A circuit board back-drilling stub detection method and detection system

Abstract

The present invention discloses a method and system for detecting stubs in back-drilled holes in circuit boards. The method comprises the following steps: obtaining a set of cross-sectional images generated by scanning the circuit board to be inspected using computed tomography technology, the cross-sectional image set comprising multiple cross-sectional images arranged in order of cross-sectional height; sequentially analyzing the cross-sectional images in the image set to determine the position of the end of the remaining stub, recorded as a first position; determining the position of the layer electrically connected to the remaining stub and closest to the end of the remaining stub, recorded as a second position; and calculating the distance between the first position and the second position to determine whether the back-drilled hole stub is qualified. The present invention utilizes tomography and artificial intelligence technologies to accurately detect the positional relationship between the stub and the signal layer within the back-drilled hole, without requiring expensive cameras or damaging the finished circuit board.

Description

Technical Field

[0001] The present invention relates to the field of circuit component detection, and in particular to a circuit board back-drilled STUB detection method and detection system. Background Art

[0002] After the through-hole metallization operation is performed on the circuit board, the stubs in the through-holes will introduce parasitic capacitance and inductance, causing signal reflection and crosstalk, affecting signal integrity.

[0003] Backdrilling is a technology used to optimize signal integrity in multi-layer PCB (printed circuit board) manufacturing. Its core principle is to precisely control the drilling depth and remove the unused portion of the through-hole (i.e., the stub) from the back of the PCB. This reduces the parasitic effects of the through-hole, minimizes reflections, crosstalk, and delays in signal transmission, and significantly reduces discontinuities in signal transmission. The remaining portion after removing the stub is usually called a stub.

[0004] The purpose of retaining the stub is to achieve electrical connection between the first several layers on the front side of the circuit board, usually to achieve electrical connection between the layers on the surface of the circuit board and the signal layers inside the circuit board.

[0005] The backdrilling depth must be strictly controlled. If the remaining stub length does not meet the requirements, the performance of the PCB will be seriously affected.

[0006] Because through-holes are relatively narrow and the thickness of the circuit board is not large, optical inspection not only requires expensive cameras but also has low precision. Current optical inspection or manual inspection only determines whether the remaining stub is connected to the signal layer and disconnected from the power layer / ground layer, but cannot make more detailed inspection of its specific location.

[0007] Another method for stub testing involves destroying the finished product. This involves cutting the first product produced by the equipment to test whether the remaining stub in the back-drilled hole meets the requirements. If it does not meet the requirements, the equipment needs to be debugged until the first product passes the test before subsequent production can proceed. This method not only damages the test product and seriously affects efficiency, but also cannot eliminate the risk of product failure caused by errors caused by long-term equipment operation.

[0008] The disclosure of the above background technology content is only used to assist in understanding the concept and technical solution of this application. It does not necessarily belong to the prior art of this application, nor does it necessarily provide technical guidance. In the absence of clear evidence that the above content has been disclosed before the filing date of this application, the above background technology should not be used to evaluate the novelty and creativity of this application. Summary of the Invention

[0009] The object of the present invention is to provide a method for accurately detecting the positional relationship between a stub and a signal layer in a back-drilled hole by using a tomography technique.

[0010] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0011] A method for detecting stubs in back-drilled holes of a circuit board is provided, wherein the remaining stubs in the holes after back-drilling are detected. The method comprises the following steps:

[0012] Acquire a cross-sectional image set generated by scanning the circuit board to be inspected using electronic computed tomography technology, wherein the cross-sectional image set includes a plurality of cross-sectional images arranged in order of heights of the cross sections;

[0013] Using a pre-built first AI model, sequentially analyzing the cross-sectional images in the image set, and determining the position of the end of the remaining stub, which is recorded as a first position;

[0014] Using the pre-built second AI model, determine the position of a layer electrically connected to the remaining stub and closest to the end of the remaining stub, which is recorded as a second position;

[0015] The distance between the first position and the second position is calculated to determine whether the back-drilled hole Stub is qualified.

[0016] Furthermore, based on any one of the above technical solutions or a combination of multiple technical solutions, the circuit board backdrilling stub detection method provided by the embodiment of the present invention further includes:

[0017] inputting the cross-sectional images of the cross-sectional image set into the first AI model in sequence;

[0018] The first AI model determines a cross-sectional image corresponding to the first position;

[0019] The first AI model is trained in the following way:

[0020] Obtaining a sample image: Scanning the sample circuit board using electronic computed tomography technology and generating a cross-sectional image set, wherein the cross-sectional image set includes a plurality of cross-sectional images arranged in order of heights of the cross sections;

[0021] Manual labeling: The cross-sectional images that reflect the end features of the remaining stub are labeled with the first type of labels;

[0022] Repeatedly acquire sample images and manually label them to form a learning sample set;

[0023] Divide the learning sample set into a training set and a validation set;

[0024] Input the learning samples of the training set into the basic model for iterative training;

[0025] The iteratively trained model is verified using the learning samples of the validation set. If the verification fails, the iterative training is continued; if the verification passes, the first AI model is obtained.

[0026] Furthermore, based on any one of the above technical solutions or a combination of multiple technical solutions, the first AI model estimates the probability of each input cross-sectional image having the end feature of the remaining stub;

[0027] The first AI model determines the cross-sectional image corresponding to the first position in the following manner:

[0028] If the estimated probability value of the first image having the end feature of the remaining stub is less than a preset first probability threshold, and the estimated probability value of the second image adjacent to the first image having the end feature of the remaining stub reaches the first probability threshold, then the cross-sectional image corresponding to the first position is determined based on the first image or the second image.

[0029] Furthermore, based on any one of the above technical solutions or a combination of multiple technical solutions, the first AI model determines the cross-sectional image corresponding to the first position in any of the following ways:

[0030] Method 1 is to select an image as the cross-sectional image corresponding to the first position from a plurality of continuous cross-sectional images, with the second image being the first image and excluding the first image;

[0031] The second method is to select an image as the cross-sectional image corresponding to the first position from a plurality of continuous cross-sectional images, with the first image being the first image and excluding the second image;

[0032] Method three is to preset a second probability threshold that is greater than or equal to the first probability threshold; the first AI model determines a third image that satisfies the following conditions: the probability value of the third image having the end feature of the remaining stub is estimated to reach the second probability threshold; and from a plurality of consecutive cross-sectional images, starting with the third image, one image is selected as the cross-sectional image corresponding to the first position;

[0033] Method 4 is to calculate the difference between two adjacent cross-sectional images in a first group of continuous cross-sectional images with the second image as the first image and excluding the first image. If the difference is greater than a preset difference threshold, calculate the difference between two adjacent cross-sectional images in the next group; repeat the above steps until the difference between the k-th and k+1-th cross-sectional images after the second image is found to be less than or equal to the preset difference threshold, where k is a positive integer; and use the [k / 2]-th cross-sectional image after the second image as the cross-sectional image corresponding to the first position, where [ ] is an operator symbol for rounding, rounding up, or rounding down.

[0034] Furthermore, according to any one of the above technical solutions or a combination of multiple technical solutions, the set value of the first probability threshold is determined by:

[0035] Inputting the test image set into the trained first AI model to obtain a probability value for each cross-sectional image in the test image set being estimated to have an end feature of a remaining stub;

[0036] Manually determining the target cross-sectional image corresponding to the position of the end of the remaining stub in the test image set;

[0037] Determining the probability values of the adjacent images on both sides of the target cross-sectional image being estimated;

[0038] A reference value of the first probability threshold is determined within an interval range of the probability values estimated for the adjacent images on both sides.

[0039] Furthermore, according to any one of the above technical solutions or a combination of multiple technical solutions, multiple test image sets are used to obtain corresponding reference values of multiple first probability thresholds;

[0040] After performing denoising processing on the reference values of the multiple first probability thresholds, an average value is obtained to serve as the first probability threshold.

[0041] Furthermore, based on any one of the above technical solutions or a combination of multiple technical solutions, the circuit board backdrilling stub detection method provided by the embodiment of the present invention further includes:

[0042] inputting the cross-sectional images of the cross-sectional image set into the second AI model in sequence;

[0043] The second AI model determines a cross-sectional image corresponding to the second position;

[0044] The second AI model is trained in the following way:

[0045] Obtaining a sample image: Scanning the sample circuit board using electronic computed tomography technology and generating a cross-sectional image set, wherein the cross-sectional image set includes a plurality of cross-sectional images arranged in order of heights of the cross sections;

[0046] Manual labeling: The second type of label is annotated on the cross-sectional images that reflect the characteristics of the layer closest to the stub end and connected to it;

[0047] Repeatedly acquire sample images and manually label them to form a learning sample set;

[0048] Divide the learning sample set into a training set and a validation set;

[0049] Input the learning samples of the training set into the basic model for iterative training;

[0050] The iteratively trained model is verified using the learning samples of the validation set. If the verification fails, the iterative training is continued; if the verification passes, the second AI model is obtained.

[0051] Further, based on any one of the above technical solutions or a combination of multiple technical solutions, the second AI model estimates the probability of each input cross-sectional image having a layer feature that is closest to and connected to the stub end, where the layer feature includes a line morphology feature connected to the stub;

[0052] The second AI model determines the cross-sectional image corresponding to the second position in the following manner:

[0053] performing sequential identification on a set of cross-sectional images whose cross-sectional order is from the back side to the front side of the circuit board, and if a probability value of a fourth image being first estimated to have the layer feature is less than a preset third probability threshold, and a probability value of a fifth image adjacent to the fourth image being estimated to have the layer feature reaches the preset third probability threshold, determining the cross-sectional image corresponding to the second position based on the fourth image or the fifth image;

[0054] A set of cross-sectional images with a cross-sectional sorting direction from the front to the back of the circuit board is identified in sequence. If the probability value of the fourth image estimated for the last time to have the layer feature is less than a preset third probability threshold, and the probability value of the fifth image adjacent to it estimated to have the layer feature reaches the third probability threshold, the cross-sectional image corresponding to the second position is determined based on the fourth image or the fifth image.

[0055] Furthermore, based on any one of the above technical solutions or a combination of multiple technical solutions, the second AI model determines the cross-sectional image corresponding to the second position in any of the following ways:

[0056] The first method is to select an image from a plurality of continuous cross-sectional images, with the fifth image being the first image and excluding the fourth image, as the cross-sectional image corresponding to the second position;

[0057] Method 2 is to preset a fourth probability threshold that is greater than the third probability threshold; the second AI model determines a sixth image that satisfies the following conditions: the probability value of the sixth image having the layer feature is estimated to reach the fourth probability threshold; and select an image from a plurality of consecutive cross-sectional images, the first of which is the sixth image, as the cross-sectional image corresponding to the second position;

[0058] Method three is to determine, among a plurality of consecutive cross-sectional images, starting with the fifth image and excluding the fourth image, a seventh image and an eighth image that meet the following conditions and are adjacent: if the second AI model first estimates that the probability value of the eighth image having the layer feature is less than the third probability threshold and the probability value of the seventh image having the layer feature reaches the third probability threshold, then determine the image centered between the fifth and seventh images as the cross-sectional image corresponding to the second position;

[0059] Method 4 is to calculate the difference in probability values of two adjacent cross-sectional images in a first group of multiple continuous cross-sectional images, starting with the fifth image and excluding the fourth image, and if the difference is less than a preset difference threshold, calculate the difference in probability values of two adjacent cross-sectional images in a next group of two cross-sectional images; repeat the above steps until the difference in corresponding probability values of the p-th and p+1-th cross-sectional images after the fifth image is found to be greater than or equal to the difference threshold, where p is a positive integer; and use the [p / 2]-th cross-sectional image after the fifth image as the cross-sectional image corresponding to the second position, wherein [ ] is an operator symbol for rounding, rounding up, or rounding down;

[0060] Method five is to determine, among a plurality of continuous cross-sectional images with the fifth image as the first one and excluding the fourth image, a ninth image that first satisfies the following conditions: the probability value of the ninth image being estimated to have the layer feature is greater than both the corresponding probability values of the previous image and the next image; and the ninth image is used as the cross-sectional image corresponding to the second position.

[0061] Furthermore, based on any one of the above technical solutions or a combination of multiple technical solutions, the set value of the third probability threshold is determined by the following method:

[0062] Inputting the test image set into the trained second AI model to obtain a probability value for each cross-sectional image in the test image set being estimated to have the layer feature;

[0063] Manually determining a target cross-sectional image corresponding to a location of a layer electrically connected to the remaining stub in the test image set;

[0064] Determining the probability values of the adjacent images on both sides of the target cross-sectional image being estimated;

[0065] A reference value of the third probability threshold is determined within an interval range of the probability values estimated for the adjacent images on both sides.

[0066] Furthermore, according to any one of the above technical solutions or a combination of multiple technical solutions, multiple test image sets are used to obtain corresponding reference values of multiple third probability thresholds;

[0067] After performing denoising processing on the reference values of the multiple third probability thresholds, an average value is obtained to serve as the third probability threshold.

[0068] Further, based on any one of the above technical solutions or a combination of multiple technical solutions, the distance between the first position and the second position is calculated by the following method:

[0069] determining the arrangement sequence numbers of the cross-sectional image corresponding to the first position and the cross-sectional image corresponding to the second position in the cross-sectional image set respectively;

[0070] Calculate the difference between the sequence numbers corresponding to the first position and the second position;

[0071] Determining a scanning step distance according to the working parameters of the electronic computed tomography scan, wherein the scanning step distance is the distance between two consecutive scans of the cross-section of the same object;

[0072] The distance between the first position and the second position is determined by calculating the product of the scanning step distance and the sequence number difference.

[0073] Furthermore, based on any one of the above technical solutions or a combination of multiple technical solutions, whether the back-drilled stub is qualified is determined by the following method:

[0074] Set the Stub qualified length range [L min ,L max ], where L min L is the lower limit of the stub qualified length range. max The upper limit of the stub qualified length range.

[0075] Define the distance between the first position and the second position as l 0, if L min ≤ l 0≤L max , the back-drilled stub is determined to be qualified, otherwise the back-drilled stub is determined to be unqualified.

[0076] According to another aspect of the present invention, a circuit board backdrilling stub detection system is provided, comprising a computerized tomography device and a processor, wherein the computerized tomography device is configured to scan a circuit board to be inspected and generate a set of cross-sectional images;

[0077] The processor is configured to execute the steps of the detection method described above.

[0078] Furthermore, based on any one of the technical solutions or a combination of multiple technical solutions described above, the electronic computed tomography device includes one or more light pipes and flat panel detectors corresponding to the light pipes.

[0079] The beneficial effects brought about by the technical solution provided by the present invention are as follows:

[0080] a. Using tomography technology instead of optical imaging, it is possible to obtain a set of cross-sectional images, enabling accurate detection of the positional relationship between the stub and the signal layer within the back-drilled hole;

[0081] b. Accurately locate the cross-sectional image of the stub end and the cross-sectional image of the signal layer in the set of cross-sectional images acquired by the tomography scan. This is converted into the actual height difference between the stub end and the signal layer, which serves as the basis for detection.

[0082] c. Ability to conduct inspections during the PCB production process, reducing the defective rate of finished PCBs and lowering the difficulty of repairs;

[0083] d. Based on an intelligent algorithm, the system can accurately identify the center of the inclined surface at the end of the stub as the first position and the center of the signal layer as the second position. This system is particularly effective in stub inspection on precision circuit boards, leveraging its high measurement accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0085] Figure 1 A flowchart of a circuit board backdrilling stub detection method provided by an exemplary embodiment of the present invention;

[0086] Figure 2A schematic diagram of a scanning process of an electronic computed tomography device of a circuit board back-drilled STUB detection system provided by an exemplary embodiment of the present invention;

[0087] Figure 3 A schematic partial cross-sectional view of a circuit board after secondary back drilling is provided as an exemplary embodiment of the present invention;

[0088] Figure 4 A flow chart of a conventional backdrilling process is provided for an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0089] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0090] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0091] The implementation process of back drilling process is as follows Figure 4 As shown:

[0092] Step 1: Initial through-hole processing: First, complete the through-hole processing that goes through the entire thickness of the PCB board.

[0093] Step 2: Metallize (copper coating) the through holes.

[0094] Step 3: Secondary back drilling: Use a larger diameter drill bit (usually 0.15~0.2mm larger than the original through-hole diameter) to drill from the back side of the PCB to remove the copper barrel stub to a specified depth.

[0095] Step 4: Remove burrs from the drilled holes and optionally fill them with resin to improve mechanical stability.

[0096] Step 5: Grind the board to remove excess material (overflowing resin) from the upper and lower surfaces of the circuit board.

[0097] The drilling depth of the secondary backdrill in step three must be strictly controlled. Excessive copper stub removal can lead to severe reflections, crosstalk, and delays in signal transmission. Excessive copper stub removal can also affect the stability of the connection between the two layers closest to the front of the circuit board. Multilayer circuit boards are precision devices with close proximity between layers, which increases the difficulty of verifying that the backdrill depth meets design requirements.

[0098] The cross section of the back-drilled hole of the circuit board obtained after step 3 is as follows Figure 3 As shown, after removing the copper tube stub of the specified depth, the remaining Stub300 is left. Figure 3 As can be seen in the figure, the end of Stub300 in the hole is inclined due to the influence of the larger diameter drill bit used in the secondary back drilling.

[0099] The present invention aims to use tomography technology to accurately detect the positional relationship between the Stub 300 and the signal layer 400 in the back-drilled hole. In one embodiment of the present invention, a circuit board back-drilled hole Stub detection method is provided to detect the remaining Stub in the hole after back-drilling. Figure 1 As shown, the detection method includes the following steps:

[0100] S1: obtaining a cross-sectional image set generated by scanning a circuit board to be inspected using electronic computed tomography technology, wherein the cross-sectional image set includes a plurality of cross-sectional images arranged in order of heights of the cross sections;

[0101] The computerized tomography (CT) device includes a light tube and a flat panel detector corresponding to the light tube. One method is to see Figure 2 There are four light pipes and four flat-panel detectors, each with a one-to-one correspondence. As indicated by the dotted arrows, the first light pipe 110, second light pipe 120, third light pipe 130, and fourth light pipe 140, distributed along a circumference, operate sequentially, illuminating the same circuit board to be inspected. Correspondingly, the first flat-panel detector 210, second flat-panel detector 220, third flat-panel detector 230, and fourth flat-panel detector 240 sequentially cooperate with the corresponding light pipes to achieve imaging. Another approach is to install only one light pipe and one flat-panel detector, which are then moved sequentially to multiple positions to illuminate and image the circuit board to be inspected at different angles. Regardless of the approach, the illumination distances of the light pipes at different positions are roughly equal, and the illumination angles are roughly the same as the angle between them and the plane of the circuit board.

[0102] CT imaging obtains an image set. In this embodiment, it is assumed that the back of the circuit board to be inspected faces the scanning device, so the cross-sectional image set is based on Figure 3 As an example, a collection of cross sections is sorted from bottom to top. The following is an example to illustrate. It should be understood that without paying any creative labor, Figure 3 The set of cross sections ordered from top to bottom in the figure can also be used as a cross-sectional image set to implement the technical solution of the present invention, and the corresponding technical solutions distinguished by this also fall within the scope of protection claimed by the present invention.

[0103] S2: Using the pre-built first AI model, sequentially analyze the cross-sectional images in the image set to determine the location of the end of the remaining Stub, which is recorded as the first location; taking a round drill bit as an example, the Stub feature is a circular ring in the cross section. Figure 3 For example, in the bottom-up cross-sectional image set, there is no stub feature in the previous cross-sectional image. However, due to the larger diameter drill bit used in the secondary backdrilling, the end of the stub 300 in the hole is inclined, and the stub feature appears at the end.

[0104] Specifically, the first position is located using a pre-built first AI model: cross-sectional images of the cross-sectional image set are sequentially input into the first AI model, and the first AI model estimates the probability of each input cross-sectional image having the end feature of the remaining stub; the first AI model determines the cross-sectional image corresponding to the first position by the following method:

[0105] If the estimated probability value of the first image having the end feature of the remaining stub is less than a preset first probability threshold, and the estimated probability value of the second image adjacent to the first image having the end feature of the remaining stub reaches the first probability threshold, then the cross-sectional image corresponding to the first position is determined based on the first image or the second image.

[0106] It can be determined that, according to the different settings of the first probability threshold, the first image and the second image can represent Figure 3 The boundary between the lower end of Stub300 and the non-Stub area (i.e., the lowest height position of the lower end inclined surface) can also represent Figure 3 The interface at any point on the lower inclined surface of Stub300 can also be characterized Figure 3 The interface at the highest height position of the lower end inclined surface of Stub300 is marked.

[0107] In different embodiments, the position corresponding to the first image recognized to have the Stub feature (ie, the second image) may be used as the first position, or Figure 3As shown, a position slightly above the second image is used as the first position 310 .

[0108] Specifically, the first AI model is trained in the following ways:

[0109] Obtaining a sample image: Scanning the sample circuit board using electronic computed tomography technology and generating a cross-sectional image set, wherein the cross-sectional image set includes a plurality of cross-sectional images arranged in order of heights of the cross sections;

[0110] Manual labeling: The cross-sectional images that reflect the end features of the remaining stub are labeled with the first type of labels;

[0111] Repeatedly acquire sample images and manually label them to form a learning sample set;

[0112] Divide the learning sample set into a training set and a validation set;

[0113] The learning samples of the training set are input into the basic model for iterative training. The training process is that the first AI model analyzes the end features of the remaining Stub for each cross-sectional image in each sample, and estimates the probability that each cross-sectional image is the cross-sectional image corresponding to the first position. The expected goal of the training is to estimate that the cross-sectional image with the highest probability is the cross-sectional image in this sample marked with the first type of label. If the requirements for model accuracy are lowered, the expected goal of the training can also be lowered to the cross-sectional image with the highest probability and the cross-sectional image marked with the first type of label being within a preset adjacent range, for example, the two are separated by no more than 3 image sets in the cross-sectional image set of the sample;

[0114] The iteratively trained model is verified using the learning samples of the validation set. If the verification fails, the iterative training is continued; if the verification passes, the first AI model is obtained.

[0115] In a specific embodiment, manual labeling of the first type of label can be performed as follows: display each cross-sectional image in the sample as an image, and visually observe the weakest cross-sectional image where the end feature of the remaining stub gradually appears, assuming that the image in the image set has the sequence number 100. As the end feature of the remaining stub becomes clearer, a cross-sectional image is observed where the end feature of the remaining stub reaches a stable clarity, assuming that the image in the image set has the sequence number 190. If the image sequence numbers in the image set are consecutive integers, the first type of label is labeled on the cross-sectional image with the sequence number 145.

[0116] The setting of the first probability threshold may affect the accuracy of the estimation result of the first AI model. In one embodiment, the setting value of the first probability threshold is determined by:

[0117] Inputting the test image set into the trained first AI model to obtain a probability value for each cross-sectional image in the test image set being estimated to have an end feature of a remaining stub;

[0118] Manually determining the target cross-sectional image corresponding to the position of the end of the remaining stub in the test image set;

[0119] Determining the probability values of the adjacent images on both sides of the target cross-sectional image being estimated;

[0120] A reference value of the first probability threshold is determined within an interval range of the probability values estimated for the adjacent images on both sides.

[0121] In a further embodiment, multiple test image sets can be input into the first AI model in batches to obtain corresponding reference values of multiple first probability thresholds; after denoising the reference values of the multiple first probability thresholds, the average value is calculated as the first probability threshold.

[0122] S3: using a pre-built second AI model, determining a position of a layer electrically connected to the remaining stub and closest to the end of the remaining stub, which is recorded as a second position;

[0123] For example, subsequent cross-sectional images are compared in a direction from the first image to the second image to find adjacent fourth and fifth images that meet the following conditions: the fourth image does not recognize layer features, the fifth image recognizes layer features, and the layer features include line morphological features connected to the stub;

[0124] The layer features are shown in the cross-sectional image as metal strips intersecting with the ring representing the Stub feature, and the second position 320 to be found is Figure 3 The height of the first layer connected to the Stub from bottom to top.

[0125] Specifically, the second position is located using a pre-built second AI model: the second AI model estimates the probability of each input cross-sectional image having a layer feature that is closest to and connected to the stub end, wherein the layer feature includes a line morphology feature connected to the stub; the second AI model determines the cross-sectional image corresponding to the second position by the following method:

[0126] performing sequential identification on a set of cross-sectional images whose cross-sectional order is from the back side to the front side of the circuit board, and if a probability value of a fourth image being first estimated to have the layer feature is less than a preset third probability threshold, and a probability value of a fifth image adjacent to the fourth image being estimated to have the layer feature reaches the preset third probability threshold, determining the cross-sectional image corresponding to the second position based on the fourth image or the fifth image;

[0127] A set of cross-sectional images with a cross-sectional sorting direction from the front to the back of the circuit board is identified in sequence. If the probability value of the fourth image estimated for the last time to have the layer feature is less than a preset third probability threshold, and the probability value of the fifth image adjacent to it estimated to have the layer feature reaches the third probability threshold, the cross-sectional image corresponding to the second position is determined based on the fourth image or the fifth image.

[0128] From this, we can determine that the fourth image and the fifth image are Figure 3 The boundary between the lower surface of the signal layer 400 and the non-signal layer is shown in FIG.

[0129] In different embodiments, the position corresponding to the first image identified to have the layer feature (ie, the fifth image) may be used as the second position, or Figure 3 As shown, a position slightly above the fifth image is used as the second position 320 .

[0130] The second AI model is trained in the following way:

[0131] Obtaining a sample image: Scanning the sample circuit board using electronic computed tomography technology and generating a cross-sectional image set, wherein the cross-sectional image set includes a plurality of cross-sectional images arranged in order of heights of the cross sections;

[0132] Manual labeling: The second type of label is annotated on the cross-sectional images that reflect the characteristics of the layer closest to the stub end and connected to it;

[0133] Repeatedly acquire sample images and manually label them to form a learning sample set;

[0134] Divide the learning sample set into a training set and a validation set;

[0135] Inputting learning samples of the training set into the basic model for iterative training, the training process is that the second AI model analyzes the layer features electrically connected to the remaining stub and closest to the end of the remaining stub for each cross-sectional image in each sample, and estimates the probability that each cross-sectional image is the cross-sectional image corresponding to the second position. The expected goal of the training is to estimate that the cross-sectional image with the highest probability is the cross-sectional image in this sample marked with the second type of label. If the requirements for model accuracy are lowered, the expected goal of the training can also be lowered to estimate that the cross-sectional image with the highest probability and the cross-sectional image marked with the second type of label are within a preset adjacent range, for example, the two are separated by no more than 3 image sets in the cross-sectional image set of the sample;

[0136] The iteratively trained model is verified using the learning samples of the validation set. If the verification fails, the iterative training is continued; if the verification passes, the second AI model is obtained.

[0137] In a specific embodiment, manual annotation of the second type of label can be performed as follows: displaying each cross-sectional image of the sample in the form of an image, visually observing a cross-sectional image in which the layer feature electrically connected to the remaining stub and closest to the end of the remaining stub gradually appears and becomes weakest, assuming that the corresponding image in the image set is numbered 1000; as the layer feature becomes clearer, a cross-sectional image is observed in which the layer feature reaches a stable clarity, assuming that the corresponding image in the image set is numbered 1150; if the image sequence numbers in the image set are consecutive integers, the second type of label is annotated on the cross-sectional image with image sequence number 1075.

[0138] The setting of the third probability threshold may affect the accuracy of the estimation result of the second AI model. In one embodiment, the setting value of the third probability threshold is determined by:

[0139] Inputting the test image set into the trained second AI model to obtain a probability value for each cross-sectional image in the test image set being estimated to have the layer feature;

[0140] Manually determining a target cross-sectional image corresponding to a location of a layer electrically connected to the remaining stub in the test image set;

[0141] Determining the probability values of the adjacent images on both sides of the target cross-sectional image being estimated;

[0142] A reference value of the third probability threshold is determined within an interval range of the probability values estimated for the adjacent images on both sides.

[0143] In a further embodiment, multiple test image sets can be input into the second AI model in batches to obtain corresponding reference values of multiple third probability thresholds; after denoising the reference values of the multiple third probability thresholds, the average value is calculated as the third probability threshold.

[0144] S4: Calculate the distance between the first position and the second position to determine whether the back-drilled hole Stub is qualified.

[0145] Specifically, the arrangement serial numbers of the cross-sectional image corresponding to the first position and the cross-sectional image corresponding to the second position in the cross-sectional image set are determined respectively; the difference in serial numbers corresponding to the first position and the second position is calculated; the scanning step distance is determined according to the working parameters of the electronic computed tomography scan, and the scanning step distance is the distance between the cross-sections of the same object scanned twice before and after; the distance between the first position and the second position is determined by calculating the product of the scanning step distance and the serial number difference.

[0146] like Figure 3 As shown, the height distance between the first position 310 and the second position 320 is l 0, set the Stub qualified length range [L min ,L max ], where L min L is the lower limit of the stub qualified length range. max The upper limit of the stub qualified length range; if L min ≤ l 0≤L max , the back-drilled stub is determined to be qualified, otherwise the back-drilled stub is determined to be unqualified.

[0147] The following is a detailed description of an embodiment of an AI model determining cross-sectional images corresponding to the first position and the second position:

[0148] First embodiment:

[0149] According to the probability estimation result of the first AI model, the first image and the second image are located. In this embodiment, the setting of the first probability threshold makes the first image and the second image represent Figure 3 The boundary between the lower end of the Stub 300 and the non-Stub area (ie, the lowest height position of the lower end inclined surface) is marked.

[0150] Among a plurality of consecutive cross-sectional images, with the second image as the first image and excluding the first image, one image is selected as the cross-sectional image corresponding to the first position. For example, the second image can be directly selected as the cross-sectional image corresponding to the first position, or the fifth image, for example, is selected upward from the second image as the cross-sectional image corresponding to the first position. The specific image to be selected can be determined by reference to the stub structure of the circuit board, such as the length of the inclined surface at the lower end of the stub formed by the drill bit used for back drilling. The longer the length, the farther the cross-sectional image from the second image is selected.

[0151] Second embodiment:

[0152] According to the probability estimation result of the first AI model, the first image and the second image are located. In this embodiment, the setting of the first probability threshold makes the first image and the second image represent Figure 3 The interface at the highest height position of the lower end inclined surface of Stub300 or the interface at any position on the inclined surface.

[0153] Among a plurality of consecutive cross-sectional images, starting with the first image and excluding the second image, one image is selected as the cross-sectional image corresponding to the first position. For example, the first image is directly selected as the cross-sectional image corresponding to the first position, or the fifth image, for example, is selected downward from the first image as the cross-sectional image corresponding to the first position. The specific image to be selected can be determined by reference to the stub structure of the circuit board, such as the length of the inclined surface at the lower end of the stub formed by the drill bit used for back drilling. The longer the length, the farther the cross-sectional image from the first image is selected.

[0154] Third embodiment:

[0155] According to the probability estimation result of the first AI model, the first image and the second image are located. In this embodiment, the setting of the first probability threshold makes the first image and the second image represent Figure 3 The boundary between the lower end of the Stub 300 and the non-Stub area (ie, the lowest height position of the lower end inclined surface) is marked.

[0156] A second probability threshold is preset, which is greater than or equal to the first probability threshold, so that the first image and the second image represent Figure 3 The first AI model determines a third image that satisfies the following conditions: a probability value of the third image having the end feature of the remaining Stub is estimated to reach the second probability threshold; and among a plurality of consecutive cross-sectional images with the third image as the first one, selects an image as the cross-sectional image corresponding to the first position, preferably selecting the third image as the cross-sectional image corresponding to the first position.

[0157] Fourth embodiment:

[0158] According to the probability estimation result of the first AI model, the first image and the second image are located. In this embodiment, the setting of the first probability threshold makes the first image and the second image represent Figure 3 The boundary between the lower end of the Stub 300 and the non-Stub area (ie, the lowest height position of the lower end inclined surface) is marked.

[0159] In a plurality of continuous cross-sectional images with the second image as the first image and excluding the first image, the difference between two adjacent cross-sectional images in the first group is calculated. If the difference is greater than a preset difference threshold, the difference between two adjacent cross-sectional images in the next group is calculated. The above steps are repeated until the difference between the k-th cross-sectional image after the second image and the k+1-th cross-sectional image is less than or equal to the preset difference threshold, where k is a positive integer. The [k / 2]-th cross-sectional image after the second image is used as the cross-sectional image corresponding to the first position, where [ ] is an operator symbol for rounding, rounding up, or rounding down.

[0160] See also Figure 3 The lower end of Stub 300 is shaped like an eight, so the corresponding cross-sectional images all have circular rings. Furthermore, the size of the circular rings decreases from large to small in the cross-sectional images from bottom to top. This embodiment utilizes this fact to locate the highest position of the inclined surface by calculating the difference between two adjacent cross-sectional images. The [k / 2]th cross-sectional image after the second image is then used as the cross-sectional image at the center of the inclined surface, i.e. Figure 3 A first position 310 is shown.

[0161] Fifth embodiment:

[0162] According to the probability estimation result of the second AI model, the fourth image and the fifth image are located. In this embodiment, the third probability threshold is set so that the fourth image and the fifth image represent Figure 3 The boundary between the lower surface of the signal layer 400 and the non-signal layer is shown in FIG.

[0163] Among the plurality of consecutive cross-sectional images, starting with the fifth image and excluding the fourth image, an image is selected as the cross-sectional image corresponding to the second position. For example, the fifth image may be directly selected as the cross-sectional image corresponding to the second position, or the fifth image may be selected upward from the fifth image as the cross-sectional image corresponding to the second position. The specific image selected may be determined based on the layer structure of the circuit board. For example, the thicker the signal layer, the farther the cross-sectional image from the fifth image is selected.

[0164] Sixth embodiment:

[0165] According to the probability estimation result of the second AI model, the fourth image and the fifth image are located. In this embodiment, the third probability threshold is set so that the fourth image and the fifth image represent Figure 3 The boundary between the lower surface of the signal layer 400 and the non-signal layer is shown in FIG.

[0166] The signal layer 400 has a certain thickness. From the lower surface of the signal layer upward, the image of the layer gradually becomes clear from blurred. Accordingly, a fourth probability threshold is preset, which is greater than the third probability threshold; the second AI model determines a sixth image that meets the following conditions: the probability value of the sixth image having the layer characteristics is estimated to reach the fourth probability threshold; among multiple consecutive cross-sectional images with the sixth image as the first one, an image is selected as the cross-sectional image corresponding to the second position. Specifically, the sixth image can be directly used as the cross-sectional image corresponding to the second position.

[0167] Seventh embodiment:

[0168] According to the probability estimation result of the second AI model, the fourth image and the fifth image are located. In this embodiment, the third probability threshold is set so that the fourth image and the fifth image represent Figure 3 The boundary between the lower surface of the signal layer 400 and the non-signal layer is shown in FIG.

[0169] The signal layer 400 has a certain thickness. Among the continuous cross-sectional images with the fifth image as the first image and excluding the fourth image, the seventh image and the eighth image that meet the following conditions and are adjacent are determined to represent Figure 3 The boundary between the upper surface of the signal layer 400 and the non-signal layer is as follows:

[0170] When the second AI model estimates for the first time that the probability value of the eighth image having the layer feature is less than the third probability threshold and the probability value of the seventh image having the layer feature reaches the third probability threshold, the center image between the fifth image and the seventh image is determined as the cross-sectional image corresponding to the second position.

[0171] Eighth embodiment:

[0172] According to the probability estimation result of the second AI model, the fourth image and the fifth image are located. In this embodiment, the third probability threshold is set so that the fourth image and the fifth image represent Figure 3 The boundary between the lower surface of the signal layer 400 and the non-signal layer is shown in FIG.

[0173] The signal layer 400 has a certain thickness. The idea of this embodiment is to determine the characterization of the signal layer 400 in a plurality of continuous cross-sectional images with the fifth image as the first image and excluding the fourth image. Figure 3 A cross-sectional image of the boundary between the upper surface of the signal layer 400 and the non-signal layer as indicated in FIG. 4 :

[0174] Calculate the difference in probability values of two adjacent cross-sectional images of the first group being estimated to have the layer feature. If the difference is less than a preset difference threshold, calculate the difference in probability values of two adjacent cross-sectional images of the next group being estimated to have the layer feature. Repeat the above steps until the difference in probability values between the p-th and p+1-th cross-sectional images after the fifth image is found to be greater than or equal to the difference threshold, where p is a positive integer. Use the [p / 2]th cross-sectional image after the fifth image as the center image of the upper and lower surfaces of the signal layer 400, and as the cross-sectional image corresponding to the second position, where [ ] is an operation symbol for rounding, rounding up, or rounding down.

[0175] Ninth embodiment:

[0176] According to the probability estimation result of the second AI model, the fourth image and the fifth image are located. In this embodiment, the third probability threshold is set so that the fourth image and the fifth image represent Figure 3 The boundary between the lower surface of the signal layer 400 and the non-signal layer is shown in FIG.

[0177] The signal layer 400 has a certain thickness. The image of the layer goes from blurry to clear and then back to blurry. The idea of this embodiment is to locate this clear high point:

[0178] Among a plurality of consecutive cross-sectional images, with the fifth image as the first one and excluding the fourth image, a ninth image is determined that satisfies the following conditions for the first time: the probability value of the ninth image estimated to have the layer feature is greater than both the corresponding probability values of the previous image and the next image; and the ninth image is used as the cross-sectional image corresponding to the second position.

[0179] Other embodiments:

[0180] Any one of the first to fourth embodiments may be combined with any one of the fifth to ninth embodiments.

[0181] Compared with the first embodiment in which the second image is used as the cross-sectional image corresponding to the first position and the second embodiment in which the first image is used as the cross-sectional image corresponding to the first position, the third embodiment in which the centered third image is determined as the cross-sectional image corresponding to the first position and the fourth embodiment in which the feature of the tapered circular ring on the cross-section is used to more accurately determine the center position of the inclined surface at the end of Stub 300 as the first position; on the other hand, compared with the fifth embodiment in which the fifth image is used as the cross-sectional image corresponding to the second position, the seventh, eighth and ninth embodiments can more accurately determine the center position of the signal layer 400 as the second position.

[0182] The above embodiment of more accurately determining the center position of the inclined surface at the end of stub 300 as the first position and more accurately determining the center position of signal layer 400 as the second position can better meet the inspection requirements of high-precision circuit boards: the overall thickness of precision circuit boards is relatively thin. With the development of technology, the spacing between layers is at the micron level. Precision technology can achieve a layer spacing of approximately 100 microns. However, the thickness of the circuit board layers themselves is generally 20 to 35 microns, and the thickness of the inclined surface at the end of the stub caused by the drilling process is generally 10 to 20 microns. The preferred embodiment of the present invention takes the thickness of the layers themselves and the inclined height of the stub end into considerations affecting detection accuracy, optimizes the algorithm in detail, and performs stub detection based on identifying the center position of the inclined surface at the end of the stub and the center position of the signal layer. This is particularly important for improving the stub detection accuracy of precision circuit boards.

[0183] Without inventive effort, the technical solutions of the new embodiments obtained by adjusting the first to ninth embodiments also fall within the scope of protection claimed by the present invention. For example, the solution for determining the fourth image and the fifth image is adjusted to:

[0184] For a set of cross-sectional images whose cross-sectional sorting direction is from the back to the front of the circuit board, sequential identification is performed. If the probability value of the fourth image having the layer feature is less than the preset third probability threshold, and the probability value of the fifth image adjacent to it is estimated to have the layer feature reaches the preset third probability threshold, then the fourth image and the fifth image represent Figure 3 The boundary between the upper surface of the signal layer 400 and the non-signal layer as indicated in FIG;

[0185] For a set of cross-sectional images whose sorting direction is from the front to the back of the circuit board, sequential identification is performed. If the probability value of the fourth image having the layer feature estimated for the second time is less than the preset third probability threshold, and the probability value of the fifth image adjacent to it being estimated to have the layer feature reaches the third probability threshold, then the fourth image and the fifth image represent Figure 3 The boundary between the upper surface of the signal layer 400 and the non-signal layer is shown in FIG.

[0186] In one embodiment of the present invention, a circuit board back-drilled hole STUB detection system is provided, comprising a computerized tomography (CT) device and a processor, wherein the CT device is configured to scan a circuit board to be inspected and generate a set of cross-sectional images; the CT device comprises one or more light pipes and a flat-panel detector corresponding to each of the light pipes.

[0187] The processor is configured with an AI model for determining the cross-sectional images corresponding to the first position and the second position, respectively. The specific training method of the AI model is not described in detail.

[0188] The processor determines the arrangement sequence numbers of the cross-sectional image corresponding to the first position and the cross-sectional image corresponding to the second position in the cross-sectional image set, and determines the difference between the sequence numbers;

[0189] Determining a scanning step distance according to the working parameters of the electronic computed tomography scan, wherein the scanning step distance is the distance between two consecutive scans of the cross-section of the same object;

[0190] The product of the scanning step distance and the sequence number difference is calculated to determine the distance between the first position and the second position, and the product is compared with a set qualified range to determine whether the back-drilled hole Stub is qualified.

[0191] For circuit boards that pass the inspection, the subsequent resin filling process can be carried out. For circuit boards that fail the inspection, they can be repaired and then re-inspected or directly discarded to reduce the defective rate of finished products.

[0192] The circuit board back-drilled hole stub detection system provided in this embodiment and the circuit board back-drilled hole stub detection method provided in the above embodiment belong to the same inventive concept. The entire content of the circuit board back-drilled hole stub detection method embodiment is incorporated into the present circuit board back-drilled hole stub detection system embodiment by reference and will not be repeated here.

[0193] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0194] The above is only a specific implementation method of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A circuit board back-drilling stub detection method, characterized in that: The remaining stub in the back-drilled hole is detected, and the detection method includes the following steps: Acquire a cross-sectional image set generated by scanning the circuit board to be inspected using electronic computed tomography technology, wherein the cross-sectional image set includes a plurality of cross-sectional images arranged in order of heights of the cross sections; Using a pre-built first AI model, the cross-sectional images in the image set are sequentially analyzed to determine the location of the end of the remaining stub, which is recorded as the first location. The first AI model estimates the probability of each input cross-sectional image having the feature of the end of the remaining stub. The first AI model determines the cross-sectional image corresponding to the first location in the following manner: if the estimated probability value of the first image having the feature of the end of the remaining stub is less than a preset first probability threshold, and the estimated probability value of the second image adjacent to the first image having the feature of the end of the remaining stub reaches the first probability threshold, then determining the cross-sectional image corresponding to the first location based on the second image using either of the following two methods: The first method is to preset a second probability threshold that is greater than or equal to the first probability threshold; the first AI model determines a third image that satisfies the following conditions: the probability value of the third image having the end feature of the remaining stub is estimated to reach the second probability threshold; and from a plurality of consecutive cross-sectional images, with the third image being the first image, one image is selected as the cross-sectional image corresponding to the first position; The second method is to calculate the difference between two adjacent cross-sectional images in a first group of multiple continuous cross-sectional images, with the second image as the first image and excluding the first image. If the difference is greater than a preset difference threshold, calculate the difference between two adjacent cross-sectional images in the next group; repeat the above steps until the difference between the k-th cross-sectional image after the second image and the k+1-th cross-sectional image is less than or equal to the preset difference threshold, where k is a positive integer; and use the [k / 2]-th cross-sectional image after the second image as the cross-sectional image corresponding to the first position, where [ ] is a rounding, rounding up, or rounding down operator. Using a pre-built second AI model, the position of the layer electrically connected to the remaining Stub and closest to the end of the remaining Stub is determined, which is recorded as the second position; the second AI model estimates the probability of each input cross-sectional image having the layer feature closest to and connected to the Stub end, and the layer feature includes the line morphology feature connected to the Stub; the second AI model determines the cross-sectional image corresponding to the second position in the following way: for a set of cross-sectional images whose cross-sectional sorting direction is from the back to the front of the circuit board, sequential recognition is performed, and if the probability value of the fourth image having the layer feature is estimated for the first time is less than a preset third probability threshold, and the probability value of the fifth image adjacent to it being estimated to have the layer feature reaches the preset third probability threshold, then determining the cross-sectional image corresponding to the second position based on the fourth image or the fifth image; performing sequential identification on a set of cross-sectional images whose cross-sectional sorting direction is from the front side to the back side of the circuit board, if the probability value of the fourth image being estimated to have the layer feature for the last time is less than a preset third probability threshold, and the probability value of the fifth image adjacent to it being estimated to have the layer feature reaches the third probability threshold, then determining the cross-sectional image corresponding to the second position based on the fourth image or the fifth image; The distance between the first position and the second position is calculated to determine whether the back-drilled hole Stub is qualified.

2. The circuit board back-drilling stub detection method according to claim 1, characterized in that: Also includes: inputting the cross-sectional images of the cross-sectional image set into the first AI model in sequence; The first AI model determines a cross-sectional image corresponding to the first position; The first AI model is trained in the following way: Obtaining a sample image: Scanning the sample circuit board using electronic computed tomography technology and generating a cross-sectional image set, wherein the cross-sectional image set includes a plurality of cross-sectional images arranged in order of heights of the cross sections; Manual labeling: The cross-sectional images that reflect the end features of the remaining stub are labeled with the first type of labels; Repeatedly acquire sample images and manually label them to form a learning sample set; Divide the learning sample set into a training set and a validation set; Input the learning samples of the training set into the basic model for iterative training; The iteratively trained model is verified using the learning samples of the validation set. If the verification fails, the iterative training is continued; if the verification passes, the first AI model is obtained.

3. The circuit board back-drilling stub detection method according to claim 1, characterized in that: The setting value of the first probability threshold is determined by: Inputting the test image set into the trained first AI model to obtain a probability value for each cross-sectional image in the test image set being estimated to have an end feature of a remaining stub; Manually determining the target cross-sectional image corresponding to the position of the end of the remaining stub in the test image set; Determining the probability values of the adjacent images on both sides of the target cross-sectional image being estimated; A reference value of the first probability threshold is determined within an interval range of the probability values estimated for the adjacent images on both sides.

4. The circuit board back-drilling stub detection method according to claim 3, characterized in that: Using a plurality of test image sets respectively, obtaining a plurality of reference values of corresponding first probability thresholds; After performing denoising processing on the reference values of the multiple first probability thresholds, an average value is obtained to serve as the first probability threshold.

5. The circuit board back-drilling stub detection method according to claim 1, characterized in that: Also includes: inputting the cross-sectional images of the cross-sectional image set into the second AI model in sequence; The second AI model determines a cross-sectional image corresponding to the second position; The second AI model is trained in the following way: Obtaining a sample image: Scanning the sample circuit board using electronic computed tomography technology and generating a cross-sectional image set, wherein the cross-sectional image set includes a plurality of cross-sectional images arranged in order of heights of the cross sections; Manual labeling: The second type of label is annotated on the cross-sectional images that reflect the characteristics of the layer closest to the stub end and connected to it; Repeatedly acquire sample images and manually label them to form a learning sample set; Divide the learning sample set into a training set and a validation set; Input the learning samples of the training set into the basic model for iterative training; The iteratively trained model is verified using the learning samples of the validation set. If the verification fails, the iterative training is continued; if the verification passes, the second AI model is obtained.

6. The circuit board back-drilling stub detection method according to claim 1, characterized in that: The second AI model determines the cross-sectional image corresponding to the second position in any of the following ways: The first method is to select an image from a plurality of continuous cross-sectional images, with the fifth image being the first image and excluding the fourth image, as the cross-sectional image corresponding to the second position; Method 2 is to preset a fourth probability threshold that is greater than the third probability threshold; the second AI model determines a sixth image that satisfies the following conditions: the probability value of the sixth image having the layer feature is estimated to reach the fourth probability threshold; and select an image from a plurality of consecutive cross-sectional images, the first of which is the sixth image, as the cross-sectional image corresponding to the second position; Method three is to determine, among a plurality of consecutive cross-sectional images, starting with the fifth image and excluding the fourth image, a seventh image and an eighth image that meet the following conditions and are adjacent: if the second AI model first estimates that the probability value of the eighth image having the layer feature is less than the third probability threshold and the probability value of the seventh image having the layer feature reaches the third probability threshold, then determine the image centered between the fifth and seventh images as the cross-sectional image corresponding to the second position; The fourth method is to calculate, among a plurality of consecutive cross-sectional images, starting with the fifth image and excluding the fourth image, a difference in probability values of two adjacent cross-sectional images in a first group being estimated to have the layer feature; if the difference is less than a preset difference threshold, then calculate a difference in probability values of two adjacent cross-sectional images in a next group being estimated to have the layer feature; Repeat the above steps until the difference in probability values between the p-th and p+1-th cross-sectional images after the fifth image is found to be greater than or equal to the difference threshold, where p is a positive integer; and use the [p / 2]-th cross-sectional image after the fifth image as the cross-sectional image corresponding to the second position, where [ ] represents a rounding, rounding up, or rounding down operator. Method five is to determine, among a plurality of continuous cross-sectional images with the fifth image as the first one and excluding the fourth image, a ninth image that first satisfies the following conditions: the probability value of the ninth image being estimated to have the layer feature is greater than both the corresponding probability values of the previous image and the next image; and the ninth image is used as the cross-sectional image corresponding to the second position.

7. The circuit board back-drilling stub detection method according to claim 1, characterized in that: The setting value of the third probability threshold is determined by: Inputting the test image set into the trained second AI model to obtain a probability value for each cross-sectional image in the test image set being estimated to have the layer feature; Manually determining a target cross-sectional image corresponding to a location of a layer electrically connected to the remaining stub in the test image set; Determining the probability values of the adjacent images on both sides of the target cross-sectional image being estimated; A reference value of the third probability threshold is determined within an interval range of the probability values estimated for the adjacent images on both sides.

8. The circuit board back-drilling stub detection method according to claim 7, characterized in that: Obtaining corresponding reference values of multiple third probability thresholds using multiple test image sets respectively; After performing denoising processing on the reference values of the multiple third probability thresholds, an average value is obtained to serve as the second probability threshold.

9. The circuit board back-drilling stub detection method according to any one of claims 1 to 8, characterized in that: The distance between the first position and the second position is calculated as follows: determining the arrangement sequence numbers of the cross-sectional image corresponding to the first position and the cross-sectional image corresponding to the second position in the cross-sectional image set respectively; Calculate the difference between the sequence numbers corresponding to the first position and the second position; Determining a scanning step distance according to the working parameters of the electronic computed tomography scan, wherein the scanning step distance is the distance between two consecutive scans of the cross-section of the same object; The distance between the first position and the second position is determined by calculating the product of the scanning step distance and the sequence number difference.

10. The circuit board back-drilling stub detection method according to claim 9, characterized in that: Determine whether the back-drilled stub is qualified by the following methods: Set the Stub qualified length range [L min ,L max ], where L min L is the lower limit of the stub qualified length range. max The upper limit of the stub qualified length range. Define the distance between the first position and the second position as l 0, if L min ≤ l 0≤L max , the back-drilled stub is determined to be qualified, otherwise the back-drilled stub is determined to be unqualified.

11. A circuit board back drilling STUB detection system, characterized in that: The invention comprises an electronic computer tomography device and a processor, wherein the electronic computer tomography device is configured to scan a circuit board to be inspected and generate a set of cross-sectional images; The processor is configured to perform the steps of the detection method according to any one of claims 1 to 10.

12. The circuit board back-drilling STUB detection system according to claim 11, characterized in that: The electronic computer tomography device includes one or more light pipes and flat panel detectors corresponding to the light pipes.

Images