Method for identifying non-standard fluxing behavior during a circuit board soldering process

Abstract

The present application relates to a kind of identification methods of the behavior of not standard fluxing in circuit board welding process, for solving the behavior of not standard fluxing in real time in circuit board welding process cannot be found.The present application includes the following steps: step one, the training of circuit board detection model and circuit board classification model;Step two, the detection rectangular frame of circuit board is obtained by carrying out circuit board detection to the i frame image of the jth circuit board obtained by real-time acquisition;Step three, the unique ID and tracking frame corresponding to each circuit board are obtained by tracking detection rectangular frame by target tracking algorithm;Step four, according to tracking frame, the rectangular area where circuit board is located is cut out, and the state of circuit board is identified by circuit board classification model;Step five, the time of welding operation and coating operation is recorded respectively, whether the difference of two times is within time limit is calculated;After judging, it is turned to step two until circuit board identification is completed.

Description

Technical Field

[0001] This invention relates to target detection, tracking, and classification in the field of computer vision, specifically to a method for identifying improper flux application during circuit board soldering. Background Technology

[0002] With the widespread use of electronic products in various fields, their reliability is receiving increasing attention. Soldering is an indispensable part of modern electronic product manufacturing, and its process often determines the reliability of the entire electronic product. Soldering quality problems are one of the main causes of circuit board failures.

[0003] Soldering methods are divided into manual soldering and machine soldering. Machine soldering can automatically complete the soldering process, is simple to operate, and is fast, which can significantly improve production efficiency, reduce human error in the soldering process, and improve production process level and product quality. However, machine soldering is mainly used for mass production of electronic components. When small-batch processing is involved, manual soldering is still the fastest and most economical method. Moreover, the repair and rework of existing circuit boards still need to be done manually.

[0004] Manual soldering is a complex and labor-intensive process. Improper operation at any stage can affect soldering quality and increase the defect rate. Applying flux is a crucial step in the soldering process; it helps fix component positions, accelerates solder melting, increases wettability, slows oxidation, and reduces the occurrence of cold solder joints and bridging. However, in manual soldering, improper operation can lead to forgetting to apply flux or applying it incorrectly, thus increasing the defect rate.

[0005] Current research on welding quality issues mainly focuses on solders, machine welding, and welding knowledge bases, with little research on real-time, automatic identification methods for non-standard manual welding operations. Summary of the Invention

[0006] This invention provides a method for identifying improper flux application during circuit board soldering. It utilizes computer vision technologies such as target detection, tracking, and classification to address the inability to detect improper flux application in real time during circuit board soldering.

[0007] A method for identifying improper flux application during circuit board soldering is characterized by the following steps:

[0008] Step 1: Training the circuit board detection model and the circuit board classification model;

[0009] Step 2: In real time, the i-th frame image of the circuit board numbered j during the soldering process is acquired. The acquired image is then identified by the circuit board detection model to obtain the detection rectangle containing the circuit board in the image.

[0010] Step 3: Track the detection rectangle using a target tracking algorithm to obtain a unique ID and tracking box for each circuit board;

[0011] Step 4: Based on the tracking frame of the circuit board, cut out the rectangular area where the circuit board is located, and identify the status of the circuit board through the circuit board classification model; if the status of the circuit board is that it is being soldered or being coated with flux, then proceed to step 5.

[0012] If the circuit board is in a state where no soldering and flux application have been performed, proceed to step two until all circuit boards have been identified.

[0013] Step 5: If the circuit board is in the process of applying flux, record the time T at this moment. j,fluxing Then proceed to step two; if the circuit board is in the process of soldering, record the time as T. j,welding And determine T j,welding The most recently recorded T on this circuit board j,fluxing Is the difference within the set time limit T? thr If the application is not completed within the time limit, the circuit board is deemed to have been improperly coated with flux. The operator is immediately alerted, and an alarm message is sent to the safety management personnel. If the application is completed within the time limit, the coating and soldering processes are considered normal, and then proceed to step two.

[0014] Furthermore, step one is detailed as follows:

[0015] The circuit board detection dataset is fed into the pre-trained circuit board model for training, and the final circuit board detection model is obtained.

[0016] The circuit board classification dataset is substituted into the circuit board classification pre-training model for training, and finally the circuit board classification model is obtained.

[0017] Furthermore, in step one, the pre-trained model for circuit board detection is a YOLOV5 pre-trained model;

[0018] The circuit board classification pre-trained model is a ResNet50 pre-trained model;

[0019] The circuit board classification dataset includes circuit boards that are being soldered, circuit boards that are being coated with flux, and circuit boards that have not undergone soldering or flux coating operations.

[0020] Furthermore, in step three, the target tracking algorithm is the DeepSORT target tracking algorithm.

[0021] Furthermore, in step five, T thr = 60 seconds.

[0022] The beneficial effects of this invention are:

[0023] This invention enables real-time and automatic identification of improper flux application during circuit board soldering, promptly reminding workers to follow proper procedures, effectively improving safety management of circuit board soldering work, ensuring product quality and production efficiency, and freeing up the productivity of safety management personnel. Attached Figure Description

[0024] Figure 1 This is a flowchart of the overall processing flow of the uncoated flux behavior monitoring algorithm in this invention;

[0025] Figure 2 This is a flowchart illustrating the behavior recognition of uncoated flux in an embodiment of the present invention. Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0027] This invention provides a method for identifying improper flux application during circuit board soldering, the process of which is as follows: Figure 1 As shown, the method of the present invention first acquires a real-time video stream from a camera; then, it identifies and tracks the circuit board using a deep learning object detection method; next, it determines whether a soldering operation or a flux application operation is being performed on the circuit board using a deep learning classification model; finally, it determines whether the operator has applied flux correctly based on the flux application during the soldering process, thus realizing the monitoring of the lack of flux application during manual soldering of circuit boards. The specific steps are as follows:

[0028] Step 1: Circuit board detection model training.

[0029] Circuit board materials are collected, labeled, and generated to obtain a circuit board detection dataset. The circuit board detection dataset is then fed into a YOLOv5 pre-trained model for training, ultimately resulting in a circuit board detection YOLOv5 model. This model can identify circuit boards from images and generate circuit board detection bounding boxes.

[0030] Step 2, circuit board inspection.

[0031] The camera or webcam captures the i-th frame image of the circuit board numbered j during the soldering process in real time. The captured image is then used to identify the circuit board using the YOLOv5 model, resulting in the detection bounding box of the circuit board in the image.

[0032] Step 3, circuit board tracing.

[0033] The DeepSORT target tracking algorithm is used to track the detection rectangles, obtaining a unique ID and tracking box for each circuit board.

[0034] Step 4: Training the circuit board classification model.

[0035] Based on the circuit board's tracking bounding box, a rectangular region containing the circuit board is cropped, and the circuit boards are categorized into three types: those undergoing coating, those undergoing soldering, and those not undergoing coating or soldering operations, resulting in a circuit board classification dataset. This classification dataset is then fed into a ResNet50 pre-trained model for training, ultimately producing a circuit board classification ResNet50 model. This model can identify the state of a circuit board based on the rectangular region containing the circuit board as defined by the tracking bounding box.

[0036] Step 5: Classify the circuit board status.

[0037] Based on the tracking frame of the circuit board, the rectangular area where the circuit board is located is cropped. The circuit board is then classified using the ResNet50 classification model to identify the state of the circuit board. The states of the circuit board are divided into: circuit board being soldered, circuit board being coated with flux, and circuit board not being soldered or coated with flux.

[0038] If the circuit board is in the process of soldering or applying flux, proceed to step six; if the circuit board is not being soldered or fluxed, proceed to step two, until all circuit boards have been identified.

[0039] Step 6: Determination of improper flux application.

[0040] Before soldering, flux needs to be applied, and soldering must be completed within a certain time limit after the flux is applied to the circuit board; otherwise, it needs to be reapplied.

[0041] like Figure 2 As shown, identify the circuit board operation type numbered j in the i-th frame image; if the circuit board is performing a flux coating operation, record the time T at this moment. j,fluxing If the circuit board is undergoing soldering, record the time of this moment as T. j,welding And through formula T j,welding -T j,fluxing >T thr Calculate T j,welding With T j,fluxing Is the difference within the set time limit T? thr Inside, T thr =60(s). If the time limit is not met, the circuit board numbered j is determined to have improper flux application (no flux applied or flux applied but expired). The operator is immediately alerted, and the alarm information is sent to safety management personnel to ensure standardized operation and improve product quality and production efficiency. After the analysis of each target in each frame is completed, proceed to step two to continue analyzing the next frame.

Claims

1. A method for identifying improper flux application during circuit board soldering, characterized in that, The steps are as follows: Step 1: Training the circuit board detection model and the circuit board classification model; details are as follows: The circuit board detection dataset is fed into the pre-trained circuit board model for training, and the final circuit board detection model is obtained. The circuit board classification dataset is substituted into the circuit board classification pre-training model for training, and the final circuit board classification model is obtained. Step 2: Real-time acquisition of the i-th frame image of the circuit board numbered j during the soldering process. The acquired image is then processed by the trained circuit board detection model to obtain the detection bounding box containing the circuit board in the image. Step 3: Track the detection rectangle using a target tracking algorithm to obtain a unique ID and tracking box for each circuit board; Step 4: Based on the tracking frame of the circuit board, cut out the rectangular area where the circuit board is located. After training, the circuit board classification model identifies the state of the circuit board. If the state of the circuit board is that it is being soldered or being coated with flux, proceed to step 5. If the circuit board is in a state where no soldering and flux application have been performed, proceed to step two until all circuit boards have been identified. Step 5: If the circuit board is in the process of applying flux, record the time T at this moment. j,fluxing Then proceed to step two; if the circuit board is in the process of soldering, record the time as T. j,welding And determine T j,welding The most recently recorded T on this circuit board j,fluxing Is the difference within the set time limit T? thr Inside; If the application is not completed within the time limit, the circuit board is deemed to have been improperly coated with flux. The operator is immediately alerted, and an alarm message is sent to the safety management personnel. If the application is completed within the time limit, the coating and soldering processes are considered normal, and then the process proceeds to step two.

2. The method for identifying improper flux application during circuit board soldering as described in claim 1, characterized in that: In step one, the pre-trained model for circuit board detection is a YOLOV5 pre-trained model; the pre-trained model for circuit board classification is a ResNet50 pre-trained model. The circuit board classification dataset includes circuit boards that are being soldered, circuit boards that are being coated with flux, and circuit boards that have not undergone soldering or flux coating operations.

3. The method for identifying improper flux application during circuit board soldering as described in claim 1 or 2, characterized in that: In step three, the target tracking algorithm is the DeepSORT target tracking algorithm.

4. The method for identifying improper flux application during circuit board soldering according to claim 2, characterized in that: In step five, T thr =60 seconds.

Images