Three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT

Abstract

The invention discloses a three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT, and the method specifically comprises the following steps: 1, photographing a to-be-detected potting product through a CT machine, and extracting CT photos at different angles; step 2, decomposing bubble defects and a circuit system structure based on the CT picture shot in the step 1; 3, performing three-dimensional reconstruction on the bubble defect and the circuit system structure; and 4, based on a reconstruction result in the step 3, constructing a fused three-dimensional diagram of the bubble defect and the circuit system structure. By means of the method, the bubble defect can be reconstructed, and follow-up product detection is facilitated.

Description

Technical Field

[0001] The invention belongs to the technical field of potting product defect detection and relates to a three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT. Background Art

[0002] At present, potting is widely used in the manufacturing process of some precision electronic products, mainly used to protect electronic products from external impact, high temperature and high pressure. If there are holes in the potting material inside the potting product, if the hole position is sensitive or the size is large, it will affect the safety performance of the electronic product after potting. For different electronic product manufacturers, the potting process has different process requirements for the presence of holes in the potting material. However, most of them restrict the location, size and number of holes to determine whether they are qualified products. Since the general electronic product shell is an opaque shell, it is difficult for the manufacturer to intuitively detect the hole condition of the internal epoxy material. Most inspections generally adopt a destructive sampling method, that is, a certain number of samples are selected for cutting and inspection of possible hole defects. This detection method is inefficient and not comprehensive and reliable. At present, with the development of science and technology, non-destructive testing has been increasingly used in practice. Industrial CT is a key device commonly used for non-destructive testing. It can obtain perspective photos of the inside of the product without damaging the shell through X-rays. However, the photos are all flat, and if the defect location is to be accurately predicted, the operator is required to be high. Therefore, by reconstructing the internal structure of the product in three dimensions through these photos, the operator can clearly judge the defects and size of the product through the internal three-dimensional image; in addition, the position of the bubble in the potted product is critical, and it mainly protects the circuit system. If too many elements are added during the inspection, it will be difficult for the operator to accurately observe the position and size of the bubble, and it will be even more difficult to judge whether the product is qualified or not. Summary of the invention

[0003] The purpose of the present invention is to provide a three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT, by which the bubble defects can be reconstructed.

[0004] The technical solution adopted by the present invention is a three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT, which specifically includes the following steps:

[0005] Step 1: Take photos of the potted product to be inspected by a CT machine and extract CT photos from different angles;

[0006] Step 2, based on the CT image taken in step 1, decomposing the bubble defect and the circuit system structure;

[0007] Step 3, three-dimensionally reconstructing the bubble defect and the circuit system structure respectively;

[0008] Step 4: Based on the reconstruction result of step 3, a fused three-dimensional image of the bubble defect and the circuit system structure is constructed.

[0009] The present invention is also characterized in that:

[0010] The specific process of step 1 is:

[0011] The potted product to be inspected is placed on a cylindrical pad on a turntable driven by a servo motor to ensure that the axis of the potted product to be inspected is concentric with the axis of the turntable. The servo motor drives the turntable to rotate. Every 0.35° rotation, the flat-panel detector will collect a CT image formed by the X-ray light source penetrating the potted product to be inspected. The turntable collects 1024 CT photos for one rotation. Among them, the potted product to be inspected includes the workpiece shell, circuit system structure, epoxy potting material and bubble defects.

[0012] The specific process of step 2 is:

[0013] Using the watershed algorithm, 1024 bubble defect images and 1024 circuit system structure images are decomposed from the 1024 potted products to be inspected obtained in step 1.

[0014] The specific process of step 3 is:

[0015] Step 3.1, assuming that the reconstructed 3D image is f(x, y, z), f j represents the pixel value of the jth pixel of the three-dimensional image f(x, y, z), then the three-dimensional image f(x, y, z) is represented by an N-dimensional vector F = [f 1 ,f 2 ,...,f N ] T It means that N = n × n × n;

[0016] Step 3.2, discretize the CT images segmented in step 2 according to the positions where the m×m rays reach the plane detector, so that the value of each pixel represents the projection value after a ray passes through the object; at the same time, each CT image is also converted into an M-dimensional vector, then M = m×m, and P = [p 1 ,p 2 ,...,p M ] T Indicates that M is the total number of rays, where p 1 ,p 2 ,...,p M The value is the pixel value in the CT photo. According to the physical process of imaging and the corresponding projection model, the relationship between the three-dimensional image vector and the projection data vector is expressed as:

[0017]

[0018] In the formula, p i is the projection value of the i-th ray, that is, the pixel value on the CT image, w ij Represents the relationship between the j-th pixel and the projection value of the i-th ray, also called the weight factor. If the ray passes through this pixel, the weight factor is 1, otherwise it is 0;

[0019] Expand equation (1) into the following linear equation system:

[0020]

[0021] Formula (2) is expressed in matrix as:

[0022] P=WF (3)

[0023] In the formula, F = [f 1 ,f 2 ,...,f N ] is the image vector, P = [p 1 ,p 2 ,...,p M ] T is the projection data vector, representing the ray projection value, and W is the projection matrix.

[0024] In step 3.2, the joint algebraic iterative reconstruction algorithm SART is used to solve formula (3), as follows:

[0025] set up:

[0026]

[0027] Where W i,+ is an M×1 vector, which represents the weight sum of all pixels for the i-th ray, and M represents the total number of rays under a projection angle;

[0028]

[0029] Where W +,j is an N×1 vector representing the weighted sum of the jth pixel for all rays. The pixel is corrected by the following process:

[0030] 1) Assign initial value to the position image vector:

[0031]

[0032] Wherein, j=1, 2, 3..., N represents the pixel index number, represents the iterative calculation value of the j-th pixel, Represents the iterative calculation value of the j-th pixel in the previous iteration. In the first iteration, it represents the initialized pixel value;

[0033] 2) Calculate the theoretical projection value after the i-th ray passes through the object

[0034]

[0035] In the formula, The last iterative calculation value of the j-th pixel;

[0036] 3) Calculate the error Δ between the theoretical projection grayscale value and the actual projection grayscale value of the i-th ray i :

[0037]

[0038] In the formula, p i Represents the actual projection grayscale value on the CT image corresponding to the i-th ray at the current projection angle;

[0039] 4) Calculate the correction value C of the jth pixel j :

[0040]

[0041] 5) Use the errors of all rays at the current projection angle to calculate the error correction value of the j-th pixel: the gray value f of the pixel at the j-th point j To make corrections:

[0042]

[0043] In the formula, λ represents the relaxation factor;

[0044] 6) Repeat steps 2) to 5) for all rays under the projection angle to complete the correction of the next generation reconstructed image in this direction, and traverse all projection angles to complete one iteration;

[0045] 7) After traversing the entire 1024 CT photos, the corresponding iteration result values ​​of each element of the three-dimensional image are obtained And according to formula (10), the gray value f is obtained after correction j ;

[0046] 8) The f obtained in step 7) j As the initial value, repeat steps 2) to 5) and iterate the calculation of 1024 CT photos in turn. The grayscale value of all elements obtained by traversing 1024 photos for the first time is F (I) The grayscale value of all elements obtained by traversing 1024 photos for the second time is F (II) After each traversal, the corresponding elements will be calculated to see if they converge. The convergence condition is:

[0047]

[0048] Here, e represents the number of times the iterative calculation traverses 1024 CT images.

[0049] In step 3.2, in step 6), after the iterative calculation of the projection angle corresponding to the first CT photo is completed, the second CT photo with an angle of +0.35° is selected for another iteration. At this time, the f of the previous iteration result is j As the initial value of this iteration, when calculating the ray passing through the three-dimensional image corresponding to the second CT photo, the ray source is rotated 0.35° around the three-dimensional image in the opposite direction of the turntable rotation direction.

[0050] In step 3.2, in step 8), if the error between two consecutive calculation results is less than 0.00001, it means that the element has converged; if an element cannot converge according to formula (11), the convergence can be determined by the maximum number of iterative calculations traversing 1024 CT images.

[0051] The specific process of step 4 is: since the constructed bubble defect and circuit system structure are three-dimensional spaces of the same size, the grayscale values ​​of the bubble defect and the circuit system structure at the same position are compared, and the minimum value of the two is taken as the grayscale value after the bubble defect and the circuit system structure are fused, thereby obtaining a three-dimensional graph in which the bubble defect and the circuit system structure are fused together.

[0052] The beneficial effect of the present invention is that the method for three-dimensional reconstruction of potting bubble defects detected by industrial CT provided by the present invention is based on the characteristics of CT photos, and uses image processing methods to extract and decompose bubble defect features and circuit system structure features in a series of CT photos of products at different angles collected by the CT machine. The decomposed bubble photos and circuit system structure photos are three-dimensionally reconstructed using the SART method, and the three-dimensionally reconstructed photos are three-dimensionally combined to form a three-dimensional combined image of the bubble and the circuit system structure, so that the position and size of the bubble relative to the circuit system structure can be observed from all aspects, so as to facilitate the operator to judge the potting quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a principle diagram of CT photo collection of potting products collected in the three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT of the present invention;

[0054] Figure 2 It is a CT photo of the potting product in the three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT of the present invention;

[0055] Figure 3The CT photos of the simple products are segmented by image processing in the three-dimensional reconstruction method for detecting the filling bubble defects based on industrial CT of the present invention;

[0056] Figure 4 It is a CT photo of bubble feature segmentation obtained by image processing in the three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT of the present invention;

[0057] Figure 5 It is a circuit feature segmentation CT photo segmented by image processing in the three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT of the present invention;

[0058] Figure 6 It is a three-dimensional characteristic image of bubbles after the bubble characteristics are individually three-dimensionally reconstructed in the three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT of the present invention;

[0059] Figure 7 It is a three-dimensional characteristic image of the circuit system structure after the circuit features are independently three-dimensionally reconstructed in the three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT of the present invention;

[0060] Figure 8 A schematic diagram showing the i-th X-ray passing through each element of a three-dimensional region in the three-dimensional reconstruction method for detecting encapsulation bubble defects based on industrial CT of the present invention;

[0061] Fig. 9 It is a three-dimensional image with bubble features and circuit features after integrating the three-dimensional feature image of the bubble and the three-dimensional feature image of the circuit in the three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT of the present invention;

[0062] Fig.10 It is another observation angle image of the three-dimensional image with bubble features and circuit features after integrating the bubble three-dimensional feature image and the circuit three-dimensional feature image in the three-dimensional reconstruction method for detecting encapsulation bubble defects based on industrial CT of the present invention.

[0063] In the figure, 1. flat panel detector, 2. turntable, 3. cylindrical pad, 4. potted product to be detected, 5. X-ray light source, 6. workpiece housing, 7. circuit system structure, 8. epoxy potting material, 9. bubble defect. DETAILED DESCRIPTION

[0064] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0065] Example 1

[0066] The present invention is based on the three-dimensional reconstruction method of industrial CT detection of potting bubble defects, and the specific process is as follows:

[0067] Step 1: Take photos of the encapsulated product using a CT machine and extract CT photos from different angles;

[0068] Step 2, decomposing the bubble defect 9 and the circuit system structure 7 from the CT image of the potted product;

[0069] Step 3, three-dimensionally reconstructing the bubble defect 9 and the circuit system structure 7 respectively;

[0070] Step 4, constructing a fused three-dimensional graph of the bubble defect and the circuit system structure, specifically:

[0071] Example 2

[0072] The specific process of step 1 is:

[0073] like Figure 1 As shown, the potted product 4 to be tested is placed on the cylindrical pad 3 on the turntable 2 driven by the servo motor, ensuring that the axis of the potted product 4 to be tested is concentric with the axis of the turntable 2. The servo motor drives the turntable 2 to rotate. Every time it rotates 0.35°, the flat panel detector 1 collects and receives the CT image formed by the X-ray light source 5 penetrating the potted product 4 to be tested. The turntable 2 collects 1024 CT photos for one rotation, which means the step length is 0.35°. In this way, a set of CT photos taken every 0.35 degrees is formed. Figure 2 is one of them (a total of 1024). For the potted product 4 to be tested by the present invention, it is composed of four elements, namely, the workpiece shell 6, the circuit system structure 7, the epoxy potting material 8 and the bubble defect 9. Figure 2 It can be seen that the circuit system structure 7 is the most densely organized, followed by the workpiece shell 6, then the epoxy potting material 8, and finally the bubble defect 9. The main function of the epoxy potting material 8 is to protect the circuit system structure 7, and the existence of the bubble defect 9 destroys this protective effect. Generally speaking, the detection process requires the detection of the position of the bubble defect 9 relative to the circuit system structure 7, as well as the size and number of the bubble defect 9, so the bubble defect 9 and the circuit system structure 7 are key elements, and the existence of other elements in the image will only affect the operator's detection, so the present invention mainly decomposes and three-dimensionally reconstructs the two elements of the bubble defect 9 and the circuit system structure 7.

[0074] Example 3

[0075] The specific process of step 2 is:

[0076] For a series of Figure 2 The CT photo shown above first needs to separate the potted product 4 to be tested, because Figure 2 The CT image of the potted product 4 to be tested is shown in FIG. Figure 3 As shown ( Figure 3 is one of the 1024 images in total), from which it can be seen that there are workpiece shell 6, circuit system structure 7, epoxy potting material 8 and bubble defect 9. In this way, the three-dimensional reconstruction is carried out. Since the density of bubble defect 9 and epoxy potting material 8 is relatively close, it is not easy to distinguish the bubble characteristics. Therefore, it is necessary to use image processing methods. The present invention uses a watershed algorithm to define the threshold range according to the gray value of each element, so as to achieve image segmentation by selecting seed points in different areas of the image, and decompose the bubble defect 9 and the circuit system structure 7 into their corresponding 1024 corresponding angle images, and decompose 1024 CT images of the bubble defect 9 respectively. Figure 4 This is one of them. There are also 1024 CT images of the circuit system structure 7 decomposed, Figure 5 The purpose of separating these two diagrams is to provide the inspector with the positional relationship between the bubble defect and the circuit system structure, so that the inspector can judge whether the potted product is qualified.

[0077] Example 4

[0078] In step 2, the specific process of using the watershed algorithm to segment the bubble defect 9 and the circuit system structure 7 is as follows:

[0079] First, the CT image taken in step 1 is preprocessed as follows:

[0080] Denoising: Gaussian blur processing is performed on the CT image to smooth the noise points in the image and reduce their interference with the watershed algorithm.

[0081] Edge detection: By calculating the gradient of the image, the edge information in the image can be highlighted, providing a more accurate segmentation basis for the watershed algorithm.

[0082] Secondly, the watershed algorithm is used to perform image segmentation on the preprocessed CT image to segment the bubble defect 9 and the circuit system structure 7;

[0083] Finally, according to the segmentation results of the watershed algorithm, the marked image is updated and adjusted to ensure the accuracy and completeness of the segmentation results. For example, if an erroneous mark is found in the segmentation result, the mark can be corrected manually or automatically. Since the watershed algorithm is prone to over-segmentation, it is necessary to merge and optimize the segmented areas to obtain a more reasonable segmentation result. Merging rules can be formulated based on the area, shape, and grayscale characteristics of the region, similar regions can be merged into one region, and the segmentation results can be visualized so that users can observe and analyze intuitively. Different segmented areas can be represented by different colors, or the segmentation boundaries can be highlighted to help users better understand the segmentation of the image.

[0084] Example 5

[0085] The specific process of step 3 is:

[0086] According to Figure 4 A total of 1024 CT images of bubble defects are shown in the figure and Figure 5 A series of 1024 CT images of the circuit system structure 7 are shown, and three-dimensional reconstruction is performed using iterative reconstruction technology to obtain a corresponding three-dimensional image of the bubble defect 9 (see Figure 6 ) and a three-dimensional diagram of the circuit system structure 7 (see Figure 7 ).

[0087] Three-dimensional image of bubble defect 9 (see Figure 6 ) and the three-dimensional reconstruction process of the three-dimensional graphics of the circuit system structure 7 are the same, and are all performed according to the following iterative reconstruction process:

[0088] First, assume that the reconstructed 3D image is f(x, y, z). Discretize the 3D image f(x, y, z) into a 3D interval of n×n×n pixel units. Figure 1 As shown in the CT photo acquisition principle diagram, the X-rays pass through this three-dimensional area. At this time, the X-rays will pass through some pixel units in this three-dimensional area, but not contact other pixel units, and finally projected onto the flat-panel detector to obtain the corresponding induced image. Figure 8 shown.

[0089] Here, the pixel value of each unit can be regarded as a constant. For the convenience of calculation, all the elements in this three-dimensional image are converted into a one-dimensional column vector in the order of rows first and columns second, and front face first and back face second, that is, in the order of f(1,1,1), f(1,2,1), f(1,3,1),…f(1,n,1), f(2,1,1), f(2,2,1),…, f(2,n,1),…, f(1,1,2), f(1,2,3),…, f(n,n,n). Then the number of elements in this vector is n×n×n. Assume f j represents the pixel value of the jth pixel in the three-dimensional image, then the three-dimensional image can be represented by an N-dimensional vector F = [f 1 ,f 2 ,...,f N ] T Represented by, where N = n×n×n. The CT images taken are discretized according to the positions where the m×m rays reach the plane detector. The value of each pixel represents the projection value after a ray passes through the object, such as Figure 8 As shown. This CT image is also converted into an M-dimensional vector, then M = m × m, and P = [p 1 ,p2 ,..., p M T It is indicated that M is the total number of rays, and here these p 1 , p 2 ,..., p M values are the pixel values in the CT image and can be directly obtained from the CT image ([[]] Figure 4 and Figure 5 ). According to the physical process of imaging and the corresponding projection model, the relationship between the three-dimensional image vector and the projection data vector can be expressed as:

[0090]

[0091] In the formula, p i is the projection value of the i-th ray, that is, the pixel value on the CT image. w ij represents the relationship value of the j-th pixel to the projection value of the i-th ray, also called the weight factor. If it passes through this pixel, the weight factor is 1; otherwise, it is 0.

[0092] Equation (1) can be expanded into the following form of a linear equation system:

[0093]

[0094] The above formula is represented by a matrix as:

[0095] P = WF (3)

[0096] In the formula, F = [f 1 , f 2 ,..., f N is the image vector, P = [p 1 , p 2 ,..., p M T is the projection data vector, representing the ray projection value, and W is the projection matrix, or weight matrix.

[0097] In practical applications, both M and N in the above equation system are very large, and in most cases M << N. Therefore, the equation system is usually inconsistent, and it is very difficult to obtain the solution of the equation system using the traditional matrix direct inversion method. The present invention uses the simultaneous algebraic reconstruction technique SART to solve this problem.

[0098] The specific implementation steps of the SART algorithm are as follows:

[0099] Suppose:

[0100]

[0101] In the formula, W i,+ ​​It is an M×1 vector, which represents the weighted sum of all pixels for the i-th ray, and M represents the total number of rays under a projection angle.

[0102]

[0103] Where W +,j is an N×1 vector representing the weighted sum of all rays for the jth pixel. The pixel is corrected by the following process:

[0104] 1) Assign initial value to the position image vector:

[0105]

[0106] Wherein, j=1, 2, 3..., N represents the pixel index number, represents the iterative calculation value of the j-th pixel, Represents the iteratively calculated value of the j-th pixel in the previous iteration. In the first iteration, it represents the initialized pixel value.

[0107] 2) Calculate the theoretical projection value after the i-th ray passes through the object

[0108]

[0109] In the formula, The last iteratively calculated value of the j-th pixel.

[0110] 3) Calculate the error Δ between the theoretical projection grayscale value and the actual projection grayscale value of the i-th ray i :

[0111]

[0112] In the formula, p i It represents the actual projection gray value on the CT image corresponding to the i-th ray at the current projection angle.

[0113] 4) Calculate the correction value C of the jth pixel j :

[0114]

[0115] The present invention uses the errors of all rays at the current projection angle to calculate the error correction value of the j-th pixel:

[0116] 5) The gray value f of the pixel at the jth point j To make corrections:

[0117]

[0118] In the formula, λ represents the relaxation factor, which usually takes a value between 0 and 2. Corresponding adjustments are made during the calculation to ensure the accuracy of the calculation.

[0119] Repeat steps 2) to 5) for all rays under this projection angle to complete the correction of the next generation of reconstructed images in this direction. Traverse all projection angles (1024x0.35°) to complete one iteration. After the iterative calculation of the projection angle corresponding to the first CT photo is completed, select the CT photo of the second angle (+0.35°) for another iteration. At this time, the f of the previous iteration result is j As the initial value of this iteration, there is no need to perform step 1). Note that when calculating the ray traversing the three-dimensional image corresponding to the second CT photo, the ray source should be rotated 0.35° in the opposite direction of the three-dimensional image and the turntable. At this time, the corresponding weight w ij It will be different depending on the pixels it passes through.

[0120] After traversing the entire 1024 CT images, the corresponding iteration result values ​​of each element of the three-dimensional image are obtained. And according to the gray value f after correction in formula (10) j , it is not finished here, we need to add the above f j As the initial value, repeat steps 2) to 5) and then iterate and calculate the 1024 CT photos in turn according to the previous method. The grayscale value of all elements obtained by the first traversal of 1024 photos is F (I) The grayscale value of all elements obtained by traversing 1024 photos for the second time is F (II) After each traversal, the corresponding elements will be calculated to see if they converge. The convergence condition is:

[0121]

[0122] Among them, e represents the number of times the iterative calculation traverses 1024 CT photos. If the error of the calculation results of two consecutive times is less than 0.00001, it means that the element has converged and can be determined. There is no need to iteratively calculate this element in future calculations, and other elements can be calculated. If an element cannot converge according to formula (11), the convergence can be determined according to the maximum number of times the iterative calculation traverses 1024 CT photos. This maximum number can be set by yourself. This method can determine the values ​​of all elements on the three-dimensional image, and then convert the one-dimensional vector into a three-dimensional vector grayscale value f (x, y, z) in sequence, so as to obtain the effect of three-dimensional reconstruction. The above is the implementation process of the SART reconstruction algorithm.

[0123] Example 6

[0124] The specific process of step 4 is as follows: since the constructed bubble defect 9 and the circuit system structure 7 are three-dimensional spaces of the same size, the two images are merged according to the corresponding relationship. The corresponding elements on the two three-dimensional images are grayscale values. When the grayscale values ​​at the same position are compared, the minimum value of the two is taken as the grayscale value after the bubble defect 9 and the circuit system structure 7 are merged. In this way, the pixel coordinates of each pixel point in the bubble defect 9 and the circuit system structure 7 are used to merge the two three-dimensional images to obtain a three-dimensional image of the bubble defect 9 and the circuit system structure 7 merged together, as shown in FIG. Fig. 9 At this time, the inspector can drag the three-dimensional graph to view the position and size information of the bubble defect 9 relative to the circuit system structure 7 from another angle. Fig.10 This will give the inspector an intuitive feeling. Based on this, the present invention will focus on displaying the three-dimensional diagram of the bubble and the circuit system, removing some minor factors, so as to be more conducive to the operator to effectively judge the product quality.

Claims

1. A three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT, characterized by: The specific steps include: Step 1: Take photos of the potted product to be inspected by a CT machine and extract CT photos from different angles; Step 2, based on the CT image taken in step 1, decomposing the bubble defect and the circuit system structure; Step 3, three-dimensionally reconstructing the bubble defect and the circuit system structure respectively; Step 4: Based on the reconstruction result of step 3, a fused three-dimensional image of the bubble defect and the circuit system structure is constructed.

2. The three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT according to claim 1 is characterized in that: The specific process of step 1 is as follows: The potted product (4) to be inspected is placed on a cylindrical pad (3) on a turntable (2) driven by a servo motor, ensuring that the axis of the potted product (4) to be inspected is concentric with the axis of the turntable (2). The servo motor drives the turntable (2) to rotate. For every 0.35° rotation, the flat panel detector (1) will collect a CT image formed by the X-ray light source penetrating the potted product (4) to be inspected. The turntable (2) collects 1024 CT images for one rotation. The potted product (4) to be inspected includes a workpiece shell (6), a circuit system structure (7), an epoxy potting material (8) and a bubble defect (9).

3. The three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT according to claim 1 is characterized in that: The specific process of step 2 is: Using the watershed algorithm, 1024 images of bubble defects (9) and 1024 images of circuit system structures (7) are decomposed from the 1024 images of potted products to be inspected (4) obtained in step 1.

4. The three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT according to claim 3 is characterized in that: The specific process of step 3 is as follows: Step 3.1, assuming that the reconstructed 3D image is f(x, y, z), f j represents the pixel value of the jth pixel of the three-dimensional image f(x, y, z), then the three-dimensional image f(x, y, z) is represented by an N-dimensional vector F = [f1, f2, ..., f N ] T It means, where N = n × n × n; Step 3.2, discretize the CT image segmented in step 2 according to the position where the m×m rays reach the plane detector (1), so that the value of each pixel represents the projection value after a ray passes through the object; at the same time, each CT image is also converted into an M-dimensional vector, then M = m×m, and P = [p1, p2, ..., p M ] T Indicates that M is the total number of rays, where p1, p2, ..., p M The value is the pixel value in the CT photo. According to the physical process of imaging and the corresponding projection model, the relationship between the three-dimensional image vector and the projection data vector is expressed as: In the formula, p i is the projection value of the i-th ray, that is, the pixel value on the CT image, w ij The relationship value between the j-th pixel and the projection value of the i-th ray is also called the weight factor. If the ray passes through this pixel, the weight factor is 1, otherwise it is 0. Expand equation (1) into the following linear equation system: Formula (2) is expressed in matrix as: P=WF (3) Where, F=[f1,f2,...,f N ] is the image vector, P = [p1, p2, ..., p M ] T is the projection data vector, representing the ray projection value, and W is the projection matrix.

5. The three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT according to claim 4 is characterized in that: In step 3.2, the joint algebraic iterative reconstruction algorithm SART is used to solve formula (3), as follows: set up: Where W i,+ is an M×1 vector, which represents the weight sum of all pixels for the i-th ray, and M represents the total number of rays under a projection angle; Where W +,j is an N×1 vector representing the weighted sum of the jth pixel for all rays. The pixel is corrected by the following process: 1) Assign initial value to the position image vector: Wherein, j=1, 2, 3..., N represents the pixel index number, represents the iterative calculation value of the j-th pixel, Represents the iterative calculation value of the j-th pixel in the previous iteration. In the first iteration, it represents the initialized pixel value; 2) Calculate the theoretical projection value after the i-th ray passes through the object In the formula, The last iterative calculation value of the j-th pixel; 3) Calculate the error Δ between the theoretical projection grayscale value and the actual projection grayscale value of the i-th ray i : In the formula, p i Represents the actual projection grayscale value on the CT image corresponding to the i-th ray at the current projection angle; 4) Calculate the correction value C of the jth pixel j : 5) Use the errors of all rays at the current projection angle to calculate the error correction value of the j-th pixel: the gray value f of the pixel at the j-th point j To make corrections: In the formula, λ represents the relaxation factor; 6) Repeat steps 2) to 5) for all rays under the projection angle to complete the correction of the next generation reconstructed image in this direction, and traverse all projection angles to complete one iteration; 7) After traversing the entire 1024 CT photos, the corresponding iteration result values ​​of each element of the three-dimensional image are obtained And according to formula (10), the gray value f is obtained after correction j ; 8) The f obtained in step 7) j As the initial value, repeat steps 2) to 5) and iterate the calculation of 1024 CT photos in turn. The grayscale value of all elements obtained by traversing 1024 photos for the first time is F (I) The grayscale value of all elements obtained by traversing 1024 photos for the second time is F (II) After each traversal, it will calculate whether the corresponding elements converge. The convergence condition is: Here, e represents the number of times the iterative calculation traverses 1024 CT images.

6. The three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT according to claim 5 is characterized in that: In step 3.2, in step 6), after the iterative calculation of the projection angle corresponding to the first CT photo is completed, the second CT photo with an angle of +0.35° is selected for another iteration. At this time, the f of the previous iteration result is j As the initial value of this iteration, when calculating the ray passing through the three-dimensional image corresponding to the second CT photo, the ray source is rotated 0.35° around the three-dimensional image in the opposite direction of the turntable rotation direction.

7. The three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT according to claim 6 is characterized in that: In step 3.2, in step 8), if the error between two consecutive calculation results is less than 0.00001, it means that the element has converged; if an element cannot converge according to formula (11), the convergence can be determined by the maximum number of iterative calculations traversing 1024 CT images.

8. The three-dimensional reconstruction method for detecting potting bubble defects based on industrial CT according to claim 7 is characterized in that: The specific process of step 4 is as follows: since the constructed bubble defect (9) and the circuit system structure (7) are three-dimensional spaces of the same size, the grayscale values ​​of the bubble defect (9) and the circuit system structure (7) at the same position are compared, and the minimum value of the two is taken as the grayscale value after the bubble defect (9) and the circuit system structure (7) are fused, thereby obtaining a three-dimensional graphic in which the bubble defect (9) and the circuit system structure (7) are fused together.

Images