A deep learning-based patch method, intelligent production equipment and storage medium

Abstract

The present invention relates to the field of intelligent manufacturing technology, and in particular to a patch method based on deep learning, intelligent production equipment and storage medium. The patch method based on deep learning is applied to a patch processing system, including: performing image acquisition for the current processing stage to obtain a processing image; analyzing the processing image through a deep learning model to obtain processing parameter information of the next processing stage of the current processing stage and / or processing quality information of the current processing stage. The present invention can improve production efficiency and accuracy, ensure product quality, and reduce errors and losses in production.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to a deep learning-based patch method, intelligent production equipment and storage medium. Background Art

[0002] In the modern electronic manufacturing industry, surface mount technology (SMT) is a key production process. With the development of technology, the distance (Pitch value) between the chips to be bonded is constantly shrinking, which poses a challenge to the traditional SMT production line. The traditional SMT production line may not be able to meet the requirements of the new process, thus the flip-chip COB process of small-pitch PCB boards was born, but the traditional manufacturing process uses manual detection, resulting in low detection efficiency, and the detection quality is also affected by multiple interference factors such as human and environmental factors, resulting in excessively high detection costs and the inability to effectively guarantee detection accuracy. The patch method, intelligent production equipment and storage medium based on deep learning of the present invention are born to solve such problems. Summary of the invention

[0003] The technical problem to be solved by the embodiments of the present invention is to provide a deep learning-based patch method, intelligent production equipment and storage medium to solve the problem in the prior art that traditional SMT production lines do not meet the new process requirements and manual inspection has high costs and low precision.

[0004] The present invention discloses a patch method based on deep learning, which is applied to a patch processing system.

[0005] The patch method based on deep learning includes:

[0006] Perform image acquisition for the current processing stage to obtain the processed image;

[0007] The processed image is analyzed by a deep learning model to obtain processing parameter information of the next processing stage of the current processing stage and / or processing quality information of the current processing stage.

[0008] Optionally, the processing stages of the patch processing system include:

[0009] The board mounting stage, printing stage, SMT stage, reflow soldering stage and lighting test stage.

[0010] Optionally, the step of acquiring processing parameter information of a next processing stage of the current processing stage and / or processing quality information of the current processing stage includes:

[0011] When the current processing stage is the board loading stage, it is detected whether the current PCB board meets the manufacturing standards. If the PCB board meets the manufacturing standards, the corresponding printing parameters are matched in combination with the current PCB board parameters for printing the PCB board.

[0012] Optionally, the step of acquiring processing parameter information of a next processing stage of the current processing stage and / or processing quality information of the current processing stage includes:

[0013] When the current processing stage is the printing stage, the corresponding patch parameters are matched in combination with the current printing parameters to perform the patch operation on the PCB board.

[0014] Optionally, the step of acquiring processing parameter information of a next processing stage of the current processing stage and / or processing quality information of the current processing stage includes:

[0015] When the current processing stage is the patch stage, the corresponding reflow parameters are matched in combination with the current patch parameters to perform a reflow soldering operation on the PCB board.

[0016] Optionally, the step of acquiring processing parameter information of a next processing stage of the current processing stage and / or processing quality information of the current processing stage includes:

[0017] When the current processing stage is the lighting test stage, the processing parameters of the test failure point are obtained, and the processing parameters include at least one of PCB board parameters, printing parameters, patch parameters and reflow parameters. Re-learning is performed based on the processing parameters to obtain knowledge accumulation of parameter matching values ​​to make predictive corrections to the operation of the next PCB board.

[0018] Optionally, after the step of acquiring the processed image, the following steps are included:

[0019] Perform image preprocessing on the processed image to obtain image features.

[0020] Optionally, the patch method based on deep learning further includes:

[0021] The quality parameters of the final product are obtained, and the deep learning model is modified based on the quality parameters.

[0022] The present invention also discloses an intelligent production device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method described above.

[0023] The present invention also discloses a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the method described above.

[0024] Compared with the prior art, the deep learning-based patch method provided by the embodiment of the present invention has the beneficial effect that: by analyzing the processing image through the deep learning model, the automatic monitoring and analysis of the current processing stage can be realized. This helps to reduce manual intervention, improve production efficiency and accuracy, and by obtaining the processing parameter information and / or quality information of the current processing stage, real-time feedback can be achieved, and the processing parameters can be automatically adjusted to ensure product quality and production efficiency, and reduce errors and losses in production. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The scheme of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments, in which:

[0026] Figure 1 It is a flowchart of an embodiment of the deep learning patch method provided by the present invention;

[0027] Figure 2 It is a structural schematic diagram of an embodiment of a patch processing system provided by the present invention.

[0028] The reference numerals in the figures are:

[0029] 10. SMT processing system; 11. Main control system; 12. Board loading machine; 13. Printing machine; 14. SMT machine; 15. Reflow equipment; 16. Lighting test equipment; 17. Rework machine; 18. Board collecting machine; 19. Image acquisition system. DETAILED DESCRIPTION

[0030] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. Now, in conjunction with the accompanying drawings, the preferred embodiments of the present invention are described in detail.

[0031] Please refer to Figure 1 and Figure 2 , Figure 1 is a flow chart of an embodiment of a deep learning patch method provided by the present invention, Figure 21 is a schematic diagram of the structure of an embodiment of the patch processing system provided by the present invention. The patch processing system 10 includes the following core components: a main control system 11, and a board loading machine 12, a printer 13, a patch machine 14, a reflow soldering device 15, a lighting test device 16 and an image acquisition system 19 connected to the main control system 11, as well as a rework machine 17 and a board receiving machine 18. In this embodiment, the processing stages of the patch processing system include a board loading stage, a printing stage, a patch stage, a reflow soldering stage and a lighting test stage in sequence. In other embodiments, the processing stages of the patch processing system may also include other stages such as a packaging stage or a cleaning stage.

[0032] It is understandable that the relationship between different stages and the quality of execution will have an impact on the subsequent stages. The quality and accuracy of the previous processing stage will directly affect the progress and results of the subsequent stages. For example, lax quality control in the previous processing stage may cause defective products to continue to flow into the subsequent stages, increasing the cost and time of repair. In addition, if there are errors or inaccurate operations in the previous processing stage, these errors may be transmitted or amplified in the subsequent processing stages, resulting in a decline in the overall product quality. Especially when the distance (Pitch value) between the bonded chips continues to shrink, it is even more necessary to strictly control the processing quality of each processing stage, and set the processing parameters of this processing stage accordingly according to the characteristics of the previous processing stage, so that the parameter settings between each processing stage are coordinated and coordinated with each other to ensure the smooth progress of the entire production process and produce high-quality products.

[0033] The deep learning-based patch method provided by the present invention comprises the following steps:

[0034] S101: performing image acquisition for the current processing stage to obtain a processed image.

[0035] In a specific implementation scenario, image acquisition is performed for the current processing stage to obtain a processed image of the current processing stage. A high-resolution industrial camera or visual inspection system can be used as an image acquisition device, and an automatic trigger or manual triggering of the image acquisition device can be set to obtain an image after the current processing stage is processed as a processed image. Multiple processed images can be acquired at one time, and all of the multiple images are used as processed images, or one of them can be selected as a processed image.

[0036] In other implementation scenarios, in order to improve the processing effect of subsequent steps, after the processed image is obtained, the processed image is preprocessed to obtain image features. The image can be processed by one or more of the following image processing methods: for example, the color processed image is converted into a grayscale image by grayscale conversion. A filter (such as a Gaussian filter, a median filter) is used to remove noise from the processed image, reduce interference and improve image quality. The visual effect of the processed image is enhanced by techniques such as histogram equalization and contrast enhancement, making the image clearer and easier to analyze. Morphological processing methods such as expansion, corrosion, opening operation, and closing operation can also be used to improve the shape and structure of objects in the processed image. Edge detection algorithms (such as Sobel and Canny operators) are used to detect edge information in the processed image to highlight the boundary features of the object.

[0037] S102: Analyze the processed image through a deep learning model to obtain processing parameter information of the next processing stage of the current processing stage and / or processing quality information of the current processing stage.

[0038] In a specific implementation scenario, in the field of image processing, deep learning models such as convolutional neural networks (CNN) or recurrent neural networks (RNN) learn the abstract features of images through multi-layer neural networks and reveal how these features affect the parameters of subsequent processing.

[0039] By analyzing the processing images through deep learning models, it is possible to extract and analyze key features in the images, such as component location, defect detection, quality assessment, etc. By training the deep learning model, the processing parameters of the next processing stage, such as pasting position, bonding pressure, temperature setting, etc., can be predicted based on the image data of the current processing stage. This helps to automate the production process and improve production efficiency and consistency. The deep learning model can also be used to analyze images of the current processing stage, identify and evaluate product quality characteristics such as defects, deviations, integrity, etc., and promptly remove unqualified products when the product quality of the current processing stage is found to be unqualified, avoiding processing of unqualified products and resulting in greater waste of resources.

[0040] In other implementation scenarios, the quality parameters of the final product are obtained, and the quality parameters of the final product may include various quality indicators, such as dimensional accuracy, surface finish, component connection strength, power-on status, etc. By associating the quality parameters of the final product with the corresponding processed image data, the deep learning model can learn the relationship between different quality parameters and image features. In this way, the model can better understand the connection between the features in the image and the quality of the final product, thereby improving the accuracy and generalization ability of the model. By correcting the deep learning model based on the quality parameters of the final product, the production system can achieve more accurate quality control and prediction, thereby improving product quality, reducing costs, and enhancing competitiveness. This closed-loop feedback system can continuously improve the production process, making it more intelligent and efficient.

[0041] If the original deep learning model does not work well, that is, the pass rate of quality parameters is not high, you can try to apply transfer learning. By leveraging the knowledge of the already trained model, you can accelerate the training of the new model and improve the performance. Choose an appropriate transfer learning strategy, such as fine-tuning some layers of the deep learning model to adapt to the specific parameter prediction task.

[0042] In one implementation scenario, the deep learning model includes an input layer, a feature extractor, a relationship modeling layer, a fully connected layer, and an output layer. The input layer is used to input processed images, each of which is 64×64 pixels in size. The feature extractor includes two convolutional layers, each of which contains 32 3×3 convolution kernels with a step size of 1 and uses a ReLU activation function. A maximum pooling layer: the pooling window size is 2×2 and the step size is 2. The output feature map size is 16×16. The relationship modeling layer is a bidirectional LSTM layer with 64 hidden units to capture the temporal relationship between features at different stages. The fully connected layer includes 128 neurons and uses a ReLU activation function. Select the corresponding output layer according to the specific task requirements, such as one neuron for a binary classification task, or multiple neurons for a multi-classification or regression task.

[0043] When the current processing stage is the board loading stage, the collected processing images are preprocessed, including grayscale conversion, denoising, image enhancement and other operations to extract the characteristic information of the PCB board, and the trained deep learning model is used to analyze and identify the PCB board features to determine whether the current PCB board meets the manufacturing standards. If the PCB board fails the quality inspection, it may need to be sent back for reprocessing or repair to ensure that the product quality meets the requirements.

[0044] If the PCB board meets the manufacturing standards, that is, passes the quality inspection, then the subsequent printing stage can proceed. Based on the association between the known image features and the corresponding printing parameters, machine learning or expert system methods are used to identify the printing parameters suitable for the current PCB board parameters. The printing parameters include at least one of printing pressure, printing temperature, printing time, printing angle, and number of printings. Based on the identified printing parameters, further optimization and adjustment can be performed to ensure the quality and efficiency of the printing process. The identified and optimized printing parameters are applied to the printing equipment to perform corresponding printing operations on the PCB board.

[0045] When the current processing stage is the printing stage, the printed product image is processed by image processing technology to extract the position, direction, size and other features of the patch. Based on the extracted patch features, machine learning or computer vision technology is used to identify the position of the patch. Target detection algorithms such as YOLO and Faster R-CNN can be used. Based on the printing parameters used by the PCB in the printing stage, the corresponding patch parameters can be obtained, including the adjustment of the patch angle, spacing, rotation and other parameters. According to the predicted patch position and patch parameters, the PCB board is patched to ensure that the patch is accurately placed at the target position of the PCB board to ensure the correctness and quality of the patch.

[0046] When the current processing stage is the patch stage, the corresponding reflow parameters are matched in combination with the current patch parameters for reflow soldering operation on the PCB board. The processed image after patching is processed to extract features related to the reflow soldering temperature curve parameters. For example, image processing technologies such as edge detection, color analysis, and shape recognition can be used. Using a trained deep learning model, the image after patching is input into the deep learning model to match the corresponding reflow parameters. The reflow parameters generally include at least one parameter in the preheating area, the heat action area, the reflow area, the peak temperature, and the cooling area.

[0047] By adjusting the reflow soldering temperature curve, you can ensure that the solder paste is completely melted and forms uniform solder joints, thereby improving the soldering quality. Proper temperature curve adjustment can reduce soldering defects such as cold soldering, warping, and cold soldering. Therefore, optimizing the reflow soldering temperature curve can reduce the incidence of soldering defects and improve the consistency and reliability of soldering.

[0048] When the current processing stage is the lighting test stage, the processing image of the chip position and whether it is lit on the PCB board after reflow soldering is obtained. After inputting the deep learning model, the model will output the prediction result of each chip position, indicating whether there is a test failure point.

[0049] Based on the model's prediction results, the chip locations identified as possible test failure points are marked. The chip locations marked as possible test failure points are manually verified. Confirm whether there are actual test failure points through manual inspection to ensure the accuracy of the model. If manual verification confirms the existence of test failure points, timely measures need to be taken to repair or process them, such as re-soldering chips, replacing components, or performing other repair operations. Record the model's prediction results and subsequent processing processes, and use this data to analyze common problems in the production process, optimize production processes, and improve deep learning models.

[0050] The deep learning model is continuously improved based on the results of manual verification and actual processing conditions. Processing parameter data including PCB board parameters, printing parameters, patch parameters and reflow parameters are collected. These parameters can cover information such as board characteristics, printer settings, patch machine parameters and reflow oven temperature. The processing parameters of the test failure point are matched with the actual results to establish a knowledge base of parameter matching values. This knowledge base can record the problems that occur under different parameter combinations and their corresponding solutions. The knowledge accumulation of parameter matching values ​​is fed back to the deep learning model for re-learning and model updating. Based on the re-learned deep learning model, the operation of the next PCB board is predicted and corrected. The model can detect possible processing parameter combinations that may cause problems in advance based on the previous knowledge accumulation and the learned association rules, and put forward corresponding correction suggestions. Continuously record new test failure point data and processing parameter information, continuously optimize the model and update the knowledge base. Through continuous learning and improvement, the accuracy and predictive ability of the model can be continuously improved, thereby achieving better production results.

[0051] The accuracy and stability of the model can be improved by adding more training data, adjusting the model structure, or optimizing hyperparameters. Furthermore, the test failure points found are incorporated into the quality control system to prevent similar problems from happening again. At the same time, timely feedback on the performance of the model and the improvement points in the production process is provided to continuously optimize the production process.

[0052] From the above description, it can be seen that in this embodiment, by analyzing the processing image through the deep learning model, automatic monitoring and analysis of the current processing stage can be achieved. This helps to reduce manual intervention, improve production efficiency and accuracy, and by obtaining the processing parameter information and / or quality information of the current processing stage, real-time feedback can be achieved, and the processing parameters can be automatically adjusted to ensure product quality and production efficiency, and reduce errors and losses in production.

[0053] The present invention provides an intelligent production device. The intelligent production device includes a processor and a memory. The processor is coupled to the memory. A computer program is stored in the memory, and the processor executes the computer program when working to implement the above method. The detailed steps can be referred to above and will not be repeated here.

[0054] The present invention provides a computer-readable storage medium. The computer-readable storage medium stores at least one computer program, which is used to be executed by a processor to implement the above method. The detailed steps can be referred to above and will not be repeated here. In one embodiment, the computer-readable storage medium can be a storage chip, a hard disk, or a mobile hard disk or a USB flash drive, an optical disk, or other readable and writable storage tools in a terminal, or a server, etc.

[0055] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0056] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0057] It should be understood that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. For those skilled in the art, the technical solutions described in the above embodiments can be modified, or some of the technical features therein can be replaced by equivalents; and all these modifications and replacements should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A patch method based on deep learning, characterized in that: Applied to patch processing system, The patch method based on deep learning includes: Perform image acquisition for the current processing stage to obtain the processed image; The processed image is analyzed by a deep learning model to obtain processing parameter information of the next processing stage of the current processing stage and / or processing quality information of the current processing stage.

2. The patch method based on deep learning according to claim 1, characterized in that: The processing stages of the patch processing system include: The board mounting stage, printing stage, SMT stage, reflow stage and lighting test stage.

3. The patch method based on deep learning according to claim 2, characterized in that: The step of obtaining processing parameter information of a next processing stage of the current processing stage and / or processing quality information of the current processing stage includes: When the current processing stage is the board loading stage, it is detected whether the current PCB board meets the manufacturing standards. If the PCB board meets the manufacturing standards, the corresponding printing parameters are matched in combination with the current PCB board parameters for printing the PCB board.

4. The patch method based on deep learning according to claim 2, characterized in that: The step of obtaining processing parameter information of a next processing stage of the current processing stage and / or processing quality information of the current processing stage includes: When the current processing stage is the printing stage, the corresponding patch parameters are matched in combination with the current printing parameters to perform the patch operation on the PCB board.

5. The patch method based on deep learning according to claim 2, characterized in that: The step of obtaining processing parameter information of a next processing stage of the current processing stage and / or processing quality information of the current processing stage includes: When the current processing stage is the patch stage, the corresponding reflow parameters are matched in combination with the current patch parameters to perform a reflow soldering operation on the PCB board.

6. The patch method based on deep learning according to claim 2, characterized in that: The step of obtaining processing parameter information of a next processing stage of the current processing stage and / or processing quality information of the current processing stage includes: When the current processing stage is the lighting test stage, the processing parameters of the test failure point are obtained, and the processing parameters include at least one of PCB board parameters, printing parameters, patch parameters and reflow parameters. Re-learning is performed based on the processing parameters to obtain knowledge accumulation of parameter matching values ​​to make predictive corrections to the operation of the next PCB board.

7. The deep learning-based patch method according to any one of claims 1 to 6, characterized in that: After the step of obtaining the processed image, the method further comprises: Perform image preprocessing on the processed image to obtain image features.

8. The patch method based on deep learning according to claim 1, characterized in that: The patch method based on deep learning also includes: The quality parameters of the final product are obtained, and the deep learning model is modified based on the quality parameters.

9. An intelligent production equipment, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 8.

Images