A smart testing method and system for interlayer peel strength of copper-aluminum composite panels

By constructing a multi-layer hidden state model and the Viterbi algorithm, combined with relative position parameters and interlayer spacing templates, the problem of accurately determining multiple crack lines in the interlayer peel test of copper-aluminum composite plates was solved, achieving precise positioning of peel boundaries and stability of strength calculation.

CN123065952APending Publication Date: 2026-11-10GRAD NEW ENERGY MATERIALS (SHANDONG) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611336833.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-31
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

In existing tests on the interlayer peel strength of copper-aluminum composite panels, the multilayer intermetallic compound structure makes it difficult to accurately determine the true peel boundary due to multiple crack lines, resulting in deviations in test data.

Method used

By constructing a multi-layer hidden state model, the optimal peeling boundary trajectory is solved from multiple crack lines using the Viterbi algorithm. Combined with relative position parameters and interlayer spacing templates, image sequences and force value sequences are acquired and processed simultaneously. Sub-pixel positioning technology and a confidence evaluation mechanism are used to remove abnormal frames.

Benefits of technology

It improves the stability and anti-interference ability of peel length measurement, ensures the accuracy and reliability of peel strength calculation, and provides a basis for quality evaluation and process optimization of copper-aluminum composite panels.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123065952A_ABST
    Figure CN123065952A_ABST
Patent Text Reader

Abstract

This invention relates to the field of peel strength testing technology for copper-aluminum composite plates, specifically to an intelligent testing method and system for the interlayer peel strength of copper-aluminum composite plates. The method includes: simultaneously acquiring image sequences and force value sequences of the peeling process; obtaining the relative position parameters of each intermetallic compound layer at the interface based on the annealing process and converting them into a pixel spacing template; extracting crack lines from the images and calculating their vertical positions; constructing three layers based on the template and assigning crack lines to each layer; defining hidden states and observed values ​​based on the interlayer spacing template and crack lines; setting state transition probabilities and emission probabilities; using the Viterbi algorithm to solve for the optimal hidden state sequence; extracting the true peel boundary position and confidence level for each frame; and calculating the average peel strength after sub-pixel localization of the boundary position and its correspondence with the force value. This invention uses relative position parameters as prior constraints, combined with Viterbi global optimization, to accurately identify the true peel boundary from multiple crack lines, improving testing accuracy and stability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of peel strength testing technology for copper-aluminum composite panels, and more specifically, to an intelligent testing method and system for interlayer peel strength of copper-aluminum composite panels. Background Technology

[0002] Copper-aluminum composite plates are metal sheets formed by combining copper and aluminum layers through a rolling process. They combine the excellent electrical and thermal conductivity of copper with the lightweight advantages of aluminum and are widely used in new energy, heat dissipation and other fields.

[0003] To ensure the performance of copper-aluminum composite panels, it is essential to focus on testing the quality of the interface bonding. The interlayer peel strength is a key mechanical indicator that can effectively prevent failure risks such as delamination and cracking.

[0004] Existing peel strength tests are all conducted in accordance with the corresponding industry standards. During the operation, the ends of the composite board are first pre-peeled, and then the sample is clamped into the upper and lower clamps of the tensile testing machine. Peel load is applied at a constant rate, and force and displacement curves are collected simultaneously. Finally, the peel strength is calculated based on these curves. Some improved solutions are equipped with a machine vision system, which uses a camera to collect images of the entire peeling process and combines edge detection algorithms to identify the peeling front and then calculate the peel length.

[0005] However, the above-mentioned solutions have certain limitations: copper-aluminum composite plates generally require annealing to strengthen the interlayer metallurgical bond during preparation. This process generates multiple intermetallic compounds at the interface, forming a brittle multilayer interface structure. During peel testing, cracks will propagate between the compound layers, forming multiple parallel and misaligned crack lines. Existing edge detection methods usually assume that the peel boundary is a single continuous boundary. When multiple crack lines actually appear, it is difficult to distinguish the true peel boundary, resulting in inaccurate peel length calculations and deviations in test data.

[0006] Therefore, there is an urgent need for an intelligent testing method that can identify multiple crack lines and accurately determine the actual peeling boundary. Summary of the Invention

[0007] One objective of this invention is to provide an intelligent testing method for the interlaminar peel strength of copper-aluminum composite plates. This method constructs a multilayer hidden state model with relative position parameters as constraints and uses the Viterbi algorithm to solve for the optimal peel boundary trajectory from multiple crack lines, thereby solving the problems mentioned in the background art. In the interlayer peel test of copper-aluminum composite plates, the presence of multiple intermetallic compound structures at the interface leads to multiple crack lines during peeling, making it difficult to accurately determine the true peel boundary.

[0008] To achieve the above objective, the method includes the following steps: S1. Simultaneously acquire image sequences and force value sequences of the interlayer peeling process of copper-aluminum composite plates; Obtain the relative position parameters between the intermetallic compound layers at the interface of the copper-aluminum composite plate, convert the relative position parameters into pixel spacing, and obtain the interlayer spacing template. S2. Perform edge detection on each frame of the image sequence to extract crack lines and calculate the vertical position of each crack line; Based on the interlayer spacing template, an aluminum-side transition layer, an intermediate diffusion layer, and a copper-side reaction layer are constructed in each frame of the image sequence to correspond to the spatial distribution of crack lines. After sorting by vertical position, the crack lines are assigned to each layer. S3. Based on the interlayer spacing template and the crack lines of each layer, define hidden states and observation values ​​for each frame of image. The hidden states represent the layer and vertical position of the actual peeling boundary, and the observation values ​​represent the vertical positions of all crack lines in the current frame. The transition probability of the hidden state between adjacent frames is defined based on the interlayer spacing template, and the emission probability is defined based on the degree of matching between the vertical position of the crack line and the vertical position in the hidden state. Based on the transition probability and emission probability, the Viterbi algorithm is used to solve the optimal hidden state sequence, and the true stripping boundary position and corresponding confidence level of each frame are extracted from the optimal hidden state sequence. S4. Perform sub-pixel localization on the extracted true peeling boundary position, match the localized position with the synchronously acquired force value sequence, remove abnormal frames based on the extraction reliability, select the effective peeling segment, and calculate the average peeling intensity based on the effective peeling segment.

[0009] In the above technical solution, the core design concept is to introduce relative position parameters as prior information into the discrimination process. Without this constraint, relying solely on the position and clarity information of crack lines is insufficient to distinguish which interface layer each crack line belongs to. The algorithm can only perform simple sorting based on proximity or clarity, which can easily lead to incorrect selection of the true boundary when multiple crack lines appear intersecting. Therefore, this solution first obtains the known interlayer spacing based on the annealing process and defines three layers accordingly. Crack lines are then distributed to each layer according to their vertical positions, ensuring that subsequent discrimination is based on a clear spatial structure. Based on this, the Viterbi algorithm is introduced for global optimization of consecutive frames. This is because the appearance of crack lines in a single frame image exhibits randomness and incompleteness, making reliable judgments difficult based solely on the information of the current frame. By defining hidden states in two dimensions—layer and position—and utilizing the transition probability between adjacent frames to constrain the continuity of the boundary and the layer change pattern, while using the emission probability to measure the matching degree between the observed value and the hidden state, the optimal peeling boundary path is finally selected from multiple candidate trajectories. This design ensures that even if the observation information is incomplete due to missing crack lines or powder obscuring in a certain frame, the algorithm can still infer a reasonable boundary position based on historical information from previous and subsequent frames, thereby guaranteeing the continuity and stability of the test.

[0010] Based on this, the specific steps of sorting the crack lines by vertical position and then allocating them to each layer are as follows: the sorted crack lines are sequentially matched according to their vertical position, the vertical position difference between adjacent crack lines is calculated, and if the difference is within the allowable error range of the relative position parameter, it is taken as the interlayer boundary point. The crack lines before the boundary point are allocated to the current layer, and the crack lines after the boundary point continue to match the next interlayer spacing until all crack lines are allocated.

[0011] In another technical solution, the transition probability is set based on the type of stratigraphic change: When the layer remains unchanged, the transition probability is constrained by a Gaussian function of the position difference between adjacent frame boundaries; When the layer advances one layer towards the copper side, the transition probability is constrained by a Gaussian function of the position difference and the interlayer spacing deviation. When the layer retreats towards the aluminum side or jumps across layers, the transfer probability is zero; The emission probability is calculated based on a Gaussian function of the distance between the vertical position of the crack line and the vertical position of the hidden state, as well as a normalized value of the crack line sharpness.

[0012] In this technical solution, the spacing-matching-based layering method and the transition probability setting based on layer change type supplement the core discrimination logic from two dimensions: spatial allocation and temporal tracking. Without the spacing-matching-based layering method, crack lines can only be simply divided according to vertical position. When multiple crack lines appear in a layer or a layer has missing crack lines, the allocation result will deviate significantly from the actual layer relationship, causing subsequent algorithms to solve based on incorrect layer assignments, and true boundary discrimination cannot be guaranteed. Without the transition probability setting based on layer change type, the positional continuity constraint alone cannot distinguish between the two physical processes of layer invariance and layer advancement, nor can it constrain state transitions that violate the stripping rules, such as regression and cross-layer jumps. The Viterbi algorithm is prone to converging to the wrong path when solving multiple consecutive frames, especially in frames with missing crack lines or high image noise, where it is more likely to deviate from the true trajectory. Together, these two methods construct a complete discrimination mechanism from single-frame allocation to multi-frame tracking, effectively ensuring the accuracy and continuity of stripping boundary identification.

[0013] The second objective of this invention is to provide an intelligent testing system for the interlayer peel strength of copper-aluminum composite plates, including an image acquisition and calibration module, a crack line extraction and parameter calculation module, a true boundary discrimination module, and a peel length calculation and strength output module. The image acquisition and calibration module is used to acquire image sequences and force value sequences, calibrate pixel equivalents, obtain relative position parameters, and convert them into interlayer spacing templates. The crack line extraction and parameter calculation module is used to perform edge detection to extract crack lines for each frame of the acquired image, calculate the vertical position of each crack line, construct an aluminum-side transition layer, an intermediate diffusion layer and a copper-side reaction layer, and distribute the crack lines to each layer after sorting them according to their vertical positions. The true boundary discrimination module is used to define hidden states and observations for each frame of image based on the interlayer spacing template and crack lines of each layer, define state transition probabilities based on the interlayer spacing template, define emission probabilities based on the degree of matching between the vertical position of the crack line and the vertical position in the hidden state, solve the optimal hidden state sequence using the Viterbi algorithm, and extract the true stripping boundary position and corresponding confidence level of each frame. The peel length calculation and strength output module is used to perform sub-pixel positioning of the actual peel boundary position to obtain the peel length, correspond the peel length to the force value sequence, select the effective peel segment after removing abnormal frames according to the confidence level, calculate the average peel strength based on the effective peel segment and output it.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention utilizes the inherent physical spacing of the multilayer intermetallic compounds at the interface of copper-aluminum composite plates as a priori constraints, integrates material structure information into the image analysis process, and solves the problem that traditional edge detection cannot distinguish multiple parallel crack lines by constructing three logical layers to spatially allocate crack lines. This elevates the identification of peeling boundaries from simple pixel feature analysis to the level of image understanding guided by the physical structure of the material.

[0015] 2. This invention models the continuous tracking of the peeling boundary as a global optimization problem of the hidden state sequence. It adopts the Viterbi algorithm to consider both the positional continuity and the layer change law in the time dimension. Even when the crack line is missing in some frames or the image is blurred, it can still infer the reasonable boundary position through the transition probability of the previous and next frames, which effectively improves the stability and anti-interference ability of peeling length measurement.

[0016] 3. This invention improves the boundary positioning accuracy to below the pixel level through sub-pixel positioning technology, and combines a credibility assessment mechanism to remove abnormal frames, ensuring the data quality on which the final peel strength calculation is based. The output average peel strength, maximum instantaneous strength and coefficient of variation can comprehensively reflect the interface bonding performance, providing a reliable basis for the quality evaluation and process optimization of copper-aluminum composite panels. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall process of the intelligent testing method for interlayer peel strength of copper-aluminum composite plates according to the present invention. Figure 2 This is a flowchart of step S2 of the present invention; Figure 3 This is a flowchart of step S3 of the present invention; Figure 4 The flowchart shows the structure of the intelligent testing system for interlayer peel strength of copper-aluminum composite plates according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Currently, in the peel strength testing of copper-aluminum composite panels, the presence of multiple intermetallic compound structures at the interface leads to multiple crack lines during peeling, making it difficult to accurately determine the true peel boundary. One objective of this invention is to provide an intelligent testing method for the peel strength of copper-aluminum composite panels. (See [link to relevant documentation]). Figure 1 As shown, it includes the following steps: S1. Simultaneously acquire image sequences and force value sequences of the interlayer peeling process of copper-aluminum composite plates; Obtain the relative position parameters between the intermetallic compound layers at the interface of the copper-aluminum composite plate, convert the relative position parameters into pixel spacing, and obtain the interlayer spacing template. S2. Perform edge detection on each frame of the image sequence to extract crack lines and calculate the vertical position of each crack line; Based on the interlayer spacing template, an aluminum-side transition layer, an intermediate diffusion layer, and a copper-side reaction layer are constructed in each frame of the image sequence to correspond to the spatial distribution of crack lines. After sorting by vertical position, the crack lines are assigned to each layer. S3. Based on the interlayer spacing template and the crack lines of each layer, define hidden states and observation values ​​for each frame of image. The hidden states represent the layer and vertical position of the actual peeling boundary, and the observation values ​​represent the vertical positions of all crack lines in the current frame. The transition probability of the hidden state between adjacent frames is defined based on the interlayer spacing template, and the emission probability is defined based on the degree of matching between the vertical position of the crack line and the vertical position in the hidden state. Based on the transition probability and emission probability, the Viterbi algorithm is used to solve the optimal hidden state sequence, and the true stripping boundary position and corresponding confidence level of each frame are extracted from the optimal hidden state sequence. S4. Perform sub-pixel localization on the extracted true peeling boundary position, match the localized position with the synchronously acquired force value sequence, remove abnormal frames based on the extraction reliability, select the effective peeling segment, and calculate the average peeling intensity based on the effective peeling segment.

[0020] This method constructs a three-layer hidden state sequence, uses known relative position parameters to constrain the state transition probability, and combines the Viterbi algorithm to solve for the global optimal path, thereby achieving continuous tracking and positioning of the real peeling boundary. Based on this, the peeling length is calculated, and then the interlayer peeling strength value is obtained.

[0021] Step S1 of this invention involves acquiring image sequences and force value sequences of the stripping process, calibrating pixel equivalents, and obtaining the relative position parameters of the intermetallic compounds. Specifically, it includes the following steps: S101: First, an image acquisition unit is deployed in the test device to acquire continuous images of the peeling process.

[0022] It should be noted that the entire testing device includes a tensile testing machine, a force sensor, and an image acquisition unit. The tensile testing machine is used to apply the peeling force, and the force sensor is used to measure the tensile force value in real time.

[0023] Specifically, a high-speed industrial camera was mounted directly above the peel test area, with the camera lens optical axis perpendicular to the surface of the copper-aluminum composite plate sample. The camera was set to external trigger mode. The output of the tensile testing machine's force sensor was connected to the camera's trigger input interface, and the force change rate threshold was set to [value missing]. .

[0024] Based on this, the control unit built into the tensile testing machine calculates the rate of change of force in real time. When detected At that time, the control unit sends a pulse trigger signal to the camera; the camera acquires a frame of the current field of view the instant it receives the rising edge of the pulse, and records the frame number of the frame in the internal buffer.

[0025] Simultaneously, at the same moment the tensile testing machine's control unit sends the trigger signal, it records the current force value F(t) and assigns this force value the same frame number as the image. Using the frame number as a unique identifier, each frame image is associated with a force measurement point, thus achieving synchronous alignment between the image and the force value.

[0026] It should be noted that the camera only triggers data acquisition when the rate of change of force exceeds a threshold, rather than using continuous shooting at a fixed frame rate. During the stripping process, the camera does not trigger acquisition in areas where the force change is gradual, thus reducing data redundancy while ensuring the integrity of critical information.

[0027] S102: After completing the synchronous acquisition of image and force values ​​in step S101, this step further calibrates the conversion relationship between image pixels and actual physical length. Specifically, before the formal test begins, two circular reflective markers with a diameter of 1 mm are pasted on the copper layer side of the copper-aluminum composite plate sample. The actual physical distance between the centers of the two markers is measured using a precision vernier caliper and recorded as follows: .

[0028] After placing the sample with the affixed markers at the test position, an image containing two markers is acquired using the aforementioned high-speed industrial camera. In the acquired image, the center coordinates of the two markers are determined using the centroid method and denoted as (x1, y1) and (x2, y2) respectively.

[0029] Furthermore, calculate the pixel distance between the two marker points in the image: ; Based on this, the pixel equivalent k is calculated using the following formula: ; In the formula, k is in millimeters per pixel, and this pixel equivalent is used to convert the location of the crack line detected in subsequent images from pixel coordinates to the actual physical length.

[0030] It should be noted that the two markers should be affixed to an area on the sample surface near the peeling path but without interfering with the peeling process, and the line connecting the two markers should be roughly parallel to the peeling direction to reduce projection error. A marker diameter of 1 mm is chosen to ensure clear visibility in the image without obstructing or interfering with the detection of crack lines.

[0031] S103: After completing the pixel equivalent calibration in step S102, this step further obtains the relative position parameters of the intermetallic compounds at the interface of the copper-aluminum composite plate and converts them into a spacing template in pixel units. Specifically, based on the annealing process parameters of the copper-aluminum composite plate, the relative position parameters of the intermetallic compounds at the corresponding annealing temperature are queried from a pre-established process database. For a copper-aluminum composite plate with an annealing temperature of 300℃, the physical spacing d between the first and second layers is obtained through metallographic calibration experiments. 12 =0.25mm, the physical distance between the second and third layers is d 23 =0.4mm.

[0032] Using the pixel equivalent k calculated by S102, the above physical spacing is converted into pixel spacing using the following formula: ; Among them, D 12 and D 23 All units are pixels.

[0033] Based on this, a tolerance δ for spacing matching is set. Considering the influence of image noise and detection error, the tolerance value is set to... That is, the allowable error is at least 2 pixels, and not less than D. 12 5%. This error tolerance is used in subsequent steps to determine whether the measured crack line spacing matches the template spacing.

[0034] It should be noted that the relative position parameter d 12 and d 23 The value of is closely related to the annealing temperature, and the above values ​​will vary under different annealing processes. In practical applications, the precise spacing value under the current process conditions can be obtained in advance through a small number of metallographic calibration experiments and stored in the process database for later retrieval.

[0035] S104: After completing the acquisition and calibration in S101 to S103, this step preprocesses the acquired raw images to enhance image quality and eliminate geometric distortion.

[0036] Specifically, for each frame of the original image acquired in S101, radial and tangential distortion corrections are first performed using pre-calibrated camera intrinsic parameters and distortion coefficients to eliminate image geometric distortion caused by lens distortion. Camera intrinsic parameters include focal length (f...x ,f y ) and principal point coordinates (c x ,c y The distortion coefficients include radial distortion coefficients k1, k2, k3 and tangential distortion coefficients p1, p2.

[0037] It should be noted that the camera parameters mentioned above were obtained in advance using the Zhang Zhengyou calibration method. This method first prints a black and white checkerboard calibration board, where the side length of each square is known. The calibration board is placed at different positions and angles within the camera's field of view, and multiple images are captured. For each image, the corner coordinates of the checkerboard are detected. Using the correspondence between the pixel coordinates of the corner points in the image and their physical coordinates in the world coordinate system, a camera imaging model equation is established. Solving the system of equations yields the camera's intrinsic parameter matrix and distortion coefficients.

[0038] Based on this, histogram equalization is performed on the corrected image. Histogram equalization transforms the gray-level histogram of the original image from a concentrated distribution to a uniform distribution, enhancing the gray-level contrast between the copper and aluminum layers and making the crack lines clearer in the image. Let the gray-level range of the original image be [0, L-1], and the transformation function be: ; In the formula, r is the original gray level; p r (j) represents the probability of gray level j appearing; s represents the transformed gray level.

[0039] After the above processing, the preprocessed image I is output. pre (t), where t is the frame number.

[0040] This preprocessed image will be used as input for the subsequent S2 step for crack line detection and extraction.

[0041] It should be noted that once the parameters for distortion correction and histogram equalization are determined, they remain unchanged within the same batch of tests and do not require repeated calibration for each frame. The preprocessing process only processes the keyframes acquired during trigger acquisition and does not process the time intervals that are not triggered, in order to ensure data processing efficiency.

[0042] like Figure 2 As shown, step S2 of the present invention involves detecting and extracting crack lines in the preprocessed image and calculating the geometric feature parameters of each crack line to provide a data basis for the subsequent S3 true boundary discrimination. Specifically, it includes the following steps: S201: Obtain the preprocessed image I output from S104. pre (t) is used to perform edge detection on the crack lines in the image to extract edge information.

[0043] Specifically, the Canny edge detection operator is first applied to I. pre (t) Processing: The Canny edge detection operator first smooths the image with a Gaussian filter to suppress noise, with the Gaussian kernel size set to 3×3 pixels and the standard deviation σ=1.0.

[0044] Then, the Sobel operator is used to calculate the gradient magnitude and gradient direction for each pixel: the Sobel operator uses two 3×3 convolution kernels to calculate the horizontal gradient G. x and vertical gradient G y The calculation formula is as follows: ; In the formula, * represents the convolution operation.

[0045] Based on this, the formulas for calculating the gradient magnitude G and the gradient direction θ are as follows: ; Furthermore, a non-maximum suppression (NMS) algorithm is used to process the gradient magnitude image. NMS compares the gradient magnitude of each pixel with the gradient magnitudes of its two neighboring pixels along the gradient direction. If the gradient magnitude of the current pixel is not a local maximum, it is set to zero; otherwise, its gradient magnitude is retained. This operation refines the edge width to a single pixel.

[0046] Then, a dual-threshold detection algorithm is used to perform threshold segmentation on the edge image after non-maximum suppression. The dual-threshold detection algorithm sets a low threshold T. low =30 and a high threshold T high =90. Gradient magnitude is greater than T. high Pixels with gradient magnitudes less than T are marked as strong edge pixels. low The pixels are suppressed to zero, and the gradient magnitude is between T. low and T high Pixels between these points are marked as weak edge pixels.

[0047] Finally, weak edge pixels are connected to strong edge pixels through edge connection operations. Specifically, for each weak edge pixel, it is checked whether there is a strong edge pixel in its eight neighborhoods. If there is, the weak edge pixel is converted into a strong edge pixel; otherwise, the weak edge pixel is set to zero.

[0048] After the above processing, a binary edge map E(t) is obtained, where white pixels represent edge points with a gray value of 255, and black pixels represent non-edge points with a gray value of 0.

[0049] It should be noted that the low threshold of 30 and the high threshold of 90 were pre-calibrated based on the grayscale statistical characteristics of the images from the copper-aluminum composite plate peeling process. Under this threshold combination, the edges of the crack lines can be completely preserved, while the speckle texture and false edges caused by uneven illumination on the sample surface are effectively suppressed. The Gaussian kernel size is set to 3x3 pixels, which can both smooth and denoise without excessively blurring the edge details of the crack lines.

[0050] S202: After obtaining the binary edge map E(t) in step S201, this step extracts the line segments of the edge points in E(t) to obtain the initial line segment representation of the crack line.

[0051] Specifically, the Hough transform line detection algorithm is used to process E(t). The basic principle of the Hough transform is to map each edge point in the image space to the parameter space. The parameter space adopts polar coordinate form ρ=xcosθ+ysinθ, where ρ is the distance from the origin to the line and θ is the angle between the normal of the line and the x-axis.

[0052] For each edge point (x, y) in the binary edge map E(t), iterate through all possible θ values, calculate the corresponding ρ value, and vote on it at the corresponding position in the accumulator array H(ρ, θ). The value of θ ranges from 0° to 180°, and a discrete value is taken every 1° for traversal. The value of ρ is determined based on the length of the image diagonal, and a discrete value is taken every pixel unit.

[0053] After the accumulator voting is completed, select the accumulator value that exceeds the threshold T. hough =40 (ρ,θ) is used as the detected line parameter. For each detected line parameter, edge points on the line are extracted in the image space, and these edge points are connected into line segments according to continuity and directional consistency. The minimum length of the line segment is set to 10 pixels, and line segments with a length less than this threshold are discarded.

[0054] Based on this, each detected line segment is represented parametrically. Let the coordinates of the two endpoints of a line segment be (x1, y1) and (x2, y2), then these two endpoints are used as the representation of the line segment. Simultaneously, the angle between the line segment and the horizontal direction is calculated using the following formula: ; It should be noted that the threshold T of the Hough transform accumulator hough The value of is related to the length and sharpness of the crack lines in the image. In this embodiment, the value is 40, which is determined in the following way: Ten preprocessed images containing typical crack lines were selected, and the actual crack segments were manually labeled. The Hough transform results were then filtered using different thresholds, and the detection accuracy and recall were statistically analyzed to select a threshold that balanced both. Experiments showed that a threshold below 40 introduced a large number of false straight segments generated by speckle noise, while a threshold above 40 missed some short but real crack lines. Therefore, 40 was chosen as the optimal threshold.

[0055] The minimum length of the straight line segment is set to 10 pixels, a value based on the actual imaging characteristics of the crack line during the peeling process of the copper-aluminum composite plate. Under the conditions of a camera resolution of 1920×1080 pixels and a field of view of 50mm×30mm, the physical length corresponding to 10 pixels is approximately 0.26 mm, which can effectively eliminate isolated noise points smaller than this size and false edges caused by speckle texture, while retaining the complete line segment of the real crack line.

[0056] The straight line segments retained after the above screening are denoted as the initial crack segments.

[0057] S203: Merge the obtained initial crack segments, connecting segments that belong to the same crack but are separated due to the fracture gap into a complete crack line.

[0058] Specifically, iterate through all the initial crack segments output in step S202, and for any two segments, calculate the distance between their nearest endpoints and the angle difference between the two segments.

[0059] Let the two line segments be l. a and l b The coordinates of their endpoints are respectively and Calculate l a The two endpoints and l b Take the minimum of all possible distances between the two endpoints of the line segment as the endpoint distance d between the two line segments. end .

[0060] At the same time, calculate the angle difference between the lines containing the two line segments. , where θ a and θ b These are the angles between the two line segments and the horizontal direction.

[0061] If both conditions are met pixels and If the angle is less than 10°, then these two line segments will be merged into a new line segment, as follows: Take l a and l bThe two endpoints furthest apart from the four endpoints are selected as the endpoints of the merged segment. This involves calculating the Euclidean distance between all pairs of endpoints and choosing the pair with the largest distance as the start and end points of the merged segment. The resulting new segment is added to the segment set, while the original two segments are removed from the set. a and l b .

[0062] Repeat the above traversal and merging process until no merging operation occurs after traversing all line segments. At this point, all line segments that meet the merging conditions have been connected into a longer complete line segment.

[0063] After merging, the set of crack lines for the current frame is obtained. , where m is the total number of crack lines in the current frame.

[0064] It should be noted that the values ​​of the endpoint distance threshold of 5 pixels and the angle difference threshold of 10° are determined based on the actual morphological characteristics of the crack lines during the peeling process of the copper-aluminum composite plate. Experimental observations show that the fracture gap of the same crack in the image usually does not exceed 5 pixels, and the directional deviation between the two crack segments before and after the fracture is small; this threshold combination can effectively connect fracture segments belonging to the same crack, while avoiding the incorrect merging of adjacent segments belonging to different cracks.

[0065] S204: Obtain the set of crack lines L(t) and calculate the geometric feature parameters of each crack line in the set, including vertical position and sharpness, for the subsequent S3 step of true boundary determination.

[0066] Specifically, for the set of crack lines any crack line in i The coordinates of its two endpoints are respectively and Take the vertical coordinate of the midpoint as the vertical position y of the crack line. i The calculation formula is as follows: ; It should be noted that the vertical position y i This indicates the vertical position of the crack line in the image. In this embodiment, the peeling direction is defined as the direction in which the y-value increases, that is, the larger the y-value, the closer the crack line is to the peeling tip.

[0067] For crack lines l i Clarity c i , defined as the average gradient magnitude of all pixels along the crack line, let crack line l i The set of pixels on is For each pixel p, the gradient magnitude is calculated using the Sobel operator. Clarity C i The calculation formula is as follows: ; In the formula, N i Crack line l i The total number of pixels on the screen.

[0068] Clarity C i The larger the value, the sharper the edge of the crack line, the more obvious the difference in gray levels on both sides, and the higher the probability that the crack line is a real peeling boundary.

[0069] Based on this, the calculated vertical position y of each crack line is... i And clarity c i The information is associated with the identification information of the crack line and stored to form a crack line feature set. .

[0070] S205: Distribute crack lines vertically to the aluminum-side transition layer, intermediate diffusion layer, and copper-side reaction layer. Specifically, the vertical position and clarity of each crack line are first obtained. To facilitate template matching based on relative position parameters, this step constructs three logical layers: an aluminum-side transition layer, an intermediate diffusion layer, and a copper-side reaction layer. The aluminum-side transition layer is used for crack lines corresponding to the CuAl2 layer, the intermediate diffusion layer is used for crack lines corresponding to the CuAl layer, and the copper-side reaction layer is used for crack lines corresponding to the Cu9Al4 layer.

[0071] Once the construction is complete, all crack lines in the current frame are sorted from smallest to largest according to their vertical position, and then distributed to the three layers mentioned above based on the sorting results.

[0072] Specifically, a set of crack lines in the aluminum side transition layer is constructed. Collection of crack lines in the intermediate diffusion layer Collection of crack lines in the copper side reaction layer During initialization, all three sets are empty sets.

[0073] For the m crack lines in the current frame crack line set L(t), the corresponding vertical position array is as follows: The corresponding resolution array is The quicksort algorithm was used to sort these crack lines according to their vertical position y. i The values ​​are sorted in ascending order, meaning that after sorting, they satisfy... .

[0074] After sorting, the corresponding resolution arrays are rearranged in the same order to obtain... .

[0075] After sorting, the sorted crack lines are assigned sequentially to the aluminum-side transition layer, intermediate diffusion layer, and copper-side reaction layer, according to the following rules: The top-ranked crack lines are assigned to the aluminum-side transition layer set. Several rows sorted in the middle are assigned to the intermediate diffusion layer set. The later-ranked ones are assigned to the copper-side reaction layer set. .

[0076] The specific allocation quantity is determined based on the relative position parameter d obtained in step S103. 12 and d 23 If the vertical position difference between adjacent crack lines after sorting matches a certain interlayer spacing, then that position is taken as the interlayer boundary point.

[0077] It should be noted that, since multiple crack lines may be generated in each layer during the peeling process, and some layers may not have crack lines temporarily, the three sets after allocation may contain zero, one, or more crack lines. The crack lines retained in each layer set are sorted according to their vertical position, with the one that is closest to the top (i.e., has the smallest y-value) being designated as the primary crack line of that layer, and the rest as secondary crack lines of that layer.

[0078] After this step, the three-layer allocation result of the current frame is output, namely the set of crack lines in the aluminum side transition layer. Collection of crack lines in the intermediate diffusion layer Collection of crack lines in the copper side reaction layer If no crack lines are detected in a certain layer, the corresponding set is empty.

[0079] like Figure 3 As shown, step S3 of the present invention involves using the known physical distance between the aluminum-side transition layer, the intermediate diffusion layer, and the copper-side reaction layer at the interface of the copper-aluminum composite plate as a constraint, and combining the Viterbi algorithm to perform global optimization on multiple consecutive frames of images to identify the true peeling boundary of each frame from multiple candidate crack lines. Specifically, this includes the following steps: S301: First, each frame of the peeling process is regarded as a time node, and two basic elements are defined for each time node: hidden state and observation value. The hidden state is used to describe the layer where the real peeling boundary of the current frame is located and its precise position, while the observation value is used to describe the vertical position and sharpness information of all crack lines detected in the current frame.

[0080] Specifically, suppose that a total of N frames of images are acquired during the entire peel test process, and the frame number is denoted as N. For frame t, its hidden state is defined as follows: ,in p is a discrete variable representing the current layer where the actual stripping boundary is located. tThis is a continuous variable representing the vertical position of the actual stripping boundary in the image frame, in pixels.

[0081] It should be noted that, The set of values ​​is These correspond to the three logical layers constructed in S205.

[0082] Specifically, This indicates that the actual peel boundary is located in the aluminum-side transition layer. This indicates that it is located in the intermediate diffusion layer. This indicates the reaction layer located on the copper side. The three layers are arranged sequentially from the aluminum side to the copper side in physical space, with the copper side reaction layer being the closest to the peeling tip.

[0083] For frame t, its observation is defined as , where m t y represents the total number of crack lines detected in frame t. i Let c be the vertical position of the i-th crack line. i Let represent the clarity of the i-th crack line.

[0084] Each crack line in the observations originates from the allocation result of step S205, that is, it has been assigned to one of the aluminum-side transition layer, intermediate diffusion layer or copper-side reaction layer respectively. However, in this step, its layer assignment is not distinguished, and it is only used as candidate observation data to participate in subsequent matching calculations.

[0085] It should be noted that the observed value o t It may contain zero, one, or more crack lines. If m t =0 means that no crack line was detected in the current frame. The subsequent emission probability will be set to a minimum value, indicating that there is no effective observation information in this frame, and inference will be made based on the state transition probability.

[0086] S302: Obtain the pixel spacing D after the relative position parameter transformation in step S1. 12 and D 23 And the allowable error tolerance δ, used to constrain the transition of hidden states between adjacent frames.

[0087] State transition probability Defined as the hidden state from the previous frame Transition to the current frame hidden state The probability of the stripping process must satisfy the physical laws governing the process.

[0088] Specifically, the setting of the transition probability follows the following three physical constraints: ① The actual stripping boundary can only move forward in the stripping direction and cannot go backward, that is, the vertical position p of the current frame. t It must be greater than or equal to the vertical position p of the previous frame. t-1.

[0089] ② The layer where the actual stripping boundary is located can only stay in the current layer or move forward one layer towards the copper side. It is forbidden to retreat towards the aluminum side or jump across layers.

[0090] ③ When the layer position changes, the difference in the vertical position of the crack line between different layers should match the physical spacing between the corresponding layers.

[0091] Based on the above constraints, the transition probability is defined in the following four cases: ① The layer recedes towards the aluminum side, that is This situation violates the physical laws of the stripping process, therefore the transition probability is set to zero. ; ② The stratigraphic level remains unchanged, that is At this point, the transition probability is determined by positional continuity, and is modeled using a Gaussian function: ; in Pixels are used to control the allowable range of positional changes. They must also meet the following requirements. Otherwise, the probability is set to zero.

[0092] ③ Advance one layer towards the copper side, i.e. In addition to the positional continuity constraint, the inter-layer spacing constraint must also be satisfied. Let the physical spacing between two layers before and after the forward movement be D. When and hour, ;when and hour, The transition probability is then defined as: ; in δ is the allowable error tolerance set in S103. This formula simultaneously penalizes deviations from the inter-story spacing due to positional changes, as well as cases of excessive positional changes, and requires... Otherwise, the probability is set to zero.

[0093] ④ Jumping across levels, i.e. This situation violates the physical laws of the stripping process, therefore the transition probability is set to zero. ; It should be noted that the coefficients 0.8 and 0.2 in the transition probability represent the prior probability of the layer remaining unchanged and the layer advancing one layer, respectively. These values ​​are determined based on the statistical characteristic that the interface cracks mainly propagate within a single layer during the peeling process of the copper-aluminum composite plate. The pixel value corresponds to a maximum peeling speed of approximately 2 millimeters per second at the camera frame rate, which is consistent with the actual situation of the copper-aluminum composite plate peeling test.

[0094] S303: This step first obtains the crack line vertical position array and sharpness array for each frame from step S2, which are used to calculate the degree of matching between the observations and the hidden states. (Emission probability) Defined as the hidden state of the current frame At that time, a set of crack lines was observed. The probability measures the degree of match between each candidate crack line in the current frame and the assumed true boundary location.

[0095] Specifically, let m be the total number of detected in frame t. t The crack lines are located vertically at the following positions: The resolutions are respectively For a given hidden state s t The emission probability is defined as the maximum value of the best-match score among all crack lines, and the calculation formula is as follows: ; In the formula, p t This represents the hypothetical true boundary perpendicular position in the hidden state; σ d =2 pixels is the distance error tolerance; C max This is the maximum value of the sharpness of all crack lines in the current frame, used to normalize the sharpness to the interval [0,1].

[0096] It should be noted that the above formula means: relative to the assumed boundary position p t The closer the crack lines are, the higher their distance matching score; the sharper the crack lines, the higher their probability of being true boundaries. The maximum value among all crack lines is taken as the emission probability of the current frame, representing the crack line that we trust best matches the assumed location and is the sharpest.

[0097] If no crack lines are detected in the current frame, i.e., m t =0, then the emission probability is set to a minimum value. This indicates that there is no valid observation information in this frame, and subsequent inferences are made based on the state transition probability.

[0098] Based on this, in order to reduce computational complexity, this step only retains the top three crack lines with the highest emission probability as valid candidates for the current frame, and the remaining crack lines will no longer participate in the subsequent Viterbi recursive calculation.

[0099] Specifically, for frame t, after calculating the emission probability score for each crack line, sorting them from highest to lowest score, and retaining only the three highest-scoring lines. Yes, the remaining crack lines are ignored in this frame. This filtering strategy is based on experimental observations: the true peel boundary usually corresponds to one of the several crack lines with the highest sharpness in each frame, and the probability of more than three high-resolution crack lines appearing in that frame is extremely low.

[0100] It should be noted that the calculation of the emission probability does not directly depend on the layer. Since the stratigraphic assignment of crack lines in the observations has already been completed in S205, this step only uses location and sharpness information for matching, and the stratigraphic constraints are applied by the state transition probabilities defined in step S302.

[0101] S304: After defining the emission probability in S303, this step further employs the Viterbi dynamic programming algorithm to solve for the optimal hidden state sequence over N consecutive frames. This maximizes the joint probability of the sequence. This step sequentially executes four sub-processes: initialization, recursive calculation, termination calculation, and backtracking.

[0102] First, initialization is performed. For frame 1, for each possible hidden state... Calculate the initial probability The initial probability is determined by the initial layer probability. The emission probability calculated in step S303 The product is determined by the following formula: ; In the formula, This represents the initial layer probability.

[0103] In this embodiment, based on experimental observations that the peel boundary is typically located at the aluminum-side transition layer at the start of the peel test, the following settings are made: Simultaneously record the optimal front drive. .

[0104] It should be noted that, due to p t Since the variables are continuous, discretization is required in actual calculations. In this step, 5 discrete points are uniformly selected within a range of ±2 pixels around each candidate crack line location as p. t For the first frame, if the number of detected crack lines is m1, then each crack line contributes 5 candidate positions. Adding the default candidate positions when there are no crack lines, the total number of candidate states does not exceed [a certain value]. indivual.

[0105] Furthermore, a recursive calculation is performed: for the t-th frame... For each possible hidden state ,calculate and optimal forerunner The recursive formula is as follows: ; ; In the formula, The state transition probability defined for S302; The emission probability defined for S303.

[0106] During the recursive process, for each candidate hidden state in the current frame, all possible hidden states in the previous frame are traversed, the transition probability is calculated, and multiplied by . Take the maximum value and multiply it by the transmission probability of the current frame.

[0107] It should be noted that only transition paths that satisfy physical constraints are considered in the recursive calculation. Specifically, if or ,but This path is directly pruned and requires no further calculation. Furthermore, if the optimal probability of the previous frame... If the preceding state is not included in the calculation, then the corresponding preceding state is also excluded.

[0108] After the recursion is complete, the calculation is terminated. The optimal hidden state of the last frame is such that... The formula for calculating the maximum value is as follows: ; Finally, backtracking is performed. Starting from frame N, the best predecessor recorded is used sequentially. The optimal hidden state for each frame is obtained by backtracking, and the backtracking formula is as follows: ; After backtracking is completed, the optimal hidden state sequence from frame 1 to frame N is obtained. .

[0109] It should be noted that the time complexity of the Viterbi algorithm is O(n log n). ,in The value of N represents the number of candidate states per frame. The value of N depends on the total duration of the stripping test and the camera trigger frequency. In this embodiment, the average frequency of camera trigger acquisition is approximately 10 frames per second, and a complete stripping test typically lasts 10 to 30 seconds. Therefore, the total number of frames N is usually between 100 and 300 frames. Under the condition of this frame range and the number of candidate states per frame being controlled within 45, the computational complexity of the Viterbi algorithm is approximately N × 45. 2 Each operation, ranging from 200,000 to 600,000 times, can be completed in milliseconds on a regular computer, meeting the requirements for real-time processing.

[0110] S305: After completing the global optimal solution of the Viterbi algorithm in S304, this step extracts the true stripping boundary position of each frame from the obtained optimal hidden state sequence and calculates the corresponding confidence level.

[0111] Specifically, for frame t, its optimal hidden state is ,in This is the actual peeling boundary location of the frame, denoted as... .

[0112] At the same time, the probability value corresponding to the optimal hidden state The confidence level r(t) is used as the result of the frame discrimination.

[0113] Based on this, this step sets a confidence threshold. If the credibility of a certain frame If the result is not found, a low reliability alarm will be triggered, indicating that the judgment result of the frame is not reliable and there may be problems such as inter-layer mismatch or abnormal image quality. It is recommended to manually review the data or remove the frame data from the subsequent peel strength calculation.

[0114] It should be noted that for paths excluded during the S304 recursion process or frames that do not match any candidate crack lines, the optimal probability value is... The values ​​are usually low and will automatically fall into the low confidence range, so there is no need to set additional anomaly detection rules separately.

[0115] After this step is completed, the actual stripping boundary position of each frame will be output. and the corresponding confidence level r(t), where It is then passed to step S4 for calculating the peel length and peel strength.

[0116] Step S4 of this invention involves calculating the peel length based on the actual peel boundary position of each frame, and calculating the peel strength by combining the synchronously acquired force data, ultimately outputting the test result. Specifically, this includes the following steps: S401: Obtain the actual stripping boundary position for each frame in step S3. The location is the integer-pixel coordinate output by the Viterbi algorithm in S304. To further improve the measurement accuracy, this step uses the gray-scale moment method to perform sub-pixel positioning of the boundary position.

[0117] Specifically, with Centered on the preprocessed image Take a rectangular window with a width of 5 pixels and a height of 1 pixel, with the window orientation perpendicular to the stripping boundary. For each pixel within the window, record its grayscale value g(j) and its vertical coordinate j, where... The formula for calculating the grayscale moments of all pixels within the window is as follows: ; In the formula, y sub This refers to the boundary position after subpixel localization, with the unit being pixels.

[0118] This formula improves the boundary positioning accuracy to the level of 0.1 pixels by using a gray-scale weighted average.

[0119] It should be noted that the gray-scale moment method assumes that the gray-scale distribution on both sides of the boundary changes abruptly. This assumption holds true during the peeling process of copper-aluminum composite panels because there is a significant difference in gray-scale between the peeled and unpeeled areas.

[0120] S402: Obtain the pixel equivalent k from step S102 and the boundary position y after sub-pixel localization in step S401. sub It also calculates the curvature k(t) of the sample surface in real time, which is used to convert the pixel coordinates in the image into the actual physical length.

[0121] Specifically, the bending curvature k(t) is calculated in real time using two markers pasted on the sample surface in step S102: the vertical coordinates of the two markers are obtained by tracking the position changes of the two markers in consecutive frames. and And calculate the difference between the two. .

[0122] At the same time, based on the actual physical distance between the two marking points on the sample Calculate the tilt angle θ(t) of the sample surface relative to the camera optical axis: ; Based on this, the curvature k(t) is calculated using the following formula: ; It should be noted that when the degree of curvature of the sample is small, k(t) approaches zero, and curvature correction is not required; when the degree of curvature of the sample is large, the absolute value of k(t) increases, and the curvature correction effect is more effective.

[0123] Based on this, the boundary position y after sub-pixel localization sub Convert to true arc length. Since the sample may bend during the peeling process, the actual peeling length is not the straight-line distance shown in the image, but rather the arc length along the bent surface. The conversion formula is as follows: ; In the formula, u is the integral variable along the stripping direction, and the unit is pixels.

[0124] Discretize the integral formula and use the trapezoidal rule for calculation: ; Where k is the pixel equivalent calibrated in step S102, in millimeters per pixel.

[0125] It should be noted that when the degree of bending of the sample is small, k(t) approaches zero. The conversion degenerates into a straight-line length transformation; when the sample is significantly curved, k(t) is not zero, and the arc length of the curved portion will be greater than its projected length. The corrected... It can more accurately reflect the actual peeling length.

[0126] After this step, the actual stripping length of each frame is output. The unit is millimeters, which is used as input for subsequent peel strength calculations.

[0127] S403: Obtain the force value sequence F(t) collected in step S101 and the actual peel length of each frame in step S402. And the credibility r(t) of each frame in step S305.

[0128] Since the force change rate triggers the acquisition in S101, each frame of the image has established a one-to-one correspondence with a force measurement point through the frame number. Therefore, the force value F(t) of the t-th frame can be directly correlated with the peeling length of the t-th frame. Pair them up to form data point pairs. .

[0129] Based on this, abnormal frames are removed according to the confidence level r(t): a confidence level threshold is set. For the t-th frame, if Then the data point pair corresponding to the frame Data points marked as invalid will not be included in subsequent peel strength calculations. The frame number and confidence value of the removed data points will be recorded for easy tracking of test results and anomaly analysis.

[0130] It should be noted that the credibility threshold The value of is determined as follows: 10 groups of samples that have completed the stripping test are selected, and the frames that are obviously abnormal are manually labeled. The confidence distribution of these abnormal frames is statistically analyzed. It is found that more than 95% of the abnormal frames have a confidence of less than 0.3. Therefore, 0.3 is used as the dividing line between normal frames and abnormal frames.

[0131] After this step, a set of valid data point pairs is output. This only includes credibility. The system outputs a list of rejected frame numbers and their corresponding confidence values ​​for subsequent result verification.

[0132] S404: Obtain the set of valid data point pairs from step S403. The test software obtains the sample width w input before the test. It should be noted that the sample width w is measured by the operator using calipers before the test begins and then input into the test software as a known parameter for calculation.

[0133] To exclude unstable data from the beginning and end of the stripping process, this step selects the effective stripping segments.

[0134] Specifically, the starting rejection length is set to... mm, with a rejection rate of 5% of the total peeling length at the end; let the total peeling length be... The end position of the last removal is .

[0135] In the set of valid data point pairs, only those that satisfy the following conditions are retained. and The data points are selected based on their specific characteristics; the remaining data points are not included in subsequent calculations.

[0136] It should be noted that the initial rejection length of 5 mm is based on practical experience in peel testing of copper-aluminum composite plates: in the early stages of peeling, the peel force has not yet reached a stable state, and there may be irregular segments with initial tearing at the peel front; the data from these segments is usually not included in the peel strength assessment. The final rejection ratio of 5% is used to exclude the force drop segment caused by the end of the clamp's travel when the sample is about to be completely peeled.

[0137] Based on this, the average peel strength is calculated using the following formula: ; In the formula, w is the width of the sample, in millimeters; F is the peeling force, measured in Newtons; L represents the peeling length, in millimeters.

[0138] Discretize the integral formula and use the trapezoidal integral method for approximate calculation: ; In the formula, F i and L i Let the force value and peel length of the i-th valid data point be respectively, and sum them to iterate through all adjacent data point pairs within the valid peel segment.

[0139] Simultaneously, calculate the maximum instantaneous peel strength: ; Calculate the coefficient of variation of the peel strength to evaluate the stability of the peeling process: ; In the formula, std represents the standard deviation; mean represents the average value.

[0140] The smaller the coefficient of variation, the more stable the peeling process and the more uniform the interface bonding; the larger the coefficient of variation, the more obvious the skip peeling or the uneven interface phenomenon exists in the peeling process.

[0141] It should be noted that the relationship between the peeling force F and the peeling length L in the above calculation formula is based on the force value triggering synchronous acquisition mechanism in step S101, and is consistent with the valid data point pairs after removing abnormal frames in step S403, to ensure the accuracy of the calculation results.

[0142] After this step, the average peel strength is output. Maximum instantaneous peel strength The peel strength variation coefficient (CV) is used as input for step S405.

[0143] S405: This step outputs the average peel strength. Maximum instantaneous peel strength Peel strength coefficient of variation (CV), initial rejection length End position of tail removal Total peel length Total test duration, list of rejected frame numbers and their corresponding confidence values, and confidence level. The frame number and its confidence value.

[0144] Let the total number of frames be N, and the number of low-confidence frames be... ,like If the test results are not found, an additional overall alarm will be output, indicating that there may be a large error in the test results and suggesting that the test be retested or manually reviewed.

[0145] It should be noted that the above output results are exported as an Excel file for subsequent data analysis.

[0146] After this step, the entire intelligent test method for interlayer peel strength of copper-aluminum composite panels is completed, and a complete test report is output.

[0147] The second objective of this invention is to provide an intelligent testing system for the interlayer peel strength of copper-aluminum composite panels, such as... Figure 4 As shown, it includes an image acquisition and calibration module, a crack line extraction and parameter calculation module, a true boundary discrimination module, and a peel length calculation and strength output module.

[0148] The image acquisition and calibration module is used to perform step S1. This module includes a high-speed industrial camera, a force sensor, and calibration components. The camera is mounted directly above the peel test area, with the lens optical axis perpendicular to the sample surface, and uses an external trigger mode. The output of the force sensor is connected to the trigger input interface of the camera. When the rate of change of force exceeds a set threshold, the tensile testing machine control unit sends a pulse trigger signal to the camera. The camera acquires a frame and records the frame number. Simultaneously, the control unit records the current force value and assigns it the same frame number, achieving synchronous alignment between the image and the force value.

[0149] The calibration component is used to attach two circular reflective markers before testing, measure the actual physical distance between the centers of the markers, acquire the marker images, calculate the pixel equivalent, and obtain the relative position parameters of the aluminum-side transition layer, intermediate diffusion layer, and copper-side reaction layer according to the annealing process parameters. These parameters are then converted into pixel spacing and used as template parameters for subsequent discrimination.

[0150] The crack line extraction and parameter calculation module is used to execute step S2. This module performs Canny edge detection on the preprocessed image, extracts straight line segments using Hough transform, merges similar line segments to form complete crack lines, calculates the vertical position and sharpness of each crack line, constructs three logical layers: an aluminum-side transition layer, an intermediate diffusion layer, and a copper-side reaction layer, and sorts all crack lines according to their vertical position before assigning them to the corresponding logical layers, outputting the crack line set and corresponding sharpness array for each layer.

[0151] The true boundary discrimination module is used to execute step S3. This module constructs a multi-layer hidden state and observation model, defining the hidden state of each frame as the layer and vertical position of the true peeling boundary, and the observation value as the vertical position and sharpness of all crack lines in that frame.

[0152] The true boundary discrimination module defines state transition probabilities based on relative position parameters. Constraints are placed on boundaries that can only advance, layers that can only remain stationary or advance one layer towards the copper side, and the spacing must match the template. Simultaneously, emission probabilities are defined based on the distance between the crack line and the hypothetical location, as well as the sharpness. This module employs the Viterbi algorithm to perform global dynamic programming across multiple consecutive frames, recursively calculating the maximum probability of each candidate state in each frame and recording the optimal predecessor. It then backtracks to obtain the optimal hidden state for each frame, extracts the true stripping boundary location, and calculates its reliability.

[0153] The peel length calculation and strength output module is used to execute step S4. This module uses the gray-scale moment method to perform sub-pixel localization of the actual peel boundary position, calculates the sample curvature in real time by tracking marker points, and converts the image coordinates into the actual arc length to obtain the peel length. This module removes abnormal frames according to the confidence threshold, selects the effective peel segment after the initial removal length and before the final removal ratio, calculates the average peel strength, maximum instantaneous peel strength, and peel strength variation coefficient in combination with the sample width, and outputs the test results and low confidence alarm records.

[0154] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A smart testing method for the interlayer peel strength of copper-aluminum composite panels, characterized in that, Includes the following steps: S1. Simultaneously acquire image sequences and force value sequences of the interlayer peeling process of copper-aluminum composite plates; Obtain the relative position parameters between the intermetallic compound layers at the interface of the copper-aluminum composite plate, convert the relative position parameters into pixel spacing, and obtain the interlayer spacing template. S2. Perform edge detection on each frame of the image sequence to extract crack lines and calculate the vertical position of each crack line; Based on the interlayer spacing template, an aluminum-side transition layer, an intermediate diffusion layer, and a copper-side reaction layer are constructed in each frame of the image sequence to correspond to the spatial distribution of crack lines. After sorting by vertical position, the crack lines are assigned to each layer. S3. Based on the interlayer spacing template and the crack lines of each layer, define hidden states and observation values ​​for each frame of image. The hidden states represent the layer and vertical position of the actual peeling boundary, and the observation values ​​represent the vertical positions of all crack lines in the current frame. The transition probability of the hidden state between adjacent frames is defined based on the interlayer spacing template, and the emission probability is defined based on the degree of matching between the vertical position of the crack line and the vertical position in the hidden state. Based on the transition probability and emission probability, the Viterbi algorithm is used to solve the optimal hidden state sequence, and the true stripping boundary position and corresponding confidence level of each frame are extracted from the optimal hidden state sequence. S4. Perform sub-pixel localization on the extracted true peeling boundary position, match the localized position with the synchronously acquired force value sequence, remove abnormal frames based on the extraction reliability, select the effective peeling segment, and calculate the average peeling intensity based on the effective peeling segment.

2. The intelligent testing method for interlayer peel strength of copper-aluminum composite plates according to claim 1, characterized in that: The synchronous acquisition of image sequences and force sequences is specifically as follows: the force change rate is monitored in real time by a force sensor. When the force change rate exceeds a set threshold, the camera is triggered to acquire an image frame and add it to the image sequence and record the frame number. At the same time, the force value at that moment is recorded and added to the force sequence and assigned the same frame number. The frame number is used as the corresponding identifier between the image sequence and the force sequence.

3. The intelligent testing method for interlayer peel strength of copper-aluminum composite plates according to claim 1, characterized in that: The method for obtaining the relative position parameters between the intermetallic compound layers at the interface of the copper-aluminum composite plate is as follows: based on the annealing process parameters of the copper-aluminum composite plate, the corresponding relative position parameters at the annealing temperature are queried from a preset database.

4. The intelligent testing method for interlayer peel strength of copper-aluminum composite plates according to claim 1, characterized in that: Before converting the relative position parameters into pixel spacing, the pixel equivalent is first calibrated by two markers pasted on the sample surface. Specifically, the actual physical distance between the centers of the two markers is measured, an image containing the two markers is acquired and its center coordinates are detected, the pixel distance is calculated, and the pixel equivalent is obtained by dividing the actual physical distance by the pixel distance.

5. The intelligent testing method for interlayer peel strength of copper-aluminum composite plates according to claim 1, characterized in that: The edge detection and crack line extraction specifically includes: performing edge detection on the image using the Canny operator, extracting straight line segments through Hough transform, merging line segments with a distance less than a threshold and an angle difference less than a threshold to form a complete crack line.

6. The intelligent testing method for interlayer peel strength of copper-aluminum composite plates according to claim 1, characterized in that: The process of sorting crack lines by vertical position and then assigning them to each layer is as follows: the sorted crack lines are sequentially matched according to their vertical position, the vertical position difference between adjacent crack lines is calculated, and if the difference is within the allowable error range of the relative position parameter, it is taken as the interlayer boundary point. Crack lines before the boundary point are assigned to the current layer, and crack lines after the boundary point continue to match the next interlayer spacing until all crack lines are assigned.

7. The intelligent testing method for interlayer peel strength of copper-aluminum composite plates according to claim 1, characterized in that: The transition probability is set based on the type of stratigraphic change: When the layer remains unchanged, the transition probability is constrained by a Gaussian function of the position difference between adjacent frame boundaries; When the layer advances one layer towards the copper side, the transition probability is constrained by a Gaussian function of the position difference and the interlayer spacing deviation. When the layer retreats towards the aluminum side or jumps across layers, the transfer probability is zero; The emission probability is calculated based on a Gaussian function of the distance between the vertical position of the crack line and the vertical position of the hidden state, as well as a normalized value of the crack line sharpness.

8. The intelligent testing method for interlayer peel strength of copper-aluminum composite plates according to claim 1, characterized in that: The subpixel positioning adopts the gray-scale moment method, taking the actual stripping boundary position as the center of the window, and calculating the subpixel-level boundary coordinates by gray-scale weighted averaging.

9. The intelligent testing method for interlayer peel strength of copper-aluminum composite plates according to claim 1, characterized in that: The calculation of average peel strength based on the effective peel segment is as follows: within the effective peel segment, the average force value between two adjacent frames is multiplied by the peel length increment within that segment, and the sum is divided by the sample width and the total length of the effective peel segment to obtain the average peel strength.

10. An intelligent testing system for the interlayer peel strength of copper-aluminum composite panels, used to perform the intelligent testing method for the interlayer peel strength of copper-aluminum composite panels as described in any one of claims 1-9, characterized in that, It includes an image acquisition and calibration module, a crack line extraction and parameter calculation module, a true boundary discrimination module, and a peel length calculation and strength output module; The image acquisition and calibration module is used to acquire image sequences and force value sequences, calibrate pixel equivalents, obtain relative position parameters, and convert them into interlayer spacing templates. The crack line extraction and parameter calculation module is used to perform edge detection to extract crack lines for each frame of the acquired image, calculate the vertical position of each crack line, construct an aluminum-side transition layer, an intermediate diffusion layer and a copper-side reaction layer, and distribute the crack lines to each layer after sorting them according to their vertical positions. The true boundary discrimination module is used to define hidden states and observations for each frame of image based on the interlayer spacing template and crack lines of each layer, define state transition probabilities based on the interlayer spacing template, define emission probabilities based on the degree of matching between the vertical position of the crack line and the vertical position in the hidden state, solve the optimal hidden state sequence using the Viterbi algorithm, and extract the true stripping boundary position and corresponding confidence level of each frame. The peel length calculation and strength output module is used to perform sub-pixel positioning of the actual peel boundary position to obtain the peel length, correspond the peel length to the force value sequence, select the effective peel segment after removing abnormal frames according to the confidence level, calculate the average peel strength based on the effective peel segment and output it.