Method and device for carrying out an x-ray inspection on soldering points on SMD printed circuit boards

Abstract

The invention relates to a method, in particular an at least partly computer-implemented method for checking the quality of soldering points on an SMD printed circuit board (3) using a 2D X-ray inspection device, having the following steps: - capturing (S1) a 2D X-ray image using an X-ray image camera (2); - evaluating (S2) the captured 2D X-ray image using an image preprocessing model in order to obtain an evaluation image; and - checking (S3) the quality of the soldering points on the basis of the evaluation image using a checking method.

Description

[0001] R. 414735

[0002] - 1 -

[0003] Description

[0004] title

[0005] Method and apparatus for X-ray inspection of solder joints on SMD printed circuit boards

[0006] Technical field

[0007] The invention relates to the X-ray inspection of SMD printed circuit boards, and in particular to methods for checking the solder joint quality after their manufacture. The invention further relates to the use of 2D X-ray inspection devices for determining the solder joint quality.

[0008] Technical background

[0009] In the production of SMD printed circuit boards, the solder pads are typically coated with solder paste, and then the components to be soldered, i.e., the SMD components, are placed onto the solder pads coated with solder paste and heated appropriately. The solder paste melts and forms a mechanically strong electrical connection between a contact point of the SMD component and the solder pad.

[0010] However, external influences, such as mechanical vibrations during the joining process, or variations in process parameters can easily lead to faulty solder joints, resulting in an interrupted current path or a current path with reduced conductivity. Therefore, testing procedures are performed after the printed circuit board is manufactured to verify the quality of the solder joints. Due to the complexity of modern electronic circuits, these defects are not always detectable by suitable electrical or electronic testing procedures, especially functional testing procedures. R. 414735

[0011] - 2 -

[0012] Optical inspection methods are no longer suitable for multilayer printed circuit boards (PCBs) or PCBs with covered solder joints, such as those used in IC packages. Therefore, X-ray inspection methods are typically employed, as they can detect and display all solder joints within the beam path. However, particularly with multilayer surface-mount devices (SMDs), all components within the beam path appear as shadows on the resulting X-ray image, which forms the basis for these inspection methods. Individual components can absorb X-rays to varying degrees, leading to shadowing on the X-ray image. This can result in inaccurate assessments of the solder quality of individual, shadowed solder joints based solely on the X-ray image.Therefore, restricted zones are currently designated for the placement of SMD components in order to avoid shadowing and to facilitate inspection based on a 2D x-ray image.

[0013] Furthermore, 3D X-ray inspection methods can be used, in which a three-dimensional X-ray image is created. This allows the quality of the solder joints to be assessed by evaluating specific plane images of the S MD printed circuit board, as the plane images are not affected by obscuring or absorbing structures. This eliminates the need for designated assembly exclusion zones and thus enables a more compact design of the S MD printed circuit board.

[0014] However, the time required to create a 3D X-ray image is considerably longer than that of a 2D X-ray image. Furthermore, the exposure to X-rays during image acquisition is significantly higher than with a 2D X-ray image, so the higher radiation dose can lead to damage in semiconductor components on the S MD circuit board. Therefore, the use of 3D X-ray inspection methods is not suitable, especially for high production volumes and semiconductor components with radiation-sensitive or small feature sizes.

[0015] It is therefore an object of the present invention to provide a method for improved R. 414735

[0016] - 3 -

[0017] To provide inspection of solder joints on S MD printed circuit boards, enabling a fast turnaround time for testing and keeping the radiation exposure on semiconductor components of the S MD printed circuit board as low as possible.

[0018] Disclosure of the invention

[0019] This problem is solved by the method for inspecting solder joints on an S MD printed circuit board according to claim 1 and by the device according to the dependent claim.

[0020] Further details are specified in the dependent claims.

[0021] According to a first aspect, a method, in particular a method at least partially computer-implemented method for checking the quality of solder joints on an S MD printed circuit board using a 2D X-ray inspection device, is provided, comprising the following steps:

[0022] Capturing a 2D X-ray image with an X-ray camera;

[0023] Evaluating the captured 2D X-ray image using an image preprocessing model to obtain an evaluation image;

[0024] Checking the quality of the solder joints using a verification procedure based on the evaluation image.

[0025] Furthermore, the image preprocessing model can include a data-based model, in particular an autoencoder or a convolutional neural network.

[0026] Due to the aforementioned problems with the inspection method based on 3D X-ray images, there is a need to provide a reliable inspection method for solder joints on SMD printed circuit boards based on 2D X-ray images. The problem that many absorbing or shadowing structures are visible in 2D X-ray images, and these obscure the image of the solder joints to be inspected, making it difficult to correlate structures detected in the 2D X-ray image with specific solder joints, is addressed by the invention. This involves performing image preprocessing based on a regularly acquired 2D X-ray image using an image preprocessing model that enables the detection of shadowing or shadowing. R. 414735

[0027] - 4 -

[0028] to remove other interfering images in the captured 2D x-ray image in order to carry out a verification or inspection procedure based on such an evaluation image, with which the quality of the solder joints shown in the evaluation image can be checked.

[0029] The inspection procedure corresponds to a conventional inspection procedure designed to detect a solder joint defect in images not affected by shadows or other interfering influences.

[0030] A major problem in inspecting solder joints on SMD printed circuit boards is that interfering structures, which are visible in the 2D X-ray image but cannot be correlated with the solder joints being inspected, appear as gray shading in the image. Approaches to reducing the shadows cast by these absorbing interfering structures through simple pixel-wise subtraction have proven highly unreliable in practice.

[0031] It is therefore proposed to provide a data-driven image preprocessing model trained to generate a refined 2D X-ray image from a captured 2D X-ray image containing interfering structures due to shadowing or absorption. This refined image should only include the structures of interest at the solder joints and should no longer show interfering structures. Such an image preprocessing model could be based on convoluted neural networks, autoencoders, or similar technologies.

[0032] The data-based model can be trained to provide a correction image, wherein the data-based model is trained to provide the correction image as a modeled image of the unpopulated circuit board when given the acquired 2D X-ray image, wherein the correction image is used within the image preprocessing model for pixel-wise subtraction from the acquired 2D X-ray image to obtain the evaluation image.

[0033] Alternatively, the data-based model can be trained to provide the evaluation image, whereby the data-based model is trained to provide the evaluation image when given the acquired 2D X-ray image. R. 414735

[0034] - 5 -

[0035] The evaluation image is performed using conventional radiographic testing, in which a projection image of the circuit board under test is generated. The degree of blackening reveals the different material properties and allows conclusions to be drawn as to whether a solder joint is correct or not.

[0036] Training such an image preprocessing model for a specific SMD printed circuit board (PCB) can be achieved by providing a large number of training datasets of initial 2D X-ray images showing fully populated SMD PCBs containing interfering structures. Additionally, one or more secondary 2D X-ray images of the unpopulated PCB are provided, showing only traces, pads, and similar features as shadowing structures. Alternatively, PCBs that are already populated on one side can be used to create a model for correction on the second reflow side.

[0037] The image preprocessing model can be trained using a generative model, such as GAN, cGAN, or pix2pix, with the first and second 2D X-ray images. The trained image processing model can then generate a second 2D X-ray image from the first, essentially resembling an X-ray image of an unpopulated or single-sided printed circuit board. This second 2D X-ray image can then be used for pixel-by-pixel image subtraction from the first 2D X-ray image of the fully populated S₂MD printed circuit board. The result is an evaluation image in which the shadowing structures resulting from the structure of the unpopulated or single-sided S₂MD printed circuit board are no longer present.

[0038] Another way to train the image preprocessing model is to first extract the image plane containing the existing structures—namely, the solder joints to be examined on the SMD circuit board—as the first 2D X-ray image, based on a 3D X-ray image acquired using a 3D X-ray inspection device. Furthermore, all acquired image planes of the 3D X-ray image can be combined to obtain a second 2D X-ray image, corresponding to the 2D X-ray image of a fully populated SMD circuit board with shadows. Alternatively, the second 2D X-ray image can be... 414735

[0039] - 6 -

[0040] A 2D X-ray image, acquired using a 2D X-ray inspection device, can be used. Using these image pairs, a corresponding generative model can be trained as before. The trained model can then be used on 2D X-ray images acquired by 2D X-ray inspection devices to remove interfering shadowing structures from the resulting 2D X-ray image.

[0041] Brief description of the drawings

[0042] The embodiments are explained in more detail below with reference to the accompanying drawings. These show:

[0043] Figure 1 shows a schematic representation of an X-ray inspection device for testing solder joints of SMD printed circuit boards;

[0044] Figure 2 is a flowchart illustrating a procedure for performing a verification of solder joints on an S MD printed circuit board after its completion; and

[0045] Figure 3 shows a comparison of a 2D x-ray image with shadows and an unpopulated circuit board.

[0046] Description of embodiments

[0047] Figure 1 schematically shows an X-ray inspection device 1 for an S-MD printed circuit board 3. The S-MD printed circuit board 3 has a circuit board with conductor tracks and solder pads, which is populated with electronic components. The X-ray inspection device 1 comprises an X-ray camera 2, which records X-rays passing through the S-MD printed circuit board 3, and a data processing device 4, which receives a (first) 2D X-ray image from the X-ray camera 2, preprocesses this image using an image preprocessing model and / or algorithm, and then evaluates the resulting image according to a known inspection procedure in order to assess the solder joints of the S-MD printed circuit board 3 with regard to their solder joint quality.

[0048] - 7 -

[0049] This can be verified, as is known, by analyzing the evaluation image.

[0050] The image preprocessing model is implemented in data processing unit 4 and is designed as a convolutional neural network or an autoencoder to determine the evaluation image based on the acquired 2D X-ray image. This image is cleaned of shadows and interfering structures, so that the evaluation of solder joint quality can be carried out more effectively using conventional testing methods.

[0051] Figure 2 shows a flowchart illustrating a procedure used to evaluate the quality of solder joints on S MD printed circuit boards.

[0052] In step S1, a 2D X-ray image of the selected S MD circuit board 3 is first created using the X-ray imaging camera 2. Such an image is shown schematically, for example, in Fig. 3 (left image). The representation in Figure 3 shows a chip 5 with solder points 6 and the shadows 7 of interfering structures that could lead to incorrect evaluations of the solder joints located in the same image area.

[0053] Using the image preprocessing model, an evaluation image (schematically shown in Figure 3, right image) can now be created from the acquired 2D X-ray image. In step S3, this evaluation image is analyzed using the solder joint inspection procedure to identify defective or defect-free solder joints, or to assess the quality of the solder joints.

[0054] The image preprocessing model can be trained using image pairs as training datasets, each showing a first 2D X-ray image of specific SMD printed circuit boards, including all shadowing effects, and a corresponding second 2D X-ray image of the unpopulated circuit board. Both X-ray images are acquired with the same 2D X-ray inspection device (in the same way, i.e., same perspective, orientation, focus, etc.) and represent data pairs for training a generative model, such as GAN, cGAN, pix2pix, and the like.

[0055] A captured 2D X-ray image can be processed using the appropriate trained R. 414735

[0056] - 8 -

[0057] The model can be converted into an evaluation image free of shadowing effects or into a correction image corresponding to an unpopulated S MD printed circuit board. The model can provide the correction image, which is subtracted pixel by pixel from the acquired 2D X-ray image to provide the evaluation image of the fully populated S MD printed circuit board without shadowing effects.

[0058] The training of such an image preprocessing model for a specific SMD circuit board can be performed by providing a large number of training datasets of initial 2D X-ray images showing fully populated SMD circuit boards containing interfering structures / shadowing features. Furthermore, one or more secondary 2D X-ray images of the unpopulated or only single-sided (of a double-sided) circuit board are provided, showing only traces, contact pads, and the like as shadowing features.

[0059] The image preprocessing model can be trained using a generative model, such as GAN, cGAN, or pix2pix, with the first and second 2D X-ray images. The trained image processing model can then generate a second 2D X-ray image from the first, essentially resembling an X-ray image of an unpopulated circuit board. This second 2D X-ray image then serves as a correction image and can be used for pixel-by-pixel image subtraction from the first 2D X-ray image of the fully populated S MD circuit board.

[0060] The result is an evaluation image in which the shadowing structures resulting from the structuring of the unpopulated or only one-sidedly populated (in the case of a double-sided printed circuit board) S MD printed circuit board are no longer present.

[0061] Another way to train the image preprocessing model is to first extract the image plane with the existing structures, namely the solder joints to be examined on the S MD circuit board, as the first 2D X-ray image, based on a 3D X-ray image that has been acquired using a 3D X-ray inspection device.

[0062] Furthermore, all captured image planes of the 3D X-ray image can be viewed. R. 414735

[0063] - 9 -

[0064] The two images are combined to obtain a second 2D X-ray image that corresponds to the 2D X-ray image of a fully populated SDM printed circuit board with shadowing. Alternatively, a 2D X-ray image generated by 2D X-ray inspection can be used as the second 2D X-ray image. Using these image pairs, a corresponding generative model can be trained as before. The trained model can then be used on 2D X-ray images acquired by 2D X-ray inspection devices to remove the interfering shadowing defects from the resulting 2D X-ray image.

Claims

R. 414735 - 10 - Claims 1. Method, in particular a method, at least partially computer-implemented, for checking the quality of solder joints on an SMD printed circuit board (3) using a 2D X-ray inspection device, comprising the following steps: Acquisition (S1) of a 2D X-ray image with an X-ray camera (2); Evaluation (S2) of the acquired 2D X-ray image using an image preprocessing model to obtain an evaluation image; Check (S3) the quality of the solder joints using a verification procedure based on the evaluation image.

2. The method of claim 1, wherein the image preprocessing model comprises a data-based model, in particular an autoencoder or a convoluted neural network.

3. Method according to claim 2, wherein the data-based model is configured to provide a correction image, wherein the data-based model is trained to provide the correction image as a modeled image of the unpopulated circuit board (3) when given the acquired 2D X-ray image, wherein the correction image is used within the image preprocessing model for pixel-wise subtraction from the acquired 2D X-ray image to obtain the evaluation image.

4. Method according to claim 2, wherein the data-based model is configured to provide the evaluation image, wherein the data-based model is trained to provide the evaluation image when the captured 2D X-ray image is specified.

5. Method for training a data-based model of an image preprocessing model to determine an evaluation image from a captured 2D X-ray image, comprising the following steps: R. 414735 - 11 - Providing training datasets that show, for a variety of different SM D printed circuit boards, a captured 2D X-ray image of a populated SM MD printed circuit board and a 2D X-ray image of the corresponding unpopulated printed circuit board, Training a data-based model with the training datasets to provide a correction image as a modeled image of the unpopulated circuit board when given a captured 2D x-ray image, where the correction image is used for pixel-wise subtraction from the captured 2D x-ray image.

6. Method for training a data-based model of an image preprocessing model to determine an evaluation image from a captured 2D X-ray image, comprising the following steps: Providing training datasets that, for a variety of different SMD circuit boards, assign a superimposed image from the images of several layers of a captured 3D X-ray image of a populated SMD circuit board to an image of one of the layers of the 3D X-ray image in which solder joints to be inspected are located, Training a data-based model with the training datasets to provide an evaluation image as a reconstructed image of the relevant plane in which the solder joints to be checked are located, given a captured 2D x-ray image.

7. Method according to claim 6, wherein the data-based model of the image preprocessing model is trained using a generative model, such as GAN, cGAN, pix2pix.

8. Device for carrying out one of the above methods according to one of claims 1 to 7.

9. Computer program product comprising instructions which, when executed by at least one data processing device, cause it to execute the steps of the method according to any one of claims 1 to 7.

10. Machine-readable storage medium comprising instructions which, when executed by at least one data processing device, cause it to execute the steps of the method according to any one of claims 1 to 7.

Images