PCBA element height measurement method based on deep learning stereo matching

Through deep learning stereo matching method and binocular camera mobile platform, the inefficiency and accuracy problems in PCBA component height measurement are solved, high-precision component height measurement is achieved, and the reliability of the test fixture design is improved.

CN119027479AActive Publication Date: 2024-11-26SOUTH CHINA UNIV OF TECH
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411086058.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-11-26
Estimated Expiration
2044-08-08

AI Technical Summary

Technical Problem

The prior art is inefficient and error-prone in PCBA component height measurement, lacks automated measurement methods, affecting the accuracy of the test fixture design.

Method used

Using a stereo matching method based on deep learning, the ERH-Stereo stereo matching network structure of ECA attention mechanism + RAFT-Stereo + UHRNet high-resolution network is designed, combined with binocular cameras and orthogonal experiments, and the two degrees of freedom are designed. The binocular camera moves the platform to achieve high-precision component height measurement.

Benefits of technology

Improves the accuracy and efficiency of PCBA component height measurement, ensuring the accuracy and reliability of the test fixture design.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027479A_ABST
    Figure CN119027479A_ABST
Patent Text Reader

Abstract

The invention discloses a PCBA element height measurement method based on deep learning stereo matching. The method comprises the following steps: acquiring a PCBA object to obtain left and right RGB images; the method comprises the following steps: designing an ERH-Stereo stereo matching network structure of an ECA attention mechanism, an RAFT-Stereo and a UHRNet high-resolution network; parameters such as a baseline distance B (mm), a lens focal length f and an object distance D of the binocular camera are selected through orthogonal experiment design, so that a minimum average error suitable for measurement of a field of view and element height is obtained.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of dimension measurement of mounted circuit board components, and in particular to a PCBA component height measurement method based on deep learning stereo matching. Background Art

[0002] Functional Test (FCT) is the process of testing and verifying the functions and indicators of the units under test (Under Unit Test, UUT) such as the printed circuit board assembly (PCBA) and electronic components. As product manufacturing tends to be customized in small batches and UUT production tends to be personalized, this requires modular and flexible design of FCT equipment. The test fixture is an important part of the FCT equipment. It is very important to quickly and reliably install and locate the PCBA to be tested and reduce the deformation of the installation force, and the selection of external test points is very critical. Measuring the height of PCBA components and determining the distribution of circuit board components are the premise and basis of test fixture design. At present, the general method is to obtain the corresponding component coordinates through the layout of the circuit board circuit diagram and manually measure the height of the circuit board components. There are problems such as low efficiency and easy errors. The automatic measurement method of PCBA component height is the key to the breakthrough of test fixture design. Summary of the invention

[0003] In order to solve the above technical problems, the purpose of the present invention is to provide a PCBA component height measurement method based on deep learning stereo matching.

[0004] The purpose of the present invention is achieved through the following technical solutions:

[0005] A PCBA component height measurement method based on deep learning stereo matching, comprising:

[0006] Step A collects the PCBA object to obtain left and right RGB images;

[0007] Step B designs the ERH-Stereo stereo matching network structure of ECA attention mechanism + RAFT-Stereo + UHRNet high-resolution network;

[0008] Step C selects the binocular camera baseline distance B (mm), lens focal length f, and object distance D parameters through orthogonal experimental design to obtain the minimum average error suitable for field of view size and component height measurement.

[0009] Furthermore, based on the binocular ranging principle, the PCBA object is captured by a binocular camera to obtain left and right RGB images of the PCBA.

[0010] Furthermore, the left and right RGB images are based on the binocular disparity ranging principle, and the PCBA disparity map is obtained through the ERH-Stereo stereo matching network structure composed of ECA attention mechanism + RAFT-Stereo + UHRNet high-resolution network. Then, the PCBA depth information is calculated in combination with the binocular camera baseline distance and focal length information to obtain the PCBA component height.

[0011] Furthermore, the core structure of the ERH-Stereo stereo matching network includes a feature extraction module, a correlation cost product module, a disparity iteration module and a disparity refinement module, wherein the correlation cost product module and the disparity iteration module inherit RAFT-Stereo, the feature extraction module adopts the feature extraction of the fused ECA attention mechanism, and the disparity refinement module adopts the UHRNet disparity refinement method.

[0012] Furthermore, the 1 / 4 resolution disparity map Disparity(1 / 4)De is upsampled to obtain the original disparity, and I r , D e The left image I is reconstructed by the distortion method l ', reprojection error err, and then by I l , err is obtained by convolution operation I fuse :

[0013]

[0014] Finally, the UHRNet network outputs a refined disparity image, namely: D' e =UHRNet(I fuse ,Conv 3×3 (disp)).

[0015] Furthermore, the step C includes: the monocular camera field of view FOV S It is related to the object distance D, the camera sensor width H (mm), and the sensor height V (mm), while the binocular camera field of view FOV D There is overlap in the width direction, which is also related to B, that is:

[0016]

[0017] Set up an orthogonal experiment with 3 factors and 4 levels and select the optimal parameters.

[0018] Furthermore, a two-degree-of-freedom binocular camera moving platform is designed, and a stepper motor is used to drive the lead screw to realize the Z-axis movement of the binocular camera to adjust the object distance. A slide table is used to realize the lateral movement of the binocular camera to adjust the baseline distance. Leveling bolts and limit guide rod mechanisms are arranged to ensure smooth movement of the left and right cameras.

[0019] Compared with the prior art, one or more embodiments of the present invention may have the following advantages:

[0020] This method designs an innovative ERH-Stereo stereo matching network structure of ECA attention mechanism + RAFT-Stereo + UHRNet high-resolution network, improves stereo matching accuracy by extracting fine features of images and refining disparity, selects optimal measurement parameters through orthogonal experiments, designs a two-degree-of-freedom binocular camera mobile platform, uses a stepper motor to drive the screw rod to realize the camera Z-axis movement, and uses a precision slide to realize the lateral movement of the binocular camera to adjust the baseline distance. It can complete high-precision measurement of PCBA component height. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of the PCBA component height measurement method based on deep learning stereo matching;

[0022] Figure 2 This is the structure diagram of the ERH-Stereo stereo matching network;

[0023] Figure 3a and 3b It is a structural diagram of a two-degree-of-freedom binocular camera mobile platform. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below in conjunction with embodiments and drawings.

[0025] like Figure 1 As shown, a PCBA component height measurement method based on deep learning stereo matching includes:

[0026] Step 10: Collect the PCBA object to obtain left and right RGB images;

[0027] Step 20 designs an innovative ERH-Stereo stereo matching network structure of ECA attention mechanism + RAFT-Stereo + UHRNet high-resolution network.

[0028] Step 30 selects the binocular camera baseline distance B (mm), lens focal length f (mm), object distance D (mm) and other parameters through orthogonal experimental design to obtain a minimum average error suitable for field of view size and component height measurement.

[0029] The monocular camera field of view FOVS is related to the object distance D, the camera sensor width H (mm), and the sensor height V (mm), while the binocular camera field of view FOVD overlaps in the width direction and is also related to B, that is:

[0030]

[0031] Set up an orthogonal experiment with 3 factors and 4 levels and select the optimal parameters.

[0032] like Figure 2 As shown, the ERH-Stereo stereo matching network structure diagram includes:

[0033] The core structure of ERH-Stereo stereo matching network includes four functional modules: feature extraction, correlation cost product, disparity iteration, and disparity refinement. The correlation cost product module and disparity iteration module inherit RAFT-Stereo, while feature extraction uses the fusion ECA attention mechanism feature extraction and disparity refinement uses the UHRNet disparity refinement method.

[0034] Sample the original disparity and use I r , D e The left image I is reconstructed by the distortion method l ', reprojection error err, and then by I l , err is obtained by convolution operation I fuse :

[0035]

[0036] Finally, the UHRNet network outputs a refined disparity image, namely: D' e =UHRNet(I fuse ,Conv 3×3 (disp))

[0037] like Figure 3a , 3b As shown in the figure, it is a structural diagram of a two-degree-of-freedom binocular camera mobile platform, including:

[0038] The designed two-degree-of-freedom binocular camera moving platform includes: stepper motor 1, limit block 2, camera assembly support plate 3, guide rod 4, lead screw 5, leveling bolt 6, lead screw fixing seat 7, precision slide rail 8, camera fixing plate 9, left and right cameras 10, etc. A stepper motor is used to drive the lead screw to realize the Z-axis movement of the camera, and a precision slide is used to realize the lateral movement of the binocular camera to adjust the baseline distance.

[0039] Although the embodiments disclosed in the present invention are as above, the above contents are only embodiments adopted for facilitating the understanding of the present invention and are not intended to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present invention, but the patent protection scope of the present invention shall still be subject to the scope defined in the attached claims.

Claims

1. A PCBA component height measurement method based on deep learning stereo matching, characterized in that: The following steps are involved: Step A collects the PCBA object to obtain left and right RGB images; Step B designs the ERH-Stereo stereo matching network structure of ECA attention mechanism + RAFT-Stereo + UHRNet high-resolution network; Step C selects the binocular camera baseline distance B (mm), lens focal length f, and object distance D parameters through orthogonal experimental design to obtain the minimum average error suitable for field of view size and component height measurement.

2. The PCBA component height measurement method based on deep learning stereo matching according to claim 1 is characterized in that: Based on the binocular ranging principle, the PCBA object is captured by a binocular camera to obtain the left and right RGB images of the PCBA.

3. The PCBA component height measurement method based on deep learning stereo matching according to claim 2 is characterized in that: The left and right RGB images are based on the binocular disparity ranging principle. The PCBA disparity map is obtained through the ERH-Stereo stereo matching network structure composed of ECA attention mechanism + RAFT-Stereo + UHRNet high-resolution network. Then, the PCBA depth information is calculated in combination with the binocular camera baseline distance and focal length information to obtain the PCBA component height.

4. The PCBA component height measurement method based on deep learning stereo matching according to claim 1 is characterized in that: The core structure of the ERH-Stereo stereo matching network includes a feature extraction module, a correlation cost product module, a disparity iteration module and a disparity refinement module. The correlation cost product module and the disparity iteration module inherit RAFT-Stereo, the feature extraction module adopts the fusion ECA attention mechanism feature extraction, and the disparity refinement module adopts the UHRNet disparity refinement method.

5. The PCBA component height measurement method based on deep learning stereo matching according to claim 4 is characterized in that: Upsample the 1 / 4 resolution disparity map Disparity(1 / 4)De to get the original disparity, and use I r , D e The left image I is reconstructed by the distortion method l ', reprojection error err, and then by I l , err is obtained by convolution operation I fuse : Finally, the UHRNet network outputs a refined disparity image, namely: e =UHRNet(I fuse ,Conv 3×3 (disp)).

6. The PCBA component height measurement method based on deep learning stereo matching according to claim 1, characterized in that: The step C includes: the monocular camera field of view FOV S It is related to the object distance D, the camera sensor width H (mm), and the sensor height V (mm), while the binocular camera field of view FOV D There is overlap in the width direction, which is also related to B, that is: Set up an orthogonal experiment with 3 factors and 4 levels and select the optimal parameters.

7. The PCBA component height measurement method based on deep learning stereo matching according to claim 1 is characterized in that: Before executing step A, the method includes designing a two-degree-of-freedom binocular camera moving platform, selecting a stepper motor to drive a lead screw to realize Z-axis movement of the binocular camera to adjust the object distance, using a slide to realize lateral movement of the binocular camera to adjust the baseline distance; arranging leveling bolts and a limit guide rod mechanism to ensure smooth movement of the left and right cameras.

Citation Information

Patent Citations

  • Binocular distance measurement method and system based on depth stereo matching algorithm

    CN117635730A

  • Binocular stereo matching method based on convolutional neural network

    CN117808862A

  • Real-time binocular stereo matching method based on adaptive candidate parallax prediction network

    WO2022089077A1