Smt production process quality control method and system
By deploying an electromagnetic coil array and infrared thermal imaging technology around the reflow oven, combined with solder paste segmentation and electromagnetic coil parameter adjustment models, the electromagnetic field parameters are dynamically adjusted, solving the problem of uneven solder paste flow and improving welding quality and stability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN KESIJIA TECHNOLOGY CO LTD
- Filing Date
- 2025-09-08
- Publication Date
- 2026-06-09
Smart Images
Figure CN121104440B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of industrial control technology, specifically to a quality control method and system for SMT production process. Background Technology
[0002] In surface mount technology (SMT) manufacturing, reflow soldering is one of the key processes, and the flowability of solder paste plays a crucial role in soldering quality. Solder paste is the bonding material between surface mount components and the PCB board. It melts and flows during heating to form solid solder joints. However, uneven or insufficient solder paste flow can lead to a series of soldering defects, such as solder bridging, cold solder joints, and uneven solder joints. These defects directly affect the reliability and performance of the product. Summary of the Invention
[0003] This invention utilizes an electromagnetic coil array deployed around a reflow oven, combined with infrared thermal imaging technology, to acquire real-time thermal images of the PCB board. These images are then processed in a solder paste segmentation model to accurately characterize the solder paste's flow state. Based on the solder paste distribution thermal map, the system can dynamically adjust the control parameters of the electromagnetic coil array, including current intensity, current frequency, and waveform, thereby precisely regulating the electromagnetic field, optimizing solder paste flow, preventing excessive or insufficient solder paste accumulation, and reducing the risk of solder bridging and cold solder joints. Through a reinforcement learning mechanism, the system automatically optimizes the electromagnetic field adjustment parameters based on feedback from changes in the solder paste distribution thermal map before each use, enhancing the system's adaptability and accuracy. This ensures that the solder paste remains in an optimal flow state throughout the soldering process, significantly improving soldering quality and process stability.
[0004] This invention provides a quality control method for SMT production process, comprising:
[0005] An electromagnetic coil array is set up around the reflow soldering oven. A thermal imager is set up in the reflow soldering oven to acquire the PCB thermal image in real time. The PCB thermal image is then sent to the solder paste segmentation model for processing and outputs a solder paste distribution thermal map.
[0006] The solder paste distribution heatmap is fed into the electromagnetic coil array parameter adjustment model for processing, and the electromagnetic coil array parameter adjustment set is output. The electromagnetic coil array parameter adjustment set includes N electromagnetic coil parameters, where N is the total number of electromagnetic coils in the electromagnetic coil array. The electromagnetic coil array parameters include the position of the electromagnetic coils and the electromagnetic coil control parameters. In addition, the electromagnetic coil array parameter adjustment model performs reinforcement learning operations based on the changes in the solder paste distribution heatmap before each use.
[0007] As a preferred approach, reinforcement learning is performed on the electromagnetic coil array parameter adjustment model based on feedback from changes in the solder paste distribution heatmap. Specifically, this includes the following operations:
[0008] The rate of change of the mean square error of pixel values in the current solder paste distribution heatmap compared to the mean square error of pixel values in the previous solder paste distribution heatmap is used as the reward value.
[0009] The previous solder paste distribution heatmap is fed into the target electromagnetic coil array parameter adjustment model for processing, outputting the target electromagnetic coil array parameter adjustment set. The previous solder paste distribution heatmap is then flattened into a one-dimensional vector, denoted as the solder paste distribution vector. Specifically, each pixel in the previous solder paste distribution heatmap is expanded into a pixel vector, including pixel coordinates and corresponding pixel values. All pixel vectors are concatenated end-to-end to form the solder paste distribution vector. This vector is then concatenated end-to-end with all electromagnetic coil parameters from the previously output electromagnetic coil array parameter adjustment set to form the adjustment strategy evaluation data. This data is fed into the adjustment strategy evaluation network for processing, outputting the adjustment strategy evaluation value. The evaluation value and reward value are then weighted and summed to form the gradient value. Based on this gradient value, the gradient ascent method is used to adjust the parameters of the electromagnetic coil array parameter adjustment model, achieving reinforcement learning of the model. The target electromagnetic coil array parameter adjustment model is also updated in real time. Initially, the target electromagnetic coil array parameter adjustment model is consistent with the target electromagnetic coil array parameter adjustment model.
[0010] As a preferred aspect, the parameter adjustment model for the target electromagnetic coil array is updated in real time, specifically including the following steps:
[0011] Each time the electromagnetic coil array parameter adjustment model is updated, the target electromagnetic coil array parameter adjustment model is updated in real time using the following formula: θ k (new) = δη k +θ k (bef), where k = 1, 2, 3, ..., K, K is the total number of parameter values in the electromagnetic coil array parameter adjustment model, θ k (new) represents the k-th parameter value in the updated electromagnetic coil array parameter adjustment model, δ is the update rate, and η is the parameter value. k To adjust the value of the k-th parameter in the updated electromagnetic coil array parameter adjustment model, θ k (bef) The kth parameter value in the target electromagnetic coil array parameter adjustment model before the update.
[0012] As a preferred approach, training the solder paste segmentation model includes the following steps:
[0013] Several solder paste segmentation training samples are obtained, including PCB thermal images. The solder paste distribution heatmaps are used to annotate the training samples. All annotated training samples are combined into a solder paste segmentation training set. The solder paste segmentation model is trained using the training set. The accuracy of the solder paste segmentation model is then determined to be higher than a set threshold. If the accuracy of the solder paste segmentation model is higher than the set threshold, the trained solder paste segmentation model is output. Otherwise, the solder paste segmentation model is trained again using the training set.
[0014] As a preferred approach, training a parameter adjustment model for an electromagnetic coil array involves the following steps:
[0015] Several training samples for adjusting electromagnetic coil array parameters are obtained, including solder paste distribution heatmaps. These training samples are then labeled using an electromagnetic coil array parameter adjustment set. All labeled training samples are combined into a training set. The electromagnetic coil array parameter adjustment model is trained using this training set. The accuracy of the model is then determined to be higher than a set threshold. If the accuracy is higher than the threshold, the trained model is output; otherwise, the model is trained again using the training set.
[0016] As a preferred approach, training the policy evaluation network involves the following steps:
[0017] Obtain several training samples for adjusting strategy evaluation, which include pre-constructed adjusting strategy evaluation data. Label the training samples using the adjusting strategy evaluation values. Combine all labeled training samples into an adjusting strategy evaluation training set. Train the adjusting strategy evaluation network using the training set. Determine if the accuracy of the adjusting strategy evaluation network is higher than a corresponding set threshold. If the accuracy is higher than the set threshold, output the trained adjusting strategy evaluation network; otherwise, continue training the adjusting strategy evaluation network using the training set.
[0018] The present invention also provides a quality control system for SMT production process, comprising:
[0019] The real-time solder paste distribution acquisition module is used to deploy an electromagnetic coil array around the reflow oven, acquire the PCB thermal image of the PCB board in real time through a thermal imager deployed in the reflow oven, and then send the PCB thermal image into the solder paste segmentation model for processing to output a solder paste distribution thermal map.
[0020] The electromagnetic coil array parameter adjustment module is used to send the solder paste distribution heat map into the electromagnetic coil array parameter adjustment model for processing and output the electromagnetic coil array parameter adjustment set. The electromagnetic coil array parameter adjustment set includes N electromagnetic coil parameters, where N is the total number of electromagnetic coils in the electromagnetic coil array. The electromagnetic coil array parameters include the position of the electromagnetic coils and the electromagnetic coil control parameters.
[0021] The reinforcement learning module is used to perform reinforcement learning operations on the electromagnetic coil array parameter adjustment model based on feedback from changes in the solder paste distribution heatmap.
[0022] The present invention has the following advantages:
[0023] This invention utilizes an electromagnetic coil array deployed around a reflow oven, combined with infrared thermal imaging technology, to acquire real-time thermal images of the PCB board. These images are then processed in a solder paste segmentation model to accurately characterize the solder paste's flow state. Based on the solder paste distribution thermal map, the system can dynamically adjust the control parameters of the electromagnetic coil array, including current intensity, current frequency, and waveform, thereby precisely regulating the electromagnetic field, optimizing solder paste flow, preventing excessive or insufficient solder paste accumulation, and reducing the risk of solder bridging and cold solder joints. Through a reinforcement learning mechanism, the system automatically optimizes the electromagnetic field adjustment parameters based on feedback from changes in the solder paste distribution thermal map before each use, enhancing the system's adaptability and accuracy. This ensures that the solder paste remains in an optimal flow state throughout the soldering process, significantly improving soldering quality and process stability. Attached Figure Description
[0024] Figure 1 This is a schematic diagram of the SMT production process quality control system used in an embodiment of the present invention. Detailed Implementation
[0025] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.
[0026] Example 1: A quality control method for SMT production process, comprising:
[0027] This implementation targets the reflow soldering process in SMT production. In SMT production, a layer of solder paste needs to be printed on the pads of the PCB using a stencil. Then, the surface mount components are accurately placed on the pads coated with solder paste using a pick-and-place machine. Finally, the mounted PCB is passed through a reflow oven to melt the solder paste and form reliable solder joints with the component leads and pads.
[0028] An array of electromagnetic coils is deployed around the reflow oven. A thermal imager installed inside the reflow oven acquires real-time thermal images of the PCB board. These images are then processed by a solder paste segmentation model to output a solder paste distribution thermal map. This map reflects the flow state of the solder paste at various locations at the current moment. Because the thermal conductivity and specific heat capacity of the solder paste differ from other objects on the PCB board, the segmentation model segments the corresponding parts of the solder paste in the PCB thermal image. To accurately obtain the solder paste state, the position of the PCB board is analyzed in actual operation to determine the wavelength selection for the thermal imager. The solder paste segmentation model is based on the U-net model.
[0029] The solder paste distribution heatmap is fed into the electromagnetic coil array parameter adjustment model for processing, outputting an electromagnetic coil array parameter adjustment set. This set includes N electromagnetic coil parameters, where N is the total number of electromagnetic coils in the array. The electromagnetic coil array parameters include the coil positions and control parameters, such as current intensity, current frequency, and current waveform. Based on this parameter adjustment set, the electromagnetic coil array is controlled to ensure uniform solder paste flow, preventing excessive or insufficient solder paste accumulation on the pads and reducing the risk of solder bridging or cold solder joints. Furthermore, the electromagnetic coil array parameter adjustment model undergoes reinforcement learning operations before each use, based on feedback from changes in the solder paste distribution heatmap. The model is composed of a convolutional neural network and a multilayer perceptron.
[0030] An electromagnetic coil array can generate a localized, transient electromagnetic field. This transient electromagnetic field acts on the solder paste at the pads. Due to the presence of metal particles (tin, lead, and silver, etc.) in the solder paste, the transient electromagnetic field changes the flow direction of the solder paste, thereby regulating its flow state. The flow state of the solder paste affects the quality of the solder joint. The specific mechanisms are as follows: uneven solder paste flow leads to uneven melting of solder paste on different pads, resulting in solder bridging or over-soldering. Meanwhile, other areas may have too little solder paste, leading to cold solder joints or incomplete soldering. Furthermore, uneven solder paste flow can result in poor wetting in certain areas of the pads, potentially causing soldering defects such as cold solder joints or poor soldering.
[0031] This invention utilizes an electromagnetic coil array deployed around a reflow oven, combined with infrared thermal imaging technology, to acquire real-time thermal images of the PCB board. These images are then processed in a solder paste segmentation model to accurately characterize the solder paste's flow state. Based on the solder paste distribution thermal map, the system can dynamically adjust the control parameters of the electromagnetic coil array, including current intensity, current frequency, and waveform, thereby precisely regulating the electromagnetic field, optimizing solder paste flow, preventing excessive or insufficient solder paste accumulation, and reducing the risk of solder bridging and cold solder joints. Through a reinforcement learning mechanism, the system automatically optimizes the electromagnetic field adjustment parameters based on feedback from changes in the solder paste distribution thermal map before each use, enhancing the system's adaptability and accuracy. This ensures that the solder paste remains in an optimal flow state throughout the soldering process, significantly improving soldering quality and process stability.
[0032] The electromagnetic coil array parameter adjustment model is subjected to reinforcement learning operations based on feedback from changes in the solder paste distribution heatmap. Specifically, the operations include the following:
[0033] The rate of change of the mean square error of pixel values in the current solder paste distribution heatmap compared to the mean square error of pixel values in the previous solder paste distribution heatmap is calculated as a reward value. This reward value characterizes the feedback of changes in the solder paste distribution heatmap. The mean square error of pixel values in the solder paste distribution heatmap reflects the distribution of temperature values. The more uniform the temperature distribution, the more uniform the flow state of the solder paste and the more uniform the heating. The rate of change reflects the change in temperature distribution after the electromagnetic coil array parameter adjustment set is applied. Considering that the temperature curve in the reflow oven is changing, the heating of the solder paste will also be in different states. Therefore, it is necessary to adjust the parameters in the electromagnetic coil array parameter adjustment model through the feedback of changes in the solder paste distribution heatmap.
[0034] The previous solder paste distribution heatmap is fed into the target electromagnetic coil array parameter adjustment model for processing, outputting the target electromagnetic coil array parameter adjustment set. The previous solder paste distribution heatmap is then flattened into a one-dimensional vector, denoted as the solder paste distribution vector. Specifically, each pixel in the previous solder paste distribution heatmap is expanded into a pixel vector, including pixel coordinates and corresponding pixel values. All pixel vectors are concatenated end-to-end to form the solder paste distribution vector, following the left-to-right and top-to-bottom order in the solder paste distribution heatmap. The solder paste distribution vector is then concatenated end-to-end with all electromagnetic coil parameters from the previously output electromagnetic coil array parameter adjustment set to form the adjustment strategy evaluation data. This adjustment strategy evaluation data is fed into the adjustment strategy evaluation network for processing, outputting the adjustment strategy evaluation value. The adjustment strategy evaluation network is built based on a BP neural network. The gradient value is then formed by weighted summation of the adjustment strategy evaluation value and the reward value. The weights in this weighted summation are set by the operator. Based on the gradient value, the gradient ascent method is used to adjust the parameters of the electromagnetic coil array parameter adjustment model, thereby achieving reinforcement learning of the electromagnetic coil array parameter adjustment model. It should be noted that performing parameter adjustment through the target electromagnetic coil array parameter adjustment set output by the target electromagnetic coil array parameter adjustment model ensures that the electromagnetic coil array parameter adjustment model is updated stably under the premise that the target electromagnetic coil array parameter adjustment model serves as a baseline. Furthermore, the target electromagnetic coil array parameter adjustment model is also updated in real time. In the initial state, the target electromagnetic coil array parameter adjustment model is consistent with the electromagnetic coil array parameter adjustment model. The initial state here refers to the state where the electromagnetic coil array parameter adjustment model has been trained and is about to be put into practical use.
[0035] The parameter adjustment model for the target electromagnetic coil array is updated in real time, specifically including the following steps:
[0036] Each time the electromagnetic coil array parameter adjustment model is updated, the target electromagnetic coil array parameter adjustment model is updated in real time using the following formula: θ k (new) = δη k +θ k (bef), where k = 1, 2, 3, ..., K, K is the total number of parameter values in the electromagnetic coil array parameter adjustment model, θ k (new) represents the k-th parameter value in the updated electromagnetic coil array parameter adjustment model, δ is the update rate, typically 0.01, and η k To adjust the value of the k-th parameter in the updated electromagnetic coil array parameter adjustment model, θ k (bef) The kth parameter value in the target electromagnetic coil array parameter adjustment model before the update.
[0037] Training the solder paste segmentation model involves the following steps:
[0038] Several solder paste segmentation training samples are obtained, including PCB thermal images. These samples are collected by operators based on actual reflow soldering processes and labeled using solder paste distribution heatmaps. These heatmaps are manually created by operators combining expert experience. All labeled training samples are combined into a solder paste segmentation training set. The solder paste segmentation model is trained using this training set. The accuracy of the model is then assessed to determine if it exceeds a set threshold. This threshold is set by the operator based on the segmentation results. If the accuracy exceeds the threshold, the trained model is output; otherwise, the model is trained again using the training set.
[0039] Training the electromagnetic coil array parameter adjustment model involves the following steps:
[0040] Several training samples for adjusting electromagnetic coil array parameters are obtained. These training samples include solder paste distribution heatmaps. These heatmaps can be manually segmented by operators based on actual PCB thermal images, or they can be output by a trained solder paste segmentation model. The training samples are then labeled using an electromagnetic coil array parameter adjustment set. This set represents the optimal strategy calculated by the operator based on the actual situation corresponding to the solder paste distribution heatmap. All labeled training samples are combined into a training set. The electromagnetic coil array parameter adjustment model is trained using this training set. The accuracy of the model is then checked against a set threshold. If the accuracy exceeds the threshold, the trained model is output; otherwise, the model continues to be trained using the training set.
[0041] Training the policy adjustment evaluation network involves the following steps:
[0042] Several training samples for adjusting strategy evaluation are obtained. These training samples include pre-constructed adjusting strategy evaluation data, which originates from the solder paste distribution heatmap and corresponding labeled electromagnetic coil array parameter adjustment set during the training of the electromagnetic coil array parameter adjustment model. The training samples are labeled with adjusting strategy evaluation values. Since the electromagnetic coil array parameter adjustment set is calculated by the operator based on the actual situation corresponding to the solder paste distribution heatmap, the adjusting strategy evaluation value is generally set to 1. All labeled adjusting strategy evaluation training samples are combined into an adjusting strategy evaluation training set. The adjusting strategy evaluation network is trained using this training set. The accuracy of the adjusting strategy evaluation network is then judged to be higher than the corresponding set threshold. If the accuracy of the adjusting strategy evaluation network is higher than the corresponding set threshold, the trained adjusting strategy evaluation network is output; otherwise, the adjusting strategy evaluation network is trained again using the training set.
[0043] Example 2, a quality control system for SMT production process, see [link / reference] Figure 1 ,include:
[0044] The real-time solder paste distribution acquisition module is used to acquire real-time thermal images of the PCB board using a thermal imager deployed in the reflow oven. The PCB thermal images are then sent to the solder paste segmentation model for processing, and a solder paste distribution thermal map is output. The solder paste distribution thermal map can reflect the flow state of the solder paste at various locations at the current moment. Since the thermal conductivity and specific heat capacity of the solder paste are different from other objects in the PCB board, the segmentation model is used to segment the part corresponding to the solder paste in the PCB thermal image. In order to accurately acquire the solder paste state, the position of the PCB board is analyzed in actual operation to determine the wavelength selection of the thermal imager.
[0045] The electromagnetic coil array parameter adjustment module is used to input the solder paste distribution heat map into the electromagnetic coil array parameter adjustment model for processing, and output the electromagnetic coil array parameter adjustment set. The electromagnetic coil array parameter adjustment set includes N electromagnetic coil parameters, where N is the total number of electromagnetic coils in the electromagnetic coil array. The electromagnetic coil array parameters include the position of the electromagnetic coils and the electromagnetic coil control parameters. The electromagnetic coil control parameters include current intensity, current frequency, and current waveform, etc. Based on the electromagnetic coil array parameter adjustment set, the electromagnetic coil array is controlled to make the solder paste flow evenly, avoid excessive or insufficient solder paste accumulation on the pads, and reduce the risk of solder bridges or cold solder joints.
[0046] The reinforcement learning module is used to perform reinforcement learning operations on the electromagnetic coil array parameter adjustment model based on feedback from changes in the solder paste distribution heatmap.
[0047] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.
Claims
1. A quality control method for SMT production process, characterized in that, include: An electromagnetic coil array is set up around the reflow soldering oven. A thermal imager is set up in the reflow soldering oven to acquire the PCB thermal image in real time. The PCB thermal image is then sent to the solder paste segmentation model for processing and outputs a solder paste distribution thermal map. The solder paste distribution heatmap is fed into the electromagnetic coil array parameter adjustment model for processing, and the electromagnetic coil array parameter adjustment set is output. The electromagnetic coil array parameter adjustment set includes N electromagnetic coil parameters, where N is the total number of electromagnetic coils in the electromagnetic coil array. The electromagnetic coil array parameters include the position of the electromagnetic coils and the electromagnetic coil control parameters. Furthermore, the electromagnetic coil array parameter adjustment model performs reinforcement learning operations based on the changes in the solder paste distribution heatmap before each use. The electromagnetic coil array parameter adjustment model is subjected to reinforcement learning operations based on feedback from changes in the solder paste distribution heatmap. Specifically, the operations include the following: The rate of change of the mean square error of pixel values in the current solder paste distribution heatmap compared to the mean square error of pixel values in the previous solder paste distribution heatmap is used as the reward value. The previous solder paste distribution heatmap is fed into the target electromagnetic coil array parameter adjustment model for processing, outputting the target electromagnetic coil array parameter adjustment set. The previous solder paste distribution heatmap is then flattened into a one-dimensional vector, denoted as the solder paste distribution vector. Specifically, each pixel in the previous solder paste distribution heatmap is expanded into a pixel vector, including pixel coordinates and corresponding pixel values. All pixel vectors are concatenated end-to-end to form the solder paste distribution vector. This vector is then concatenated end-to-end with all electromagnetic coil parameters from the previously output electromagnetic coil array parameter adjustment set to form adjustment strategy evaluation data. This data is fed into the adjustment strategy evaluation network for processing, outputting the adjustment strategy evaluation value. The evaluation value and reward value are then weighted and summed to form the gradient value. Based on this gradient value, the gradient ascent method is used to adjust the parameters of the electromagnetic coil array parameter adjustment model, achieving reinforcement learning of the model. The target electromagnetic coil array parameter adjustment model is also updated in real time. Initially, the target electromagnetic coil array parameter adjustment model is consistent with the target electromagnetic coil array parameter adjustment model. The parameter adjustment model for the target electromagnetic coil array is updated in real time, specifically including the following steps: Each time the electromagnetic coil array parameter adjustment model is updated, the target electromagnetic coil array parameter adjustment model is updated in real time using the following formula: θ k (new) = δη k +θ k (bef), where k = 1, 2, 3, ..., K, K is the total number of parameter values in the electromagnetic coil array parameter adjustment model, θ k (new) represents the k-th parameter value in the updated electromagnetic coil array parameter adjustment model, δ is the update rate, and η is the parameter value. k To adjust the value of the k-th parameter in the updated electromagnetic coil array parameter adjustment model, θ k (bef) The kth parameter value in the target electromagnetic coil array parameter adjustment model before the update.
2. The SMT production process quality control method according to claim 1, characterized in that, Training the solder paste segmentation model involves the following steps: Several solder paste segmentation training samples are obtained, including PCB thermal images. The solder paste distribution heatmaps are used to annotate the training samples. All annotated training samples are combined into a solder paste segmentation training set. The solder paste segmentation model is trained using the training set. The accuracy of the solder paste segmentation model is then determined to be higher than a set threshold. If the accuracy of the solder paste segmentation model is higher than the set threshold, the trained solder paste segmentation model is output. Otherwise, the solder paste segmentation model is trained again using the training set.
3. The SMT production process quality control method according to claim 2, characterized in that, Training the electromagnetic coil array parameter adjustment model involves the following steps: Several training samples for adjusting electromagnetic coil array parameters are obtained, including solder paste distribution heatmaps. These training samples are then labeled using an electromagnetic coil array parameter adjustment set. All labeled training samples are combined into a training set. The electromagnetic coil array parameter adjustment model is trained using this training set. The accuracy of the model is then determined to be higher than a set threshold. If the accuracy is higher than the threshold, the trained model is output; otherwise, the model is trained again using the training set.
4. The SMT production process quality control method according to claim 3, characterized in that, Training the policy adjustment evaluation network involves the following steps: Obtain several training samples for adjusting strategy evaluation, which include pre-constructed adjusting strategy evaluation data. Label the training samples using the adjusting strategy evaluation values. Combine all labeled training samples into an adjusting strategy evaluation training set. Train the adjusting strategy evaluation network using the training set. Determine if the accuracy of the adjusting strategy evaluation network is higher than a corresponding set threshold. If the accuracy is higher than the set threshold, output the trained adjusting strategy evaluation network; otherwise, continue training the adjusting strategy evaluation network using the training set.
5. A quality control system for SMT production process, characterized in that, The system applies a quality control method for SMT production process according to any one of claims 1-4, comprising: The real-time solder paste distribution acquisition module is used to deploy an electromagnetic coil array around the reflow oven, acquire the PCB thermal image of the PCB board in real time through a thermal imager deployed in the reflow oven, and then send the PCB thermal image into the solder paste segmentation model for processing to output a solder paste distribution thermal map. The electromagnetic coil array parameter adjustment module is used to send the solder paste distribution heat map into the electromagnetic coil array parameter adjustment model for processing and output the electromagnetic coil array parameter adjustment set. The electromagnetic coil array parameter adjustment set includes N electromagnetic coil parameters, where N is the total number of electromagnetic coils in the electromagnetic coil array. The electromagnetic coil array parameters include the position of the electromagnetic coils and the electromagnetic coil control parameters. The reinforcement learning module is used to perform reinforcement learning operations on the electromagnetic coil array parameter adjustment model based on feedback from changes in the solder paste distribution heatmap.