. c Find its predecessor index r in the path matrix. c-1 : ; By tracing back to the first column, we obtain the complete index sequence. .

[0075] By utilizing the correspondence between the aforementioned index sequence and the depth coordinate axis z of the interface ultrasound image, the original interface contour curve can be obtained. , where x c is the horizontal coordinate corresponding to column c.

[0076] To reduce the impact of local noise and discrete fluctuations on the interface curve, this embodiment also smooths the original interface contour curve. The smoothing methods used in generating the final interface contour in this embodiment include Rloess smoothing based on local regression, Savitzky-Golay smoothing, Gaussian filtering smoothing, and wavelet denoising smoothing. Specifically, when using the Rloess smoothing method based on local regression, a smoothing window of 30 sampling points can be selected to obtain the final interface curve z. final (x)

[0077] Through the above methods, the interface recognition method for titanium-steel composite plates based on physical enhancement and dynamic programming provided in this embodiment can achieve global optimization of the interface contour under various complex wavy interface conditions such as small wave distance and large wave amplitude, effectively improving the continuity, stability and accuracy of the interface extraction results.

[0078] Example 2

[0079] Based on the solution in Embodiment 1, this embodiment further provides a computer program product, a storage medium, and an ultrasonic nondestructive testing device.

[0080] The computer program product includes a computer program. When the computer program is executed by the processor, it implements the interface recognition method for titanium-steel composite plates based on physical enhancement and dynamic programming as in Example 1. It performs interface recognition on the full-focus imaging image of the titanium-steel composite plate obtained by ultrasonic testing, and then extracts the interface contour from the interface ultrasonic image.

[0081] The storage medium contains a computer program. When the computer program is executed by the processor, it implements the interface recognition method for titanium-steel composite plates based on physical enhancement and dynamic programming, as described above. It performs interface recognition on the full-focus imaging image of the titanium-steel composite plate obtained by ultrasonic testing, and then extracts the interface contour from the interface ultrasonic image.

[0082] The ultrasonic nondestructive testing equipment includes an ultrasonic probe, a data processing module, and a display module. The data processing module includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it generates an ultrasonic image of the interface of the test piece based on the detection signal from the ultrasonic probe, similar to existing equipment, and displays it through the display module. Furthermore, it introduces interface contour recognition and visualization functions. For example, the data processing module can implement the aforementioned interface recognition method for titanium-steel composite plates based on physical reinforcement and dynamic programming, extract the interface contour from the interface ultrasonic image, and then overlay it onto the interface ultrasonic image for display; thus, it visually demonstrates the interface extraction effect.

[0083] Furthermore, in practical applications, the ultrasonic nondestructive testing equipment of this embodiment can also export the spatial coordinate data of the extracted interface contour as needed, such as outputting the horizontal coordinates and corresponding depth values ​​of the interface in text or table form for subsequent analysis and evaluation.

[0084] Performance verification

[0085] To verify the practical effect of the interface recognition method for titanium-steel composite plates based on physical reinforcement and dynamic programming provided by this invention, technicians tested the performance of the proposed solution (denoted as PI-DP) in various scenarios and compared it with the traditional interface recognition scheme (denoted as the maximum value method). The experimental process included the following: I. Simulation Testing This experiment first creates experimental scenarios with interface contours of different complexities and scales through simulation experiments, and then uses different schemes to perform interface recognition in order to compare the performance of different schemes.

[0086] 1.1 Medium-complexity wavy interface profile

[0087] This experiment first verifies the applicability of the proposed solution under moderately complex wavy interface conditions. During the experiment, the root mean square error (RMSE) (in mm) was used as the evaluation metric to assess the similarity between the extracted interface contour and the actual interface contour. The corresponding calculation formula is as follows: ; In the above formula, M represents the number of sampling points participating in the statistics. To extract the interface outline in x m The depth value at that location; For a realistic interface outline in x m The depth value at that location.

[0088] Specifically, this embodiment uses COMSOL Multiphysics to establish... Figure 2 The finite element model of the titanium-steel composite plate is shown. The interface shape and position in the model can be described by the following equation (unit: mm): ; The interface has a wavelength of approximately 2.09 mm and an amplitude of 0.2 mm, and is located at a depth of 12 mm in the titanium steel plate.

[0089] Echo signals were acquired using a probe incorporating a phased array transducer, generating a corresponding full matrix dataset (FMC). The transducer element width was 0.5 mm, the element spacing was 0.6 mm, the center frequency was 5 MHz, and the system sampling frequency was determined based on the detection equipment settings. The acquired data was processed according to the method in Example 1 to achieve interface contour recognition. Furthermore, this embodiment also employs the traditional maximum value method for comparison, performing TFM imaging based on the experimental data of both methods to generate a visualized image of the interface contour, as shown below. Figure 3 and Figure 4 As shown.

[0090] analyze Figure 3 and Figure 4Data shows that the traditional maximum value method suffers from severe interface jumps and trough identification failures under small wave distance conditions, while the method of this invention successfully extracts the complete continuous wave interface path through path continuity constraints and physical weight correction by dynamic programming.

[0091] Furthermore, the RMSE of the two schemes was calculated within the same range, and the results are as follows. Figure 5 and Figure 6 As shown, the traditional maximum value method yields a result of 0.1704 mm, while the RMSE of the proposed method is 0.0670 mm. This demonstrates that the proposed method maintains high interface contour extraction accuracy and good path continuity even under complex interface conditions with small wavelet distances, fully verifying its superior performance in such scenarios.

[0092] 1.2 Large Amplitude Interface

[0093] This experiment further verifies the performance of the proposed solution in a scenario with large amplitude and complex interfaces. During the experiment, a finite element model of the titanium-steel composite plate was established using COMSOL Multiphysics, and its interface morphology is as follows: ; The interface has a wavelength of approximately 2.09 mm, an amplitude of 0.3 mm, and a maximum local tilt angle of approximately 45°, representing a typical scenario where traditional methods are prone to response attenuation and path breakage. Under these simulation conditions, echo signals were acquired using the same phased array transducer parameters as in the previous experiments, generating the corresponding full matrix dataset (FMC), and data processing was performed according to the same steps. Based on the experimental data, TFM imaging was performed, and a visual image of the interface contour was generated, as shown below. Figure 7 As shown.

[0094] Furthermore, the actual interface outline is drawn on the same image as the interface outline extracted by the present invention and the traditional maximum value method, resulting in the following: Figure 8 The comparison diagram is shown. Analysis of the data in the diagram shows that the traditional maximum value method exhibits significant interface response interruptions and path jumps in the high-angle region. In contrast, the solution of this invention effectively corrects the reflection energy attenuation by using the physical weight of reflection energy based on Zoeppritz, and successfully obtains a continuous and stable interface profile by combining the global continuity constraint of dynamic programming.

[0095] Furthermore, calculating the RMSE of the two schemes within the same range reveals that the RMSE of the maximum value method is 0.1858 mm, while the RMSE of the scheme of this invention is 0.0969 mm. This demonstrates that the scheme of this invention can maintain high interface contour extraction accuracy and good path continuity even under complex interface conditions with large amplitude, verifying its robustness in such complex scenarios.

[0096] 1.3 Wavelet-small interface

[0097] This experiment further verifies the performance of the invention in scenarios with complex interfaces at small wavelet distances. During the experiment, a finite element model of the titanium-steel composite plate was established using COMSOL Multiphysics, and the interface morphology is as follows: ; The interface has a wavelength of approximately 1.05 mm and an amplitude of 0.2 mm. Under these simulation conditions, echo signals were acquired using the same phased array transducer parameters as in the previous experiments, generating the corresponding full matrix dataset (FMC), and data processing was performed following the same steps. Based on the experimental data, TFM imaging was performed, and a visual image of the interface contour was generated, as shown below. Figure 9 As shown.

[0098] Furthermore, the actual interface outline is drawn on the same image as the interface outline extracted by the present invention and the traditional maximum value method, resulting in the following: Figure 10 The comparison diagram is shown. Analysis of the data in the diagram shows that the traditional maximum value method suffers from severe interface jumps and trough identification failures under small wave distance conditions, while the solution of this invention successfully extracts the complete continuous wave-like interface contour through path continuity constraints and physical weight correction by dynamic programming.

[0099] Furthermore, calculating the RMSE of the two schemes within the same range reveals that the maximum value method yields a result of 0.1704 mm, while the PI-DP method has an RMSE of 0.0670 mm. This demonstrates that the proposed solution maintains high interface contour extraction accuracy and good path continuity even under complex interface conditions with small wavelet distances, fully validating its superior performance in such complex scenarios.

[0100] II. Real Test

[0101] Based on the aforementioned simulation experiments, this experiment further employs the scheme of this invention to verify the actual explosive composite titanium-steel composite plate sample through phased array ultrasonic testing. By observing and analyzing the sample cross-section using a metallographic microscope, the true interface morphology was obtained as follows: Figure 11 As shown, its trajectory curve is approximately as follows: ; The interface has a wavelength of approximately 2.09 mm and an amplitude of 0.18 mm. The interface profile exhibits typical periodic wavy undulation characteristics found in practical engineering.

[0102] In this experiment, a 32-element phased array transducer with an element spacing of 0.6 mm and a center frequency of 5 MHz was used to acquire FMC data from the sample, and the data processing was performed according to the same scheme as in the previous experiment. Based on the experimental data, TFM imaging was performed to generate a visualized image of the interface contour, as shown below. Figure 12 As shown.

[0103] Furthermore, the actual interface outline is drawn on the same image as the interface outline extracted by the present invention and the traditional maximum value method, resulting in the following: Figure 13 The comparison diagram is shown. Analysis of the data in the diagram reveals that the traditional maximum value method exhibits local jumps in the interface undulation region, resulting in poor continuity of the extraction results. In contrast, the method proposed in this invention can extract a complete wavy interface profile that is highly consistent with the metallographic analysis results. This demonstrates that the method of this invention has good engineering applicability and accuracy stability in real engineering samples.

[0104] In summary, the above experiments verified the proposed solution under different scenarios, including medium-complexity wavy interfaces, large-amplitude complex interfaces, small-span complex interfaces, and real sample interfaces, and compared it with traditional solutions. The experimental results show that the proposed solution has good effectiveness and robustness under different complex wavy interface conditions and can be applied in practical engineering.

[0105] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An interface recognition method for titanium-steel composite plates based on physical reinforcement and dynamic programming, characterized in that, It includes: Acquire full-focus imaging images of the internal structure of titanium-steel composite plates; The high-energy response band in the full-focus imaging image is identified according to the preset intensity threshold, and the corresponding candidate search area is constructed; the interface depth is set in combination with the specimen structure, and the imaging data of the corresponding depth range is extracted from the candidate search area to form the interface ultrasound image. Solving the interface of titanium-steel composite plate at different incident angles based on the Zoeppritz equation. Longitudinal wave reflection coefficient Set the interface reflection energy weight for: ; and then in and Establish mappings between them; utilize The image response is weighted to form a data cost term, and a cumulative cost function containing the data cost term and the path continuity constraint penalty term is constructed. In the ultrasound image of the interface, each candidate pixel in the candidate search area is determined according to the preset sampling interval, with the depth direction as the row and the horizontal direction as the column. The path between the candidate pixels in the candidate search area is dynamically planned, and the cumulative cost update and path recording of all column pixels are completed from left to right by recursively solving column by column. The pixel with the lowest cumulative cost in the last column is selected as the endpoint of the interface path. Starting from the endpoint, the path is traced back column by column. The path with the lowest cumulative cost is smoothed and used as the identified interface outline.

2. The interface recognition method for titanium-steel composite plates based on physical reinforcement and dynamic programming as described in claim 1, characterized in that, The cumulative cost function D(r,c) is expressed as follows: ； In the above formula, r and k represent row indices; c represents column indices; D(r,c) represents the cumulative cost of the r-th candidate pixel in the c-th column of the candidate search region; D(k,c-1) represents the cumulative cost of the k-th candidate pixel in the (c-1)-th column of the candidate search region; Ω(r) represents the set of candidate pixels in the previous column connected to the current candidate pixel r; C(r,c,k) represents the transfer cost from the previous candidate pixel k to the r-th candidate pixel in the c-th column; C d (r,c,k) represents the data cost of moving from candidate pixel k in the previous column to candidate pixel r in the c-th column; The penalty coefficient represents the continuity constraint; |rk| represents the depth index difference between two candidate pixels r and k in adjacent columns.

3. The interface recognition method for titanium-steel composite plates based on physical reinforcement and dynamic programming as described in claim 2, characterized in that: The depth jump variables between the candidate pixels included in Ω(r) and the current candidate pixel are limited to a preset maximum jump range. The corresponding discriminant function is: ； In the above formula, Take 8 pixels; nRows represents the total number of rows in the candidate search area.

4. The interface recognition method for titanium-steel composite plates based on physical reinforcement and dynamic programming as described in claim 2, characterized in that, Data cost C d The formula for calculating (r,c,k) is as follows: ； In the above formula, A(r,c) represents the normalized image intensity of the r-th candidate pixel in the c-th column of the candidate search region; Indicates the local angle of incidence Interface reflection energy weight, This represents the correction parameter; Δz is the image depth sampling interval, and Δx is the image horizontal sampling interval.

5. The interface recognition method for titanium-steel composite plates based on physical reinforcement and dynamic programming as described in claim 4, characterized in that: and Methods for establishing inter-mapping include: Within a preset angle range, the incident angle Perform discrete sampling; Solve for each discrete angle value Corresponding longitudinal wave reflection coefficient Then calculate the corresponding interface reflection energy weights. ; Create representation and A parameter lookup table for one-to-one mapping relationships.

6. The interface recognition method for titanium-steel composite plates based on physical reinforcement and dynamic programming as described in claim 5, characterized in that: In the recursive solution process for the path with the minimum cumulative cost, based on any local incident angle... The corresponding interface reflection energy weights are obtained by querying the parameter lookup table. Then according to Calculate the data cost C for transferring from candidate pixel k in the previous column to the r-th candidate pixel in the c-th column. d (r,c,k); Wherein, the parameter lookup table does not contain At that time, based on the two closest adjacent angle values and Calculated using interpolation Corresponding interface reflection energy weight .

7. The interface recognition method for titanium-steel composite plates based on physical reinforcement and dynamic programming as described in claim 1, characterized in that: The smoothing methods used to generate the interface contour include Rloess smoothing based on local regression, Savitzky-Golay smoothing, Gaussian filtering smoothing, and wavelet denoising smoothing.

8. A computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the interface recognition method for titanium-steel composite plates based on physical enhancement and dynamic programming as described in any one of claims 1-7, performs interface recognition on the full-focus imaging image of the titanium-steel composite plate obtained by ultrasonic testing, and then extracts the interface contour from the interface ultrasonic image.

9. A storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the interface recognition method for titanium-steel composite plates based on physical enhancement and dynamic programming as described in any one of claims 1-7, performs interface recognition on the full-focus imaging image of the titanium-steel composite plate obtained by ultrasonic testing, and then extracts the interface contour from the interface ultrasonic image.

10. An ultrasonic nondestructive testing device, comprising an ultrasonic probe, a data processing module, and a display module, wherein the data processing module includes a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, on the one hand, it generates an interface ultrasound image of the test piece based on the detection signal of the ultrasound probe and displays it through the display module; on the other hand, it implements the interface recognition method for titanium-steel composite plates based on physical enhancement and dynamic programming as described in any one of claims 1-7, and then extracts the interface contour from the interface ultrasound image and visualizes it in the interface ultrasound image.

Images