A laser ball welding control system, method and medium

Abstract

The application relates to a laser ball planting welding control system, a method and a medium, and relates to the technical field of battery integrated busbar assembly welding. The laser ball planting welding control system comprises a tin ball monitoring module, a laser control module, a temperature monitoring module, a pressure regulating and spraying module and a visual detection module, which are bidirectionally connected with a central control module, are used for collecting and detecting welding point positions, tin ball positions, tin material spraying point positions and welding point forming states according to image detection instructions in combination with a preset ball planting welding model, and generating welding detection results; the central control module is used for receiving real-time data of the monitoring modules, sending control instructions to the execution modules, realizing action time sequence linkage of the modules, dynamic parameter adjustment and welding abnormality alarm; through precise control of laser ball planting welding, the tensile strength and the anti-peeling performance of each branch of the busbar are improved, and the performance stability of the integrated busbar when collecting automobile battery information is enhanced.

Description

Technical Field

[0001] This application relates to the field of battery integrated busbar assembly and welding technology, and in particular to a laser ball-planting welding control system, method and medium. Background Technology

[0002] Currently, most mainstream integrated busbar (CCS) solutions use flexible printed circuit boards (FPCs) or flexible flat cables (FFCs) combined with die-cut circuits (FDCs) to acquire voltage and temperature signals. The acquisition branches are connected to conductive aluminum busbars by welding to form an electrical circuit.

[0003] Currently, laser soldering is mainly used to weld and fix the acquisition branches. During the welding process, solder balls are first transported to the welding station in batches by vibration feeding. The laser module emits laser at a fixed power and frequency to melt the solder balls, and the gas jet module sprays nitrogen at a constant pressure to push the molten solder to the contact point between the integrated busbar acquisition branch and the aluminum busbar to complete the welding.

[0004] Existing patents disclose an intelligent adaptive laser ball-mounting system, device, and method. The intelligent adaptive laser ball-mounting system includes a multi-source laser module, a dual-vision positioning module, a dynamic air pressure ball supply module, a real-time monitoring and feedback module, and an adaptive control module that are electrically connected to each other. This invention improves positioning accuracy through the dual-vision positioning module, reduces the impact of thermal stress through dual-source collaboration, reduces the aluminum content caused by ball blockage through the dynamic air pressure ball supply module, and improves ball-mounting efficiency and stability. The adaptive control module enables automatic parameter matching and real-time calibration, eliminating the need for frequent manual adjustments. It is compatible with ball-mounting in ultra-fine pitch scenarios, adapts to various substrate materials, and meets the requirements of 3D packaging, HBM stacking, and Chiplet integration. It achieves high-precision, low-damage, high-yield, and high-compatibility ball-mounting operations and has high promotional value.

[0005] The existing technical solutions mentioned above have the following defects: 1. The welding process of the existing dry cell integrated busbar assembly lacks process monitoring and dynamic parameter control, which easily leads to problems such as insufficient melting of solder balls or laser burns on the aluminum busbar substrate, resulting in a decrease in the conductivity of the aluminum busbar and insufficient tensile strength of the collecting branches. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the purpose of this application is to provide a laser ball bonding control system, method, and medium. By precisely controlling the laser ball bonding, the tensile strength and anti-peeling properties of each acquisition branch are effectively improved, and the performance stability of the integrated busbar when acquiring automotive battery information is enhanced.

[0007] This was achieved using the following technical solutions: In a first aspect, this application provides a laser ball bonding control system, comprising: The solder ball monitoring module is bidirectionally connected to the central control module and to the laser control module. It is used to monitor the remaining amount of solder balls in the loading container, obtain the solder ball conveying status, and sort and transfer the solder balls in the loading container to the target position according to the solder ball distribution instructions, and detect the solder ball arrival status. The laser control module is bidirectionally connected to the central control module and also connected to the temperature monitoring module. It is used to adjust the laser pulse parameters and detect the laser output status according to the laser output command. The temperature monitoring module is bidirectionally connected to the central control module and also connected to the pressure-regulating injection module. It is used to monitor the temperature changes during the solder ball melting process according to the temperature sensing command and to collect the solder ball temperature and solder joint temperature. The pressure-regulating injection module is bidirectionally connected to the central control module and is used to correct the injection pressure and injection timing of the inert gas according to the welding control command and the gas pressure control parameters. The vision inspection module is bidirectionally connected to the central control module. It is used to collect and inspect the position of solder joints, the position of solder balls, the position of solder spraying points and the solder joint forming state according to the image inspection instructions and the preset ball-mounting welding model, and generate welding inspection results. The central control module is used to receive real-time data from the monitoring modules and send control commands to the execution modules to realize the linkage of action sequence of each module, dynamic adjustment of parameters, and welding abnormality alarm.

[0008] By adopting the above technical solution, multi-source sensing data is integrated through the central control module, and PID temperature control, visual feedback and time-series collaborative algorithms are used to dynamically adjust laser output, jet pressure and solder ball sorting, realizing closed-loop control of the entire ball bonding process, which has the advantages of high precision, adaptive adjustment and real-time abnormal early warning.

[0009] Furthermore, the solder ball monitoring module includes: The material level monitoring unit is used to scan the solder balls in the loading container, obtain the solder ball specification parameters, and count the remaining solder ball quantity and solder ball conveying status. The solder ball sorting unit is used to transfer solder balls to the target position according to the solder ball specifications and the solder ball conveying status, and to determine the solder ball arrival status.

[0010] By adopting the above technical solution, the specifications and remaining status of solder balls are obtained in real time through machine vision and image segmentation algorithms. Based on rule matching or decision tree algorithms, the solder balls are sorted and transferred to the target station, realizing accurate monitoring and efficient sorting of solder ball supply, and significantly improving the stability and automation level of the solder ball loading process.

[0011] Furthermore, the laser control module includes: The parameter matching unit is used to match the solder ball specifications and solder joint materials according to the preset laser configuration library to obtain the laser target parameters; The laser adjustment unit is used to adjust the parameters of the laser pulse according to the laser target parameters and the laser output command, and output the target laser. The laser detection unit is used to detect the state of the target laser, determine the laser output state, and generate state feedback parameters.

[0012] By adopting the above technical solution, the specifications of solder balls and the material of solder joints are dynamically matched through parameter matching algorithms to set laser target parameters. The laser output status is monitored in real time using a closed-loop feedback mechanism, thereby achieving precise control and adaptive optimization of welding energy and improving the consistency of welding quality and process stability.

[0013] Furthermore, the visual inspection module includes: The analysis and decomposition unit is used to analyze the laser ball bonding process, determine the visual inspection nodes, and obtain the corresponding task execution identifiers; The visual invocation unit is used to generate a task switching timestamp based on the visual inspection nodes and the welding task sequence. The solder ball positioning unit is used to acquire images of the solder ball spraying position according to the first task execution identifier, obtain a spraying point position map, and mark the spraying point position with a label; The substrate positioning unit is used to acquire images of the position to be welded according to the second task execution identifier, obtain a substrate position map, and mark the substrate inspection label. The weld point inspection unit is used to acquire images of the finished welded product according to the third task execution identifier, obtain a finished weld point image, and mark the weld point inspection label. The image cleaning unit is used to filter and denoise the spray point location map, the substrate location map to be welded, and the finished weld point map, and to perform time-series correlation with the corresponding tags to construct a ball-planting welding image tuple; The parameter matching unit is used to perform modal coordination on the preset ball-planting welding model according to the task switching timestamp and match the corresponding modal-specific parameters. Anomaly identification unit is used to identify the ball-planting welding element group based on modal-specific parameters and corresponding status labels, determine abnormal nodes, and calculate abnormal deviation values. The anomaly control unit is used to determine the anomaly correction strategy based on the type of execution device and the anomaly node, and to construct the deviation correction sequence based on the anomaly deviation value. The sequence classification unit is used to classify the deviation correction sequence pairs according to the status label to obtain the stage adjustment sequence, and combine it with the task execution identifier to generate the welding inspection result.

[0014] By adopting the above technical solution, through time-series scheduling and multi-node image acquisition strategies, combined with modal matching and anomaly detection algorithms, primitive association and deviation quantification are performed on the spraying points, substrates and finished welds, and stage adjustment sequences are generated according to equipment type, so as to realize precise visual monitoring and adaptive control of the entire laser ball bonding process, significantly improving welding yield and process stability.

[0015] Furthermore, the pre-defined ball-mounting welding model includes: The timing analysis layer is used to perform timing segmentation of the welding backbone network according to the laser ball bonding process, resulting in a solder ball determination branch, a substrate discrimination branch, and a solder joint discrimination branch. The feature extraction layer is used to perform feature analysis and edge detection on historical ball-mounted welding images, extracting features of spraying points, substrates to be welded, and finished weld points. The collaborative training layer is used to iteratively train the features of the injection point based on the preset single-branch training parameters and the solder ball determination branch, and generate the injection point positioning weight matrix. The substrate discrimination branch is combined with the features of the substrate to be welded for iterative training to generate the substrate localization weight matrix; Simultaneously, the features of finished solder joints are iteratively trained by combining the solder joint discrimination branch to generate a weight matrix of formed solder joints. Based on preset collaborative training parameters and a spatiotemporal attention mechanism, any two different weight matrices are jointly trained and spliced ​​to generate a global welding weight matrix. The time-discontinuity segmentation layer is used to segment the global welding weight matrix according to the task switching timestamp, forming a local time-series weight matrix; The identification optimization layer is used to identify and detect historical ball-planting welding images based on the local time-series weight matrix, calculate stage index parameters, and compare them with preset stage index thresholds. If all the stage indicator parameters are greater than the stage indicator threshold, it indicates that the current ball-planting welding model is the optimal welding model. If not, then the difference between the stage indicator parameter and the stage indicator threshold is calculated to obtain the stage indicator difference, and the global welding weight matrix is ​​corrected to obtain the optimal welding weight matrix. The modal interaction layer is used to perform modal transformation on the optimal welding weight matrix according to the stage task objectives and output the modal-specific parameters for each stage.

[0016] By adopting the above technical solution, through multi-branch collaborative training and spatiotemporal attention mechanism, the features of spraying point, substrate and weld point are jointly iterated to generate a global weight matrix and optimize it in segments, so as to realize high-precision modeling and dynamic modal adaptation of the welding process, and significantly improve the generalization ability and process adaptability of the ball-planting welding model.

[0017] Furthermore, the central control module includes: The data receiving unit is used to receive and store tin monitoring data, laser control parameters, welding temperature parameters, pressure-adjusted spraying parameters and welding detection results based on the laser ball bonding process, and to build a welding timing parameter library; The instruction conversion unit is used to convert and distribute the welding timing parameter library according to the equipment protocol type, and generate equipment execution instructions; The coordinated control unit is used to coordinate the timing of each module according to the equipment execution command and modal-specific parameters, generate welding abnormality alarms, and complete the timing linkage of actions between modules.

[0018] By adopting the above technical solutions, a timing parameter library is constructed through multi-source data fusion and protocol conversion algorithms. Combined with modal-specific parameters, inter-module collaborative scheduling and abnormal linkage are realized, which significantly improves the automation level and process stability of the laser ball bonding process.

[0019] Secondly, this application also provides a laser ball bonding control method, which adopts the following technical solution; A laser ball bonding control method includes: Monitor the remaining amount of solder balls in the loading container, obtain the solder ball conveying status, and sort and transfer them to the target position according to the solder ball distribution instructions, and detect the solder ball arrival status; Adjust the laser pulse parameters according to the laser output command, and detect the laser output status; Monitor temperature changes during the solder ball melting process according to temperature sensing instructions, and collect solder ball temperature and solder joint temperature; Based on the welding control command and gas pressure control parameters, the injection pressure and injection timing of the inert gas are corrected. Based on the image detection instructions and the ball-mounting welding model, the solder joint position, solder ball position, solder spray point and solder joint forming state are collected and detected to generate welding inspection results. It receives and analyzes solder ball delivery status, laser output status, solder ball temperature, solder joint temperature, spray pressure, spray timing, and welding inspection results, and generates and distributes laser output commands, temperature sensing commands, welding control commands, and image inspection commands.

[0020] By adopting the above technical solutions, multi-sensor fusion and closed-loop PID control algorithms are used to monitor solder ball balance, laser status, temperature and spraying parameters in real time. A visual recognition model is used to detect solder joint formation, dynamically generate and distribute control commands, and achieve adaptive coordination and precise control of the entire welding process, which significantly improves process stability and yield.

[0021] Thirdly, this application also provides a storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the laser ball bonding control method as described above.

[0022] In summary, the beneficial technical effects of this application are as follows: By using multi-sensor fusion and closed-loop PID control algorithm, the solder ball balance, laser status, temperature and spraying parameters are monitored in real time. The visual recognition model is used to detect solder joint formation, dynamically generate and distribute control commands, realize adaptive coordination and precise control of the entire welding process, and significantly improve process stability and yield. By using multi-branch collaborative training and spatiotemporal attention mechanism, the features of spraying points, substrate and weld points are jointly iterated to generate a global weight matrix and optimize it in segments, thereby achieving high-precision modeling and dynamic modal adaptation of the welding process, which significantly improves the generalization ability and process adaptability of the ball-planting welding model. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the laser ball bonding control system in this application; Figure 2 This is a schematic diagram of the visual inspection module architecture in this application; Figure 3 This is a schematic diagram of the ball-planting welding model architecture in this application; Figure 4 This is a schematic diagram of the laser ball bonding control method in this application; Figure 5 This is a flowchart of the laser ball bonding process in this application; Figure 6 This is a schematic diagram of the welding positions in this application. Detailed Implementation

[0024] The present application will be further described in detail below with reference to the accompanying drawings.

[0025] Reference Figure 1 The laser ball bonding control system disclosed in this application includes: The solder ball monitoring module is bidirectionally connected to the central control module and to the laser control module. It is used to monitor the remaining amount of solder balls in the loading container, obtain the solder ball conveying status, and sort and transfer the solder balls in the loading container to the target position according to the solder ball distribution instructions, and detect the solder ball arrival status. The laser control module is bidirectionally connected to the central control module and also connected to the temperature monitoring module. It is used to adjust the laser pulse parameters and detect the laser output status according to the laser output command. The temperature monitoring module is bidirectionally connected to the central control module and also connected to the pressure-regulating injection module. It is used to monitor the temperature changes during the solder ball melting process according to the temperature sensing command and to collect the solder ball temperature and solder joint temperature. The pressure-regulating injection module is bidirectionally connected to the central control module and is used to correct the injection pressure and injection timing of the inert gas according to the welding control command and the gas pressure control parameters. The vision inspection module is bidirectionally connected to the central control module. It is used to collect and inspect the position of solder joints, the position of solder balls, the position of solder spraying points and the solder joint forming state according to the image inspection instructions and the preset ball-mounting welding model, and generate welding inspection results. The central control module is used to receive real-time data from the monitoring modules and send control commands to the execution modules to realize the linkage of action sequence of each module, dynamic adjustment of parameters, and welding abnormality alarm.

[0026] In this embodiment, the central control module, the core control unit of the system, communicates bidirectionally with all other modules, receives real-time data from each monitoring module, sends control commands to each execution module, realizes the linkage of action sequence of each module, dynamic adjustment of parameters and alarm of abnormal situation, and has a built-in parameter database to store the optimal configuration of laser parameters, nitrogen parameters and other parameters corresponding to different solder ball specifications and different solder joint materials.

[0027] The solder ball monitoring module consists of a level sensor, a solder ball sorting unit, and a data transmission unit. It is used to monitor the remaining amount of solder balls in the solder ball container and the conveying status of solder balls at the outlet in real time. The monitoring data is transmitted to the central control module. When the remaining amount of solder balls is lower than the preset threshold or when a ball is stuck at the outlet, an abnormal signal is sent to the central control module and an alarm is triggered.

[0028] The solder ball sorting unit receives instructions from the central control module to accurately sort individual solder balls in the solder ball container, so that each individual solder ball is accurately transferred to the preset position of the spray head, and the position sensor detects the position status of the solder ball in real time and feeds it back to the central control module.

[0029] The laser control module receives instructions from the central control module, precisely adjusts the laser pulse power, pulse width, and pulse frequency according to the solder ball specifications and solder joint material, and detects the laser output status in real time and feeds it back to the central control module to achieve instantaneous and precise melting of the solder ball while avoiding burning the solder joint substrate. The pressure-regulating injection module receives instructions from the central control module. At the moment the solder ball melts, it controls the pressure regulating valve and flow meter to precisely regulate the injection pressure and flow rate of nitrogen. It also controls the timing of nitrogen injection through the injection valve. Using inert gas pressure, the molten solder is precisely injected onto the surface of the solder joint. At the same time, nitrogen acts as an anti-oxidant to ensure the quality of the solder joint.

[0030] The vision inspection module is used for solder joint pre-positioning and solder joint formation inspection. First, it captures an image of the solder joint position and performs image processing, transmitting the precise coordinates of the solder joint to the central control module, providing a basis for the central control module to adjust the position of the spray head. Second, it detects whether the position of the solder ball at the spray head is accurate. Then, after the solder joint is formed, it captures an image of the solder joint after soldering to determine whether the solder spraying point is accurate and whether the solder joint formation is good, and feeds back the inspection results to the central control module.

[0031] The temperature monitoring module is used to monitor the temperature changes during the solder ball melting process in real time, as well as the temperature changes and cooling process of the solder joint after the solder is sprayed onto the surface of the solder joint. The temperature data is transmitted to the central control module, which can dynamically adjust the laser parameters according to the temperature data to ensure that the solder ball is in the optimal melting temperature range, while avoiding damage to the substrate caused by excessively high solder joint temperature.

[0032] Preferably, the solder ball monitoring module includes: The material level monitoring unit is used to scan the solder balls in the loading container, obtain the solder ball specification parameters, and count the remaining solder ball quantity and solder ball conveying status. The solder ball sorting unit is used to transfer solder balls to the target position according to the solder ball specifications and the solder ball conveying status, and to determine the solder ball arrival status.

[0033] In this embodiment, during the wafer-level micro solder ball soldering process in a semiconductor packaging workshop, the material level monitoring unit scans the solder balls in the loading container at a frequency of 100 times per second using a laser displacement sensor to obtain specifications such as solder ball diameter (e.g., 0.3mm), roundness, and surface defects in real time, and also counts the remaining solder balls and the conveying status such as jamming and shortage in the conveying pipeline. The solder ball sorting unit selects qualified solder balls according to the specifications and, combined with the arrival signal in the conveying status, transfers the solder balls one by one to the target pad position on the substrate through a combination of vacuum suction and electromagnetic vibration. At the same time, the fiber optic sensor confirms that the solder balls have been accurately placed, thus determining the solder ball arrival status.

[0034] Preferably, the laser control module includes: The parameter matching unit is used to match the solder ball specifications and solder joint materials according to the preset laser configuration library to obtain the laser target parameters; The laser adjustment unit is used to adjust the parameters of the laser pulse according to the laser target parameters and the laser output command, and output the target laser. The laser detection unit is used to detect the state of the target laser, determine the laser output state, and generate state feedback parameters.

[0035] In this embodiment, the parameter matching unit of the laser control module first performs rapid matching of the current solder ball specifications (e.g., diameter 0.25mm, tin content 96.5%) and pad material (e.g., copper-based gold plating) according to the preset laser configuration library, and outputs the laser target parameters (wavelength 1064nm, pulse width 2.5ms, peak power 120W). The laser adjustment unit, based on the target parameters and the laser output command issued by the host computer, precisely controls the waveform and energy of the laser pulse by adjusting the acousto-optic modulator and drive current, and outputs a stable target laser. The laser detection unit uses the built-in photodiode and energy meter to monitor the peak power and pulse width of the target laser in real time, determines the laser output status (e.g., normal, weak, or overshoot), and generates status feedback parameters containing the actual energy value and deviation amount, which are sent back to the control system for dynamic closed-loop correction of the laser parameters for the next cycle, ensuring that each solder ball is heated evenly and melted fully, and achieving highly reliable spatter-free soldering.

[0036] Reference Figure 2 Preferably, the visual inspection module includes: The analysis and decomposition unit is used to analyze the laser ball bonding process, determine the visual inspection nodes, and obtain the corresponding task execution identifiers; The visual invocation unit is used to generate a task switching timestamp based on the visual inspection nodes and the welding task sequence. The solder ball positioning unit is used to acquire images of the solder ball spraying position according to the first task execution identifier, obtain a spraying point position map, and mark the spraying point position with a label; The substrate positioning unit is used to acquire images of the position to be welded according to the second task execution identifier, obtain a substrate position map, and mark the substrate inspection label. The weld point inspection unit is used to acquire images of the finished welded product according to the third task execution identifier, obtain a finished weld point image, and mark the weld point inspection label. The image cleaning unit is used to filter and denoise the spray point location map, the substrate location map to be welded, and the finished weld point map, and to perform time-series correlation with the corresponding tags to construct a ball-planting welding image tuple; The parameter matching unit is used to perform modal coordination on the preset ball-planting welding model according to the task switching timestamp and match the corresponding modal-specific parameters. Anomaly identification unit is used to identify the ball-planting welding element group based on modal-specific parameters and corresponding status labels, determine abnormal nodes, and calculate abnormal deviation values. The anomaly control unit is used to determine the anomaly correction strategy based on the type of execution device and the anomaly node, and to construct the deviation correction sequence based on the anomaly deviation value. The sequence classification unit is used to classify the deviation correction sequence pairs according to the status label to obtain the stage adjustment sequence, and combine it with the task execution identifier to generate the welding inspection result.

[0037] In this embodiment, on a wafer-level micro solder ball placement production line in an advanced semiconductor packaging plant, the vision inspection module analyzes and determines three key inspection nodes according to the laser solder ball placement process: before solder ball spraying, during substrate positioning and alignment, and after soldering, corresponding to task execution identifiers "S1", "S2", and "S3" respectively. When the soldering task progresses to the first node, the vision invocation unit generates a task switching timestamp T1, and the solder ball positioning unit immediately triggers a high-magnification industrial camera to acquire images of the solder ball spraying position, obtains a spraying point location map, and marks the spraying point location with a label; subsequently, at time T2, the substrate positioning unit acquires images of the copper pads on the substrate to be soldered, obtains a substrate location map, and marks the substrate with an inspection label; after soldering is completed, at time T3, the solder joint inspection unit acquires a finished solder joint image of the finished product area and marks the solder joint with an inspection label. After the image cleaning unit performs median filtering to remove noise from the three images, it constructs a solder ball placement welding tuple according to the time sequence association of the timestamp and the label. The parameter matching unit matches the modal-specific parameters for the three stages of spray positioning, substrate alignment, and weld point formation from the preset ball-planting welding model library based on the task switching timestamps T1, T2, and T3.

[0038] The anomaly identification unit uses these parameters to perform feature comparisons on the primitives at each stage: identifying solder ball offset (e.g., deviation of 0.02mm) in the spray point location map, detecting pad oxidation or contamination (grayscale anomaly value > 10%) in the substrate map, and determining bridging or cold solder joints (solder joint area less than 80% of the standard) in the finished solder joint map, and calculating the abnormal deviation value. The anomaly control unit, based on the type of execution equipment (e.g., sprayer, motion platform, laser), formulates a "adjusting the XY coordinates of the spray head" correction strategy for "spray offset" anomalies, constructing a deviation correction sequence including the stepper motor pulse count based on the 0.02mm deviation value; for "cold solder joint" anomalies, it adjusts the laser power and pulse width.

[0039] The sequence classification unit categorizes the deviation correction sequence into stage adjustment sequences according to the status labels (before spraying, during welding, and after welding), and associates them with task execution identifiers. Finally, it generates a welding inspection result report containing the anomaly type, location coordinates, and recommended correction parameters, which is fed back to the production line control system in real time to adjust subsequent operation parameters, ensuring that the ball placement yield of millions of solder balls is controlled at over 99.95%.

[0040] Reference Figure 3 Preferably, the preset ball-planting welding model includes: The timing analysis layer is used to perform timing segmentation of the welding backbone network according to the laser ball bonding process, resulting in a solder ball determination branch, a substrate discrimination branch, and a solder joint discrimination branch. The feature extraction layer is used to perform feature analysis and edge detection on historical ball-mounted welding images, extracting features of spraying points, substrates to be welded, and finished weld points. The collaborative training layer is used to iteratively train the features of the injection point based on the preset single-branch training parameters and the solder ball determination branch, and generate the injection point positioning weight matrix. The substrate discrimination branch is combined with the features of the substrate to be welded for iterative training to generate the substrate localization weight matrix; Simultaneously, the features of finished solder joints are iteratively trained by combining the solder joint discrimination branch to generate a weight matrix of formed solder joints. Based on preset collaborative training parameters and a spatiotemporal attention mechanism, any two different weight matrices are jointly trained and spliced ​​to generate a global welding weight matrix. The time-discontinuity segmentation layer is used to segment the global welding weight matrix according to the task switching timestamp, forming a local time-series weight matrix; The identification optimization layer is used to identify and detect historical ball-planting welding images based on the local time-series weight matrix, calculate stage index parameters, and compare them with preset stage index thresholds. If all the stage indicator parameters are greater than the stage indicator threshold, it indicates that the current ball-planting welding model is the optimal welding model. If not, then the difference between the stage indicator parameter and the stage indicator threshold is calculated to obtain the stage indicator difference, and the global welding weight matrix is ​​corrected to obtain the optimal welding weight matrix. The modal interaction layer is used to perform modal transformation on the optimal welding weight matrix according to the stage task objectives and output the modal-specific parameters for each stage.

[0041] In this embodiment, the ball-mounting welding model first divides the welding backbone network into three parallel sub-networks according to the time axis based on the laser ball-mounting welding process through the temporal analysis layer: the solder ball determination branch (before solder ball spraying), the substrate discrimination branch (when the pad is aligned), and the solder joint discrimination branch (after welding). The feature extraction layer performs Canny edge detection and SIFT feature analysis on 2000 sets of ball-mounting welding images accumulated in history, and extracts the solder ball spraying point position offset features, substrate pad oxidation grayscale features, and bridging and cold solder joint morphology features of finished solder joints.

[0042] The collaborative training layer uses preset single-branch training parameters to perform 200 rounds of iterative training on the solder ball judgment branch and the spray point features to generate a spray point positioning weight matrix (positioning error ≤ 0.01mm), 150 rounds of iterative training on the substrate discrimination branch and the substrate to be soldered features to generate a substrate positioning weight matrix (alignment deviation ≤ 0.005mm), and 300 rounds of iterative training on the solder joint discrimination branch and the finished solder joint features to generate a formed solder joint weight matrix (defect recognition rate ≥ 99.5%). Subsequently, based on the collaborative training parameters (learning rate 0.001, batch size 32), a spatiotemporal attention mechanism is introduced to perform feature joint training and matrix concatenation on each pair of the three weight matrices to generate a global soldering weight matrix. The time-discontinuity segmentation layer divides the global matrix into three local temporal weight matrices according to the task switching timestamps in actual production (e.g., T1=0ms, T2=120ms, T3=450ms), corresponding to the spray positioning, substrate alignment, and solder joint inspection stages, respectively.

[0043] The recognition optimization layer uses a local temporal weight matrix to perform inference detection on a batch of new ball-mounted welding images, calculating stage index parameters (such as spray point positioning accuracy of 98.5%, substrate recognition accuracy of 99.2%, and weld defect detection rate of 99.8%), and comparing them with preset stage index thresholds (all must be ≥99.0%). It was found that the spray point positioning accuracy was lower than the threshold. The convolutional layer parameters related to spray positioning in the global welding weight matrix were fine-tuned by calculating the stage index difference (-0.5%). After 5 iterations, the three indicators reached 99.3%, 99.1%, and 99.8% respectively, meeting the threshold requirements and obtaining the optimal result. The optimal welding weight matrix is ​​optimized. Finally, the modal interaction layer performs modal transformation on the optimal weight matrix according to the task objectives of each stage (such as high-precision coordinates in the spraying stage, grayscale thresholds in the alignment stage, and defect classification in the post-weld stage). It outputs spraying point positioning modal-specific parameters (feature map size 256×256, non-maximum suppression threshold 0.45), substrate alignment modal-specific parameters (template matching similarity 0.85, grayscale difference threshold 8%), and solder joint inspection modal-specific parameters (cross-over ratio threshold 0.75, confidence level 0.9). These parameters are available for real-time use by the online visual inspection module to achieve highly reliable ball placement and soldering process control for tens of thousands of solder balls on each GPU substrate.

[0044] Preferably, the central control module includes: The data receiving unit is used to receive and store tin monitoring data, laser control parameters, welding temperature parameters, pressure-adjusted spraying parameters and welding detection results based on the laser ball bonding process, and to build a welding timing parameter library; The instruction conversion unit is used to convert and distribute the welding timing parameter library according to the equipment protocol type, and generate equipment execution instructions; The coordinated control unit is used to coordinate the timing of each module according to the equipment execution command and modal-specific parameters, generate welding abnormality alarms, and complete the timing linkage of actions between modules.

[0045] In this embodiment, the central control module, as the scheduling core of the entire process, first uses the data receiving unit to follow the laser ball bonding process (refer to...). Figure 5 The timing requirements are met, and the system receives and stores in real time the solder ball balance and position status from the solder ball monitoring module, the current pulse parameters and feedback status from the laser control module, the welding temperature parameters collected by the infrared thermal imager, the air pressure and flow data from the pressure-regulating spray module, and the welding inspection results (including spray offset, pad alignment deviation and solder joint defect type) output by the visual inspection module. All data are indexed by millisecond-level timestamps to build a welding timing parameter library.

[0046] Subsequently, the instruction conversion unit converts key control quantities in the parameter library according to fieldbus protocols (such as EtherCAT): converting the "solder ball in place" signal into an opening command for the injector solenoid valve, converting the "laser target parameters" into a pulse width modulation command for the laser driver, and converting the "substrate positioning deviation" into an XY compensation command for the motion platform, and distributing them to the corresponding execution devices via industrial Ethernet. The collaborative control unit uses modal-specific parameters (such as injection positioning needing to be completed at T0+0ms, laser welding needing to be triggered at T0+120ms, and post-weld inspection needing to be executed at T0+450ms) as a benchmark to coordinate the timing of the actions of each module: when the solder ball monitoring module reports a shortage of material, the collaborative control unit immediately generates a "pause feed" command and triggers a welding abnormality alarm, while simultaneously notifying the vision inspection module to skip the current solder joint; when the welding temperature parameter exceeds the threshold, the collaborative control unit dynamically adjusts the laser pulse width and simultaneously extends the cooling waiting time, ensuring strict linkage between the timing of actions between modules, ultimately achieving a fully closed-loop control of the highly reliable ball placement and welding process for tens of thousands of solder balls on each substrate.

[0047] Reference Figure 4 This application discloses a laser ball bonding control method, comprising: S1: Monitor the remaining amount of solder balls in the loading container, obtain the solder ball conveying status, and sort and transfer them to the target position according to the solder ball distribution instruction, and detect the solder ball arrival status; S2: Adjust the laser pulse parameters according to the laser output command and detect the laser output status; S3: Monitor the temperature change during the melting process of the solder ball according to the temperature sensing command, and collect the temperature of the solder ball and the solder joint. S4: Based on the welding control command and gas pressure control parameters, correct the inert gas injection pressure and injection timing; S5: Based on the image detection instructions and the ball-mounting welding model, collect and detect the solder joint position, solder ball position, solder spray point and solder joint forming state, and generate welding inspection results; S6: Receives and analyzes solder ball delivery status, laser output status, solder ball temperature, solder joint temperature, spray pressure, spray timing, and welding inspection results; generates and distributes laser output commands, temperature sensing commands, welding control commands, and image inspection commands.

[0048] In this embodiment: the laser ball bonding control system is started, the central control module calls the built-in parameter database, and the operator inputs relevant parameters into the central control module according to the specifications (diameter, material) of the solder ball to be soldered and the substrate material of the solder joint. The central control module generates the initial laser parameters, nitrogen parameters and solder ball delivery parameters.

[0049] The solder ball monitoring module detects the remaining amount of solder balls in the solder ball container in real time. If the remaining amount is higher than the preset threshold, proceed to the next step; if the remaining amount is lower than the preset threshold, the central control module triggers a material shortage alarm, prompting the operator to replenish the material.

[0050] The central control module sends a conveying command to the solder ball separation control module, transferring a single solder ball to the preset position of the spray head (see reference). Figure 6 Once the position sensor detects that the solder ball is in place, it sends a signal back to the central control module.

[0051] The industrial camera in the vision positioning and detection module captures images of the weld points at the welding station. The image processing unit identifies and calibrates the images, obtains the precise coordinates of the weld points, and transmits them to the central control module. The central control module then fine-tunes the position of the spray head based on the coordinates to ensure that the spray head is aligned with the weld point.

[0052] Based on the initial parameters or feedback data from the previous welding, the central control module sends parameter adjustment commands to the laser control module and the nitrogen pressure regulating injection module. The laser control module adjusts to the preset pulse power, pulse width, and pulse frequency, while the nitrogen pressure regulating injection module adjusts to the preset injection pressure and flow rate, waiting for the melting command.

[0053] The central control module sends a laser emission command to the laser control module, and the laser generator emits a laser to instantly melt the solder ball at the nozzle. At the same time, the central control module sends an injection command to the nitrogen pressure regulating injection module, the injection valve opens, and nitrogen is ejected at a preset pressure and flow rate. This achieves millisecond-level timing coordination with the solder ball melting action, accurately spraying the molten solder onto the surface of the solder joint.

[0054] The temperature monitoring module monitors the temperature at the melting point of the solder ball and the surface temperature of the solder joint in real time, and transmits the temperature data to the central control module. If the temperature deviates from the preset optimal range, the central control module immediately and dynamically adjusts the parameters of the laser control module to ensure that the solder ball is fully melted and the solder joint substrate is not burned.

[0055] The industrial camera in the vision positioning and detection module captures images of the solder joints after welding. The image processing unit analyzes and judges the position and forming state of the solder joints. If the solder joints are well formed and the position is accurate, the welding is completed. If there are problems such as position deviation or poor solder forming, the central control module marks the solder joint as a defective product and adjusts the parameters of the next welding (such as the position of the nozzle and the nitrogen pressure) according to the detection data.

[0056] After a single soldering operation is completed, the central control module sends the next solder ball delivery instruction to the solder ball separation control module to achieve continuous automated soldering. If situations such as ball jamming, laser failure, or abnormal nitrogen pressure occur during the soldering process, the central control module will immediately trigger a shutdown alarm and record the abnormal information.

[0057] This application discloses a storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the laser ball bonding control method as described above.

[0058] The embodiments described in this specific implementation are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A laser ball bonding control system, characterized in that, include: The solder ball monitoring module is bidirectionally connected to the central control module and to the laser control module. It is used to monitor the remaining amount of solder balls in the loading container, obtain the solder ball conveying status, and sort and transfer the solder balls in the loading container to the target position according to the solder ball distribution instructions, and detect the solder ball arrival status. The laser control module is bidirectionally connected to the central control module and also connected to the temperature monitoring module. It is used to adjust the laser pulse parameters and detect the laser output status according to the laser output command. The temperature monitoring module is bidirectionally connected to the central control module and also connected to the pressure-regulating injection module. It is used to monitor the temperature changes during the solder ball melting process according to the temperature sensing command and to collect the solder ball temperature and solder joint temperature. The pressure-regulating injection module is bidirectionally connected to the central control module and is used to correct the injection pressure and injection timing of the inert gas according to the welding control command and the gas pressure control parameters. The vision inspection module is bidirectionally connected to the central control module. It is used to collect and inspect the position of solder joints, the position of solder balls, the position of solder spraying points and the solder joint forming state according to the image inspection instructions and the preset ball-mounting welding model, and generate welding inspection results. The central control module is used to receive real-time data from the monitoring modules and send control commands to the execution modules to realize the linkage of action sequence of each module, dynamic adjustment of parameters, and welding abnormality alarm.

2. The laser ball bonding control system according to claim 1, characterized in that, The solder ball monitoring module includes: The material level monitoring unit is used to scan the solder balls in the loading container, obtain the solder ball specification parameters, and count the remaining solder ball quantity and solder ball conveying status. The solder ball sorting unit is used to transfer solder balls to the target position according to the solder ball specification parameters and the solder ball conveying status, and to determine the solder ball arrival status.

3. The laser ball bonding control system according to claim 1, characterized in that, The laser control module includes: The parameter matching unit is used to match the solder ball specifications and solder joint materials according to the preset laser configuration library to obtain the laser target parameters; The laser adjustment unit is used to adjust the parameters of the laser pulse according to the laser target parameters and the laser output command, and output the target laser. The laser detection unit is used to detect the state of the target laser, determine the laser output state, and generate state feedback parameters.

4. The laser ball bonding control system according to claim 1, characterized in that, The visual detection module includes: The analysis and decomposition unit is used to analyze the laser ball bonding process, determine the visual inspection nodes, and obtain the corresponding task execution identifiers; The visual invocation unit is used to generate a task switching timestamp based on the visual detection node and the welding task sequence. The solder ball positioning unit is used to acquire images of the solder ball spraying position according to the first task execution identifier, obtain a spraying point position map, and mark the spraying point position with a label; The substrate positioning unit is used to acquire images of the position to be welded according to the second task execution identifier, obtain a substrate position map, and mark the substrate inspection label. The weld point inspection unit is used to acquire images of the finished welded product according to the third task execution identifier, obtain a finished weld point image, and mark the weld point inspection label. The image cleaning unit is used to filter and denoise the spray point location map, the substrate location map to be welded, and the finished weld point map, and to perform time-series association with the corresponding tags to construct a ball-mounting welding image tuple.

5. The laser ball bonding control system according to claim 4, characterized in that, The visual inspection module further includes: The parameter matching unit is used to perform modal coordination on the preset ball-planting welding model according to the task switching timestamp and match the corresponding modal-specific parameters. An anomaly identification unit is used to identify the ball-planting welding element group based on the modal-specific parameters and the corresponding status labels, determine the abnormal nodes, and calculate the abnormal deviation value. An anomaly control unit is used to determine an anomaly correction strategy based on the type of the execution device and the anomaly node, and to construct a deviation correction sequence based on the anomaly deviation value. The sequence classification unit is used to classify the deviation correction sequence pairs according to the status labels to obtain the stage adjustment sequence, and generate welding inspection results by combining the task execution identifier.

6. The laser ball bonding control system according to claim 1, characterized in that, The preset ball-planting welding model includes: The timing analysis layer is used to perform timing segmentation of the welding backbone network according to the laser ball bonding process, resulting in a solder ball determination branch, a substrate discrimination branch, and a solder joint discrimination branch. The feature extraction layer is used to perform feature analysis and edge detection on historical ball-mounted welding images, extracting features of spraying points, substrates to be welded, and finished weld points. The collaborative training layer is used to iteratively train the features of the injection point based on the preset single-branch training parameters and the solder ball determination branch, and generate the injection point positioning weight matrix. The substrate discrimination branch is combined with the features of the substrate to be welded for iterative training to generate the substrate localization weight matrix; Simultaneously, the features of finished solder joints are iteratively trained by combining the solder joint discrimination branch to generate a weight matrix of formed solder joints. Based on preset collaborative training parameters and a spatiotemporal attention mechanism, any two different weight matrices are jointly trained and spliced ​​to generate a global welding weight matrix. The time-discontinuity segmentation layer is used to segment the global welding weight matrix according to the task switching timestamp to form a local time-series weight matrix; The identification optimization layer is used to identify and detect historical ball-planting welding images based on the local time-series weight matrix, calculate stage index parameters, and compare them with preset stage index thresholds. If all the stage indicator parameters are greater than the stage indicator threshold, it indicates that the current ball-planting welding model is the optimal welding model. If not, then the difference between the stage indicator parameter and the stage indicator threshold is calculated to obtain the stage indicator difference, and the global welding weight matrix is ​​corrected to obtain the optimal welding weight matrix. The modal interaction layer is used to perform modal transformation on the optimal welding weight matrix according to the stage task objectives and output the modal-specific parameters for each stage.

7. The laser ball bonding control system according to claim 1, characterized in that, The central control module includes: The data receiving unit is used to receive and store tin monitoring data, laser control parameters, welding temperature parameters, pressure-adjusted spraying parameters and welding detection results based on the laser ball bonding process, and to build a welding timing parameter library; The instruction conversion unit is used to convert and distribute the welding timing parameter library according to the equipment protocol type, and generate equipment execution instructions; The coordinated control unit is used to coordinate the timing of each module according to the device execution instructions and modal-specific parameters, generate welding abnormality alarms, and complete the timing linkage of actions between modules.

8. A laser ball bonding control method, applied to the system described in any one of claims 1-7, characterized in that, include: Monitor the remaining amount of solder balls in the loading container, obtain the solder ball conveying status, and sort and transfer them to the target position according to the solder ball distribution instructions, and detect the solder ball arrival status; Adjust the laser pulse parameters according to the laser output command, and detect the laser output status; Monitor temperature changes during the solder ball melting process according to temperature sensing instructions, and collect solder ball temperature and solder joint temperature; Based on the welding control command and gas pressure control parameters, the injection pressure and injection timing of the inert gas are corrected. Based on the image detection instructions and the ball-mounting welding model, the solder joint position, solder ball position, solder spray point and solder joint forming state are collected and detected to generate welding inspection results. It receives and analyzes solder ball delivery status, laser output status, solder ball temperature, solder joint temperature, spray pressure, spray timing, and welding inspection results, and generates and distributes laser output commands, temperature sensing commands, welding control commands, and image inspection commands.

9. A storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the laser ball bonding control method as claimed in claim 8.

Images