SMT patch deviation correction system linked with AOI and method thereof

Abstract

The invention provides an SMT patch correction system linked with AOI and a method thereof. The SMT patch correction system comprises a stokehole AOI detection module, a stokehole AOI detection module, a multi-dimensional sensing module, an intelligent data center module, an AI decision optimization module and a process simulation verification module. According to the method, the AOI detection data in front of and behind the furnace are fused through the intelligent data center module, the mapping relation of knowledge graph associated elements, the technology and offset data is constructed, a linkage closed loop is formed, and offset source tracing and full-process dynamic deviation correction are achieved; environmental parameters such as temperature, vibration and suction nozzle pressure are collected in real time through the multi-dimensional sensing module, and the environmental parameters are standardized through an intelligent data platform and then are included into correction logic of the AI decision optimization module, so that dynamic compensation of the influence of environmental factors on the mounting precision is realized; a deep learning and reinforcement learning cooperation strategy is adopted through the AI decision optimization module, parameters are dynamically optimized and corrected in combination with thermal deformation simulation and other multi-source information of the process simulation verification module, and the adaptability to different element types and production scenes is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of surface mounting technology, in particular to an SMT patch deviation correction system linked with AOI and a method thereof. BACKGROUND

[0002] Surface Mount Technology (SMT) is the core process of high-density integration of components in the field of electronic manufacturing, and its mounting precision directly determines the performance and reliability of electronic equipment. As a key quality control link of SMT production line, Automatic Optical Inspection (AOI) detects the pre-mounting component position deviation before the furnace and the actual state of the component after welding, providing data support for mounting quality optimization. With the development of electronic components to 0201, 01005 and other miniaturized specifications, and the stringent requirements for mounting precision (within ±0.03mm) in the fields of 5G communication and automotive electronics, however, the traditional deviation correction method based on single AOI data or static parameters still has some problems: Firstly, the pre-furnace and post-furnace AOI detection data are isolated and do not form a closed loop, the pre-mounting deviation data before the furnace and the actual deviation data after welding lack fusion analysis, and the root cause of the deviation cannot be traced and dynamic correction cannot be achieved throughout the process. Secondly, there is a lack of real-time perception and compensation of environmental parameters. Temperature fluctuations, vibrations and nozzle pressure changes during the operation of the mounting machine will affect the mounting precision, and the existing system does not include such parameters in the deviation correction decision. Thirdly, the adaptability of the correction algorithm is insufficient. The traditional algorithm only generates static correction parameters based on a single data source, and cannot dynamically optimize with multi-source information such as thermal deformation simulation, making it difficult to adapt to different component types and production scenarios. Therefore, an SMT patch deviation correction system linked with AOI and a method thereof are proposed. SUMMARY

[0003] Therefore, the embodiments of the present application hope to provide an SMT patch deviation correction system linked with AOI and a method thereof to solve or alleviate the technical problems in the prior art, and at least provide a beneficial choice.

[0004] To solve the above technical problems, one technical solution adopted by the present application is: an SMT patch deviation correction system linked with AOI, comprising a pre-furnace AOI detection module, a post-furnace AOI detection module, a multi-dimensional perception module, an intelligent data platform module, an AI decision optimization module, a deviation correction execution module and a process simulation verification module. The pre-furnace AOI detection module is used to obtain pre-mounting deviation data and transmit it to the intelligent data platform module. The post-furnace AOI detection module is used to obtain actual deviation data after welding and transmit it to the intelligent data platform module. The multi-dimensional perception module is configured to collect environmental parameter data in real time and transmit the data to the intelligent data hub module. The intelligent data hub module is configured to receive pre-pasting offset data, actual offset data and environmental parameter data, establish a unified data model, build a knowledge graph, and generate a standardized data set. The AI decision optimization module is configured to receive the standardized data set from the intelligent data hub module, generate correction parameters based on deep learning and reinforcement learning, and send the correction parameters to the deviation correction execution module. The deviation correction execution module is configured to receive the correction parameters from the AI decision optimization module and adjust the position parameters and attitude parameters of the chip mounter according to the correction parameters. The process simulation verification module is configured to generate optimization suggestions through thermal deformation simulation and feed back to the AI decision optimization module.

[0005] The intelligent data hub module includes a protocol conversion unit, a knowledge graph construction unit and a data storage unit, which are further preferred in the technical solution. The protocol conversion unit is configured to convert detection data of different formats into a unified format. The knowledge graph construction unit is configured to associate the mapping relationship between component models, process parameters and offset data. The data storage unit is configured to store the standardized data set and historical production data.

[0006] The furnace AOI detection module includes a camera unit, a multi-spectral light source unit and an image preprocessing unit, which are further preferred in the technical solution. The camera unit is configured to collect image data of pre-pasting components. The multi-spectral light source unit is configured to provide red and white band light sources, which are respectively adapted to tin paste detection and component identification. The image preprocessing unit is configured to perform feature enhancement processing on the collected images.

[0007] The post-furnace AOI detection module includes a three-dimensional vision unit, a defect identification unit and a data output unit, which are further preferred in the technical solution. The three-dimensional vision unit is configured to scan three-dimensional coordinate data of components after welding. The defect identification unit is configured to identify welding defects based on a pre-set feature library. The data output unit is configured to transmit the actual offset data and defect information after integration.

[0008] The multi-dimensional perception module includes a temperature sensing unit, a pressure sensing unit and a vibration sensing unit, which are further preferred in the technical solution. The temperature sensing unit is used for detecting the real-time temperature of the motion shaft of the chip mounter and the reflow soldering furnace. The pressure sensing unit is used for monitoring the pressure value of the contact between the suction nozzle and the component. The vibration sensing unit is used for collecting the vibration frequency and amplitude data of the equipment during operation.

[0009] Further preferably, the AI decision optimization module comprises a deep learning unit, a reinforcement learning unit and a policy output unit. The deep learning unit is used for training an offset prediction model based on historical data. The reinforcement learning unit is used for generating correction parameters through dynamic iterative optimization. The policy output unit is used for converting the correction parameters into executable control signals.

[0010] Further preferably, the deviation correction execution module comprises a servo control unit, a suction nozzle adjustment unit and a parameter calibration unit. The servo control unit is used for driving the motor to adjust the X / Y axis displacement and rotation angle of the chip mounter. The suction nozzle adjustment unit is used for adjusting the height and pressure of the suction nozzle according to the correction parameters. The parameter calibration unit is used for periodically verifying the motion accuracy of the chip mounter.

[0011] Further preferably, the process simulation verification module comprises a thermal deformation model unit, a simulation calculation unit and a suggestion generation unit. The thermal deformation model unit is used for constructing a thermal expansion coefficient model of the PCB material. The simulation calculation unit is used for performing thermal deformation simulation based on the finite element analysis method. The suggestion generation unit is used for generating temperature curve optimization parameters and patching pressure compensation values according to the simulation results.

[0012] To solve the above technical problems, another technical solution adopted by the present application is an SMT patch deviation correction method linked with AOI, comprising the following steps: Step one, obtain pre-patching offset data, actual offset data after welding and environmental parameter data, and transmit them to an intelligent data hub module; Step two, the intelligent data hub module receives the pre-patching offset data, actual offset data and environmental parameter data, establishes a unified data model and constructs a knowledge graph, and generates a standardized data set; Step three, the AI decision optimization module receives the standardized data set, trains the offset prediction model through the deep learning unit, dynamically iteratively optimizes and corrects the parameters generated by the reinforcement learning unit, and outputs the correction parameters by the strategy output unit; Step four, the deviation correction execution module receives the correction parameters, adjusts the X / Y axis displacement and rotation angle of the chip mounter through the servo control unit, adjusts the height and pressure of the suction nozzle through the suction nozzle adjustment unit, and periodically checks the motion accuracy through the parameter calibration unit. Step five, the process simulation verification module constructs a PCB material thermal expansion coefficient model through the thermal deformation model unit, performs thermal deformation simulation based on finite element analysis through the simulation calculation unit, generates temperature curve optimization parameters and patch pressure compensation values according to the simulation results through the suggestion generation unit, and feeds back to the AI decision optimization module to form a closed loop.

[0013] As a further preferred technical solution of the present application, in step three, when the AI decision optimization module generates correction parameters, a deep learning and reinforcement learning collaborative strategy is adopted, specifically as follows: When the production batch is less than or equal to 5 batches, the deep learning unit weight ratio is 70%, and the reinforcement learning unit ratio is 30%. When the production batch is greater than 5 batches, the reinforcement learning unit weight is dynamically increased to 60%, and is adapted to the mounting characteristics of different component types through dynamic iterative optimization.

[0014] The embodiment of the present application has the following advantages due to the adoption of the above technical solution: 1. The present application fuses the pre-furnace and post-furnace AOI detection data through the intelligent data platform module, constructs the mapping relationship of the knowledge graph associated with the components, processes and offset data, forms a linkage closed loop, realizes the offset root tracing and full-process dynamic deviation correction; 2. The present application realizes the dynamic compensation of the influence of environmental factors on mounting accuracy through the multi-dimensional perception module, which collects environmental parameters such as temperature, vibration and suction nozzle pressure in real time, and after standardization by the intelligent data platform, the environmental parameters are included in the deviation correction logic of the AI decision optimization module; 3. The present application adopts a deep learning and reinforcement learning collaborative strategy through the AI decision optimization module, dynamically optimizes and corrects the parameters in combination with multi-source information such as thermal deformation simulation of the process simulation verification module, and improves the adaptability to different component types and production scenarios.

[0015] The above summary is only for the purpose of the description and is not intended to limit in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present application will be apparent from the drawings and the following detailed description. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced. Obviously, the accompanying drawings in the following description only represent some embodiments of the present application, and for those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 A functional module schematic diagram of an SMT patch rectification system linked with AOI according to the present application; Figure 2 A flowchart schematic diagram of an SMT patch rectification method linked with AOI according to the present application. DETAILED DESCRIPTION

[0018] The embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0019] It should be apparent that the following describes the embodiments of the present disclosure through specific concrete examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the disclosure. Obviously, the described embodiments are only some of the embodiments of the present disclosure, not all. The present disclosure can also be implemented or applied by other different specific embodiments, and each detail in the specification can be modified or changed based on different views and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present disclosure.

[0020] It should be noted that the following describes various aspects of the embodiments within the scope of the appended claims. It should be apparent that the aspects described herein can be embodied in a wide variety of forms and that any particular structure and / or function described herein is merely illustrative. Based on the disclosure, those skilled in the art should understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, devices and / or methods can be implemented using any number of the aspects set forth herein. In addition, this device and / or method can be implemented using other structures and / or functionality in addition to or instead of one or more of the aspects set forth herein.

[0021] It is also necessary to note that the drawings provided in the following embodiments only illustrate the basic concept of the present disclosure in a schematic manner, and only show the components related to the present disclosure in the drawings, not the number, shape and size of the components when actually implemented. The actual implementation of each component may be a random change in type, number and proportion, and the component layout pattern may also be more complex.

[0022] In addition, in the following description, specific details are provided in order to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the aspects described can be practiced without these specific details.

[0023] Figure 1 is a functional module schematic diagram of an SMT patch rectification system associated with AOI according to an embodiment of the present application, as shown in Figure 1 An SMT patch rectification system associated with AOI, comprising a pre-furnace AOI detection module, a post-furnace AOI detection module, a multi-dimensional perception module, an intelligent data platform module, an AI decision optimization module, a rectification execution module and a process simulation verification module. The pre-furnace AOI detection module is configured to obtain pre-pasting offset data and transmit the pre-pasting offset data to the intelligent data platform module. Specifically, first, the multi-spectral light source unit of the pre-furnace AOI detection module is started, and the corresponding waveband is switched according to the detection object - the red waveband light source is enabled for tin paste detection, and the white waveband light source is enabled for component identification, to ensure the clarity of image acquisition; then, the camera unit acquires images of the PCB board after pre-pasting under the cooperation of the light source, to obtain original image data of the pre-pasted components; next, the image preprocessing unit performs feature enhancement processing on the collected original image, including denoising, contrast adjustment, etc., to highlight the component edge and position features; thereafter, based on the pre-processed image, the actual position of the component and the design coordinate deviation are analyzed by an image recognition algorithm, and the pre-pasting offset data is calculated; finally, the pre-furnace AOI detection module transmits the calculated pre-pasting offset data to the intelligent data platform module through the GigE Vision protocol.

[0024] The post-furnace AOI detection module is configured to obtain actual offset data after welding and transmit the actual offset data to the intelligent data platform module. Specifically, first, the three-dimensional vision unit of the post-furnace AOI detection module is started, and the PCB after welding is scanned to collect three-dimensional coordinate data of the components after welding; then, the defect recognition unit calls the preset welding defect feature library to analyze the three-dimensional coordinate data and image information obtained by scanning, and identify welding defects such as virtual welding and tin connection; then, based on the three-dimensional coordinate data, the actual position of the component after welding is calculated and the deviation of the design coordinate is generated. Actual offset data; thereafter, the data output unit integrates the actual offset data and the identified welding defect information to form complete detection data; finally, the post-furnace AOI detection module transmits the integrated detection data to the intelligent data platform module through the proprietary protocol.

[0025] The multi-dimensional perception module is used to collect environmental parameter data in real time and transmit it to the intelligent data platform module; Specifically, first, the temperature sensing unit, pressure sensing unit and vibration sensing unit of the multi-dimensional perception module are started synchronously, the temperature sensing unit detects the real-time temperature of the motion shaft of the chip mounter and the reflow soldering furnace in real time, the pressure sensing unit monitors the pressure value of the contact between the suction nozzle and the component, and the vibration sensing unit collects vibration frequency and amplitude data when the equipment is running. Then, each sensing unit filters and calibrates the collected raw data to remove noise interference and ensure data accuracy; then, the module integrates the processed temperature, pressure, vibration and other environmental parameter data into a unified format data set; finally, the multi-dimensional perception module transmits the integrated environmental parameter data to the intelligent data platform module through the CAN bus.

[0026] The intelligent data platform module is used to receive pre-mounting offset data, actual offset data and environmental parameter data, establish a unified data model and build a knowledge graph, and generate a standardized data set; Specifically, first, the intelligent data platform module receives pre-mounting offset data from the pre-furnace AOI detection module, actual offset data from the post-furnace AOI detection module, and environmental parameter data from the multi-dimensional perception module through corresponding interfaces; then, the protocol conversion unit converts and processes the received data of different formats (such as AOI detection data and sensing parameter data) to a preset data format, eliminating format differences; then, the knowledge graph construction unit associates the mapping relationship between component models, process parameters and offset data based on the converted standardized data, and constructs a knowledge graph containing multi-dimensional associated information; at the same time, the module establishes a unified data model based on the knowledge graph and the integrated various data, realizing the structured organization of data; thereafter, the data storage unit classifies and stores the original data, converted data and knowledge graph data to ensure data traceability; finally, the module generates a standardized data set based on the unified data model, providing input data for the AI decision optimization module.

[0027] An AI decision optimization module is configured to receive the standardized data set from the intelligent data middle station module, generate correction parameters based on deep learning and reinforcement learning, and send the correction parameters to the deviation rectification execution module. Specifically, first, the AI decision optimization module receives the standardized data set transmitted by the intelligent data middle station module, which includes pre-attachment offset data, actual offset data, environmental parameter data, and knowledge graph association information. Then, the deep learning unit calls historical production data and trains the offset prediction model in combination with the standardized data set to output preliminary offset prediction results based on historical rules. Next, the reinforcement learning unit generates correction parameters based on the preliminary prediction results through dynamic iterative optimization correction parameter generation logic and adjusts parameter weights in combination with real-time production scenarios (such as component types and production batches) to generate correction parameters that adapt to the current scenario. At the same time, the strategy output unit converts the correction parameters into control signal formats recognizable by the placement machine. Finally, the AI decision optimization module sends the converted control signals to the deviation rectification execution module.

[0028] The deviation rectification execution module is configured to receive the correction parameters from the AI decision optimization module and adjust the position parameters and attitude parameters of the placement machine according to the correction parameters. Specifically, first, the deviation rectification execution module receives the correction parameters from the AI decision optimization module, which include position compensation, attitude adjustment values, and suction nozzle control parameters. Then, the servo control unit drives the motor to adjust the X / Y axis displacement of the placement machine according to the position compensation in the correction parameters and corrects the rotation angle according to the attitude adjustment values to achieve precise calibration of the placement position and attitude. Next, the suction nozzle adjustment unit synchronously adjusts the height and pressure of the suction nozzle based on the suction nozzle control parameters in the correction parameters to ensure that the component attachment force adapts to the component type and PCB board state. At the same time, the parameter calibration unit checks the motion accuracy of the placement machine at a preset period (such as every 1000 PCBs) and generates calibration compensation values by comparing the actual adjustment results with the deviation of the correction parameters. Finally, the module feeds back the status information of the adjustment completed to the AI decision optimization module to form an execution closed loop.

[0029] The process simulation verification module is configured to generate optimization suggestions through thermal deformation simulation and feed back to the AI decision optimization module. Specifically, first, the thermal deformation model unit of the process simulation verification module is started, the PCB design file and material parameters (such as thermal expansion coefficient) are obtained from the intelligent data platform module, the thermal expansion coefficient model of the PCB material is constructed, and the basic physical model is provided for the thermal deformation simulation; then, the simulation calculation unit calls the thermal expansion coefficient model, and the thermal deformation of the PCB in the reflow soldering process is simulated and calculated based on the finite element analysis method, and the position offset trend of the components under different temperature curves is simulated; then, the suggestion generation unit analyzes the simulation results, generates temperature curve optimization parameters (such as temperature adjustment value of each temperature zone) and patch pressure compensation value according to the component type and mounting requirements; finally, the process simulation verification module transmits the generated optimization suggestions to the AI decision optimization module, provides simulation basis for the generation of corrected parameters, and forms a closed loop feedback.

[0030] In one embodiment, specifically, the intelligent data platform module includes a protocol conversion unit, a knowledge graph construction unit and a data storage unit; The protocol conversion unit is configured to convert detection data in different formats into a unified format. Specifically, first, the protocol conversion unit receives the pre-mounting offset data from the pre-furnace AOI detection module, the actual offset data from the post-furnace AOI detection module and the environmental parameter data from the multi-dimensional perception module through the adaptive interface. These data adopt different formats (such as image derived data format of pre-furnace AOI, three-dimensional coordinate data format of post-furnace AOI, and sensing signal format of multi-dimensional perception module) due to different source modules. Then, the unit analyzes the format of the received data, identifies the data field meaning, coding rule and transmission protocol characteristics; then, according to the unified data standard (including data structure, field definition, precision requirement, etc.) preset by the intelligent data platform module, the different format data after analysis is converted and processed, and the image derived data, three-dimensional coordinate data and sensing signal are uniformly mapped to structured data format; finally, the protocol conversion unit outputs the converted unified format data, and provides standardized input for the knowledge graph construction unit and the data storage unit.

[0031] The knowledge graph construction unit is configured to associate the mapping relationship between component model, process parameters and offset data. Specifically, first, the knowledge graph construction unit obtains the unified format data from the protocol conversion unit, extracts key entity information from it, including component model (such as 0201, BGA, etc.), process parameters (such as mounting speed, suction nozzle pressure, reflow soldering temperature curve, etc.), offset data (pre-mounting offset, actual offset after soldering); then, the unit defines the correlation rules between entities, for example: the mapping relationship between “0201 component + suction nozzle pressure 0.3N + reflow soldering temperature 260℃” and “pre-mounting offset ±10μm, actual offset ±8μm”, the mapping relationship between “BGA chip + mounting speed 50mm / s + reflow soldering temperature rising rate 3℃ / s” and “pre-mounting offset ±20μm, actual offset ±15μm”; then, based on the extracted entities and defined correlation rules, an initial knowledge graph is constructed to form a ternary tuple network of “component model-process parameter-offset data”, where each node represents an entity and the edge represents the mapping relationship and correlation strength between entities; after that, as new production data (such as new component model, adjusted process parameters, new offset data) is input, the unit dynamically updates the knowledge graph, optimizes the weight of the correlation rules, and enhances the accuracy of the mapping relationship; finally, the knowledge graph construction unit outputs the updated knowledge graph to generate a standardized data set for the intelligent data center module and provide correlation relationship support for the AI decision optimization module to train the model.

[0032] A data storage unit for storing the standardized data set and historical production data; Specifically, first, the data storage unit receives the unified format data (i.e. the standardized data set) from the protocol conversion unit, and the past data imported from the historical production records (including the offset data, process parameter settings, environmental parameter changes, and corresponding production yield of the historical batches, etc.); then, the unit classifies and labels the received data, stores the standardized data set in zones according to “pre-furnace AOI data”, “post-furnace AOI data”, and “environmental parameter data”, and indexes the historical production data according to “component model”, “production batch”, and “process version”, to ensure clear data classification; then, the time series database is used to store the real-time generated standardized data set (adapted to the fast writing and querying of time series data such as environmental parameters), and the relational database is used to store the structured historical production data (to facilitate the correlation query of the corresponding relationship between process parameters and offset data of different batches); after that, data storage rules are set, including data retention period (such as retaining production data for nearly 12 months), backup frequency (such as daily incremental backup and weekly full backup), to prevent data loss and meet the traceability requirements; finally, the data storage unit opens a data query interface to support the knowledge graph construction unit of the intelligent data center module to call historical data for correlation analysis, and the AI decision optimization module to extract historical data to train the offset prediction model.

[0033] In one embodiment, specifically, the pre-reflow AOI detection module comprises a camera unit, a multi-spectrum light source unit and an image preprocessing unit; The camera unit is configured to acquire image data of the pre-mounted components. Specifically, first, the camera unit performs parameter initialization after starting, calibrates the lens focal length, exposure time, white balance, etc., and ensures the adaptation to the multi-spectrum light source unit of the pre-reflow AOI detection module (e.g., calibrates the imaging parameters to adapt to the red and white band light sources); then, according to the detection scene (solder paste detection or component identification), the light source switching of the multi-spectrum light source unit is synchronously responded: when detecting the solder paste, the exposure parameters are adjusted (e.g., the exposure time is shortened to avoid overexposure) in cooperation with the red band light source; when identifying the components, the focal length is optimized (e.g., the macro mode is adjusted for tiny components such as 0201) in cooperation with the white band light source; then, the camera unit acquires continuous images of the pre-mounted components through a preset scanning path (e.g., scans row by row according to the partition of the PCB), ensures that all mounted components on the PCB are covered, and the acquired original images contain the position, contour and pin features of the components; finally, the camera unit transmits the acquired image data to the image preprocessing unit of the pre-reflow AOI detection module in real time, providing original image input for subsequent feature enhancement and offset data calculation.

[0034] The multi-spectrum light source unit is configured to provide red and white band light sources, respectively adapted to solder paste detection and component identification. Specifically, first, the multi-spectrum light source unit performs light source calibration after starting, detects the wavelength purity and intensity stability of the red band (620-660 nm) and the white band (400-760 nm) through the built-in spectrum analyzer, ensures that the peak wavelength deviation of the red band is ≤5 nm, and the color temperature of the white band is stable at 5500-6500 K. Then, according to the detection task instruction of the pre-reflow AOI detection module, the light source is switched: when performing solder paste detection, the unit activates the red band light source, and uses the high reflection characteristic of the solder paste to highlight the contrast between the solder paste area and the PCB substrate; when performing component identification, the white band light source is switched, the color, texture and pin details of the components are restored through full-spectrum illumination, and the contour identification of different types of components such as 0201 and SOP is adapted; Then, the light source irradiation angle is adjusted (such as using 45° oblique + coaxial combined illumination), the light is uniformly covered on the pre-mounted component area through the diffusion plate to avoid shadow interference caused by uneven illumination; at the same time, the unit and the camera unit establish a synchronous triggering mechanism, the light source switching is linked with the camera exposure time (such as setting the exposure time to 20 ms under red light and 15 ms under white light), to ensure the brightness adaptation during image acquisition; finally, the multi-spectral light source unit continuously monitors the light source intensity, and automatically compensates the power when the detected attenuation exceeds 10%, to provide stable lighting support for the camera unit to collect clear pre-mounted component images.

[0035] An image preprocessing unit is configured to perform feature enhancement processing on the collected images. Specifically, first, the image preprocessing unit receives the pre-mounted component original image data transmitted by the camera unit, and loads the preset processing parameter template (parameters are configured for different component types such as 0201, BGA, and solder paste area) when starting; then, noise suppression processing is performed on the original image, and an adaptive median filter algorithm is used to remove sensor noise and environmental light interference, and to retain component edge details (such as for the granular noise of the solder paste image, the filter window is set to 3x3 pixels; for the smooth noise of the component image, the window is adjusted to 5x5 pixels); then, contrast enhancement processing is performed, the local contrast of the image is improved through the limited contrast adaptive histogram equalization (CLAHE) algorithm, the light and dark differences caused by uneven light source are solved, and the gray scale features of the solder paste area and the outline boundary of the component are highlighted; then, edge enhancement processing is performed, the Sobel operator is used to calculate the gradient of the image, the pixel difference of the component pin and edge outline is strengthened, and an edge feature map is generated; at the same time, the processed image is subjected to coordinate normalization, the image pixel coordinates are mapped and calibrated with the physical coordinates of the PCB design file, to ensure the coordinate consistency of subsequent offset calculation; finally, the image preprocessing unit outputs the feature-enhanced image data, to provide a clear and identifiable image basis for the calculation of pre-mounted offset data.

[0036] In one embodiment, specifically, the furnace post-AOI detection module includes a three-dimensional vision unit, a defect identification unit, and a data output unit. The three-dimensional vision unit is configured to scan the three-dimensional coordinate data of the post-welding component. Specifically, first, the three-dimensional vision unit starts and performs hardware initialization, calibrates the built-in laser scanning components, industrial cameras and displacement platforms to ensure scanning accuracy (such as calibration to within ±2μm error range); then, the reference point (such as Mark point) of the PCB after welding is identified by the positioning sensor to determine the boundary of the scanning area and ensure that all welded elements on the PCB are covered; then, the laser triangulation method combined with structured light projection technology is used to scan the elements after welding point by point: the laser transmitter projects a line of laser light onto the element surface, and the industrial camera synchronously collects the reflected light signal, and the depth information of each point on the element surface is obtained by calculating the offset of the light; then, the unit splices, denoises (removes the edge and background interference points of the PCB) and coordinates the collected raw point cloud data, converts the relative coordinates into an absolute coordinate system consistent with the PCB design file, and extracts the three-dimensional coordinates (X, Y axis plane coordinates and Z axis height coordinates) of each element; finally, the three-dimensional vision unit transmits the processed three-dimensional coordinate data of the elements after welding to the defect recognition unit and data output unit of the post-furnace AOI detection module, providing a basis for subsequent actual offset data calculation.

[0037] The defect recognition unit is configured to recognize welding defects based on a preset feature library. Specifically, first, the defect recognition unit loads the preset welding defect feature library when it starts, which contains standard feature parameters of various typical defects: such as virtual welding (contact area of element and pad <60%), tin bridge (width of tin bridge between adjacent pads >0.1mm), less tin (tin amount on pad <50% of design value), excessive offset (actual offset > design threshold), etc., and labels the image features (gray scale distribution, contour shape) and three-dimensional features (height difference, volume deviation) of each defect; then, the unit receives the three-dimensional coordinate data of the elements after welding and the supporting image information transmitted by the three-dimensional vision unit, and extracts features from the data: calculates the adhesion of the element and the pad, the tin volume and height from the three-dimensional coordinates, and extracts two-dimensional features such as pad edge contour and tin paste diffusion shape from the image; then, the actual features extracted are compared one by one with the standard features in the preset feature library by using the combination of template matching and feature vector comparison, and the defect type is recognized by calculating the feature similarity (such as setting the similarity threshold ≥85% to determine the match); at the same time, the identified defects are quantified in terms of parameters, such as the length of the tin bridge, the area ratio of the virtual welding, and the specific value of the offset, to form the defect details; finally, the defect recognition unit transmits the recognition results (including defect type, location, and quantitative parameters) to the data output unit, providing a basis for the integration of actual offset data and defect information in the subsequent process.

[0038] The data output unit is configured to integrate and transmit actual offset data and defect information. Specifically, first, the data output unit receives the actual offset data after welding (including the X / Y / Z axis coordinate deviation values of the components) from the three-dimensional vision unit and the welding defect information (including defect type, location, quantitative parameters, etc.) from the defect recognition unit; then, the unit associates and matches the two types of data, binds the actual offset data of the same component with the corresponding defect information through the unique identification of the component (such as the coordinate position coding on the PCB), forms the associated record of “component position-offset- defect details”, and adds metadata to the associated record, including detection time, PCB batch number, post-furnace AOI equipment number, etc., to construct a structured data set containing complete context information; after that, the structured data set is format-verified to ensure that the data fields are complete (such as no missing offset or defect type) and the format is unified (in line with the receiving standard of the intelligent data center module); finally, the data output unit transmits the integrated data after verification to the intelligent data center module through a preset transmission protocol (such as TCP / IP), providing a basis for subsequent data fusion analysis.

[0039] In one embodiment, specifically, the multi-dimensional perception module includes a temperature sensing unit, a pressure sensing unit, and a vibration sensing unit. The temperature sensing unit is used to detect the real-time temperature of the motion shaft of the chip mounter and the reflow soldering furnace. Specifically, first, the temperature sensing unit performs sensor calibration after starting, performs zero correction and linearity compensation on the built-in thermocouple sensor (adapted to the range of -50℃~300℃), and ensures that the detection accuracy is controlled within ±0.5℃; then, the unit detects the target through distributed sensing nodes: 2 sensors are deployed at each of the X-axis, Y-axis, and Z-axis motion guide rails of the chip mounter to monitor the temperature change of the motion shaft in real time; 3 sensors are respectively deployed in the preheating zone, constant temperature zone, and reflow zone of the reflow soldering furnace to collect real-time temperatures in different temperature zones; then, the temperature data of each detection point is synchronously collected at a sampling frequency of 10Hz, and the original data includes detection point number, collection timestamp, and temperature value; after that, the collected original data is filtered (using a sliding average filtering algorithm with a window size of 5 sampling points) to remove temperature jumps caused by transient interference, and a smooth real-time temperature curve is output; finally, the temperature sensing unit encapsulates the processed temperature data in JSON format (including detection point identification, real-time temperature, and timestamp) and transmits it to the integration unit of the multi-dimensional perception module through the RS485 bus to provide temperature information for the integration of subsequent environmental parameter data.

[0040] The pressure sensing unit is used to monitor the pressure value of the contact between the suction nozzle and the component. Specifically, first, the pressure sensing unit is started after the sensor calibration, the built-in micro pressure sensor (such as a strain gauge sensor with a range of 0-5N and an accuracy of ±0.01N) is calibrated to zero and linearly corrected to ensure that the output is stable when the nozzle is not in contact, and the error is ≤0.02N when the contact pressure is in the range of 1-3N; then, the calibrated pressure sensor is integrated at the end of the patch machine nozzle or the connecting arm to ensure that the sensor can directly sense the pressure change during the contact of the nozzle with the component and the fitting process; then, the unit sets the sampling trigger mechanism: when the nozzle is lowered to 5mm from the component surface, high-frequency sampling (sampling frequency 50Hz) is started, and the whole process of pressure data from the nozzle contacting the component, applying pressure to the mounting is captured in real time; then, the collected raw pressure data is filtered, low-pass filtering algorithm is used to remove high-frequency noise caused by mechanical vibration, and effective pressure value (i.e. the pressure peak value and duration when the nozzle and the component are in stable contact) is extracted; at the same time, the unit associates the effective pressure data with the nozzle number, component pickup / placement timestamp to form structured data (such as "Nozzle No. 3-Component pickup pressure 1.2N-Placement pressure 0.8N-Timestamp xxxx"); finally, the pressure sensing unit transmits the processed pressure monitoring data to the integration unit of the multi-dimensional perception module to provide pressure dimension information for real-time integration of environmental parameter data.

[0041] The vibration sensing unit is used to collect vibration frequency and amplitude data when the device is running. Specifically, first, the vibration sensing unit is started after the sensor calibration, the built-in piezoelectric vibration sensor (such as a sensor with a range of 0-500Hz and an amplitude measurement range of ±50μm) is calibrated for sensitivity and phase compensation to ensure that the measurement error is ≤2% in the frequency range of 10-300Hz, and the amplitude measurement accuracy is ≤1μm; then, the calibrated vibration sensor is deployed at the key vibration source position of the patch machine: one sensor is installed at each of the machine body base, X / Y axis motor base, and nozzle driving mechanism to comprehensively collect vibration signals when the device is running; then, the unit sets a dynamic sampling mechanism: when the patch machine is in a static state, low-frequency sampling (10Hz) is used; when in dynamic stages such as component pickup, high-speed movement, and mounting, it automatically switches to high-frequency sampling (1000Hz) to capture the vibration frequency change and instantaneous amplitude peak value in real time; then, the collected raw vibration signal is analyzed and filtered by Fourier transform to extract the main vibration frequency components (such as motor operating characteristic frequency, mechanical resonance frequency, etc.), and band-pass filtering is used to remove environmental low-frequency interference (such as ground vibration) to separate the effective vibration data generated by the device itself; Meanwhile, the unit associates the processed vibration frequency and amplitude data with the patching machine operation stage (such as pickup, movement, and placement) and the corresponding component model to form a structured record (such as "X-axis movement stage-vibration frequency 50Hz-amplitude 15μm-component model 01005-time stamp xxxx"); finally, the vibration sensing unit transmits the integrated vibration data to the integration unit of the multi-dimensional perception module to provide vibration dimension information for real-time aggregation of environmental parameter data.

[0042] In one embodiment, specifically, the AI decision optimization module includes a deep learning unit, a reinforcement learning unit, and a strategy output unit. The deep learning unit is configured to train an offset prediction model based on historical data. Specifically, first, the deep learning unit extracts historical production data, including historical pre-pasting offset data, actual offset data after welding, component model of the corresponding batch, process parameters (such as mounting speed, suction nozzle pressure), and environmental parameters (temperature, vibration, etc.), from the storage unit of the intelligent data platform module through the data interface to construct a training data set; then, the training data set is preprocessed: abnormal values (such as extreme offset data caused by equipment failure) are removed, numerical value type parameters (such as temperature, pressure) are normalized (mapped to the 0-1 interval), category type data (such as component model) is one-hot encoded, and the training set and validation set are divided in a ratio of 8:2; next, a hybrid offset prediction model based on convolutional neural network (CNN) and long short-term memory network (LSTM) is constructed: the CNN layer is used to extract the spatial correlation of environmental parameters, component model, etc. (such as the difference in temperature sensitivity of different components), the LSTM layer is used to capture the dynamic features of time series data (such as the offset trend of consecutive batches), and the output layer uses a fully connected layer to predict the pre-pasting and post-welding offset (X, Y axis deviation); then, the prediction error (mean square error MSE) of the validation set is used as the optimization target, and the model is iteratively trained through the backpropagation algorithm: the initial learning rate is set to 0.001, the training epoch is 50, the batch size is 32, the model accuracy is evaluated with the validation set after each training, and the training is stopped when the validation set error does not decrease for 5 consecutive times, and the optimal model parameters are saved; finally, the trained offset prediction model is deployed to the AI decision optimization module to receive the standardized new data set from the intelligent data platform module, and to predict the component offset trend in the current production scenario in real time, providing a basis for the reinforcement learning unit to optimize and correct parameters.

[0043] The reinforcement learning unit is configured to generate logic for dynamically iterating and optimizing correction parameters. Specifically, first, the reinforcement learning unit receives the preliminary offset prediction results output by the deep learning unit, and at the same time acquires real-time production data (including the current component model, process parameter settings, environmental parameters, and generated offset data) of the intelligent data center module, and integrates these information into the initial state space of reinforcement learning; then, the core reinforcement learning elements are defined: the state space includes the current production scene characteristics (component type, environmental temperature, vibration frequency, etc.) and historical offset trends; the action space is the adjustment range of the correction parameters (such as X / Y axis compensation ± 50 μm, suction nozzle pressure adjustment ± 0.2 N, etc.); the reward function takes the deviation of the actual offset and the target threshold (the smaller the deviation, the higher the reward value), the production yield improvement amplitude, and the parameter adjustment cost (such as the smaller the adjustment amplitude, the lower the cost) as the comprehensive index; then, the deep Q network (DQN) algorithm is used for dynamic iterative training: the correction parameter generation logic is initialized (such as the initial parameter generation formula based on rules), after each batch of production, the actual execution result (offset after correction, yield) is calculated to calculate the immediate reward, and the parameter generation logic is optimized through Q value update, so that the correction parameter is more suitable for the current scene; at the same time, the exploration and utilization balance mechanism is set: when the new component type or process parameter changes, the exploration rate is increased (randomly try new parameter combination) to find better logic; in the stable production stage, the utilization rate is increased (generate parameters based on the historical optimal strategy) to ensure stability; finally, after multiple iterations (such as updating the logic once every 500 PCBs produced), the reinforcement learning unit outputs the optimized correction parameter generation logic, ensuring that the generated correction parameters can dynamically adapt to different production scenes (such as component size changes, environmental fluctuations), and improving the correction accuracy.

[0044] The strategy output unit is configured to convert the correction parameters into executable control signals. Specifically, first, the strategy output unit receives the optimized correction parameters output by the reinforcement learning unit, including position compensation (X / Y axis offset compensation value, rotation angle correction value), suction nozzle control parameters (pressure adjustment value, height adjustment amount), and motion axis speed correction value, etc., in the form of numerical data (such as "X-axis compensation + 15 μm, suction nozzle pressure 0.35 N"); Then, the unit loads the control signal protocol template of the chip mounter (such as the signal format specification based on Modbus or EtherCAT protocol), converts the correction parameters into executable control signals according to the protocol requirements: converts the position compensation into the pulse number of the servo motor (such as 1 μm corresponding to 10 pulses), converts the pressure adjustment value into the voltage control signal of the air pressure valve (such as 0-5V corresponding to 0-1N pressure), and converts the speed correction value into the frequency signal of the driver; Then, the converted control signal is checked for effectiveness to check whether the signal is within the safe threshold range of the equipment (such as the pulse number does not exceed the maximum stroke corresponding value of the motor, and the voltage signal does not exceed the upper limit of 5V), to avoid equipment failure caused by abnormal parameters; At the same time, the unit establishes a signal synchronization mechanism to ensure that the position control signal, pressure control signal and speed control signal are sent to the corresponding execution components (X / Y axis servo system, suction nozzle drive module, motion axis controller) according to the preset timing, ensuring the synergy of multi-parameter adjustment; Finally, the strategy output unit feeds back the conversion results of the control signal (such as signal sending state, whether it meets the protocol specification) to the AI decision optimization module, completes the conversion closed loop of correction parameters to control signals, and provides direct instruction support for the precise action of the deviation correction execution module.

[0045] In one embodiment, specifically, the deviation correction execution module includes a servo control unit, a suction nozzle adjustment unit and a parameter calibration unit; The servo control unit is used to drive the motor to adjust the X / Y axis displacement and rotation angle of the chip mounter. Specifically, first, the servo control unit receives control signals from the strategy output unit, which contain pulse instructions corresponding to X / Y axis displacement (such as 10 pulses per μm) and the number of pulses corresponding to the rotation angle (such as 500 pulses per degree); then, the unit decodes the pulse signals, combines the current position information of the chip mounter (obtained through the built-in encoder), calculates the actual distance that the X / Y axis needs to move and the angle that the rotation axis needs to rotate; next, the drive module outputs drive current to the X-axis, Y-axis and rotation axis servo motors according to the calculation results, controls the motor to start according to the preset acceleration, and in the moving process, real-time acquisition of motor position feedback signals is realized through the encoder, which is compared with the instruction position, and the drive current is dynamically adjusted through the PID algorithm to ensure the displacement accuracy (such as within ±1 μm) and the rotation angle accuracy (such as within ±0.01°); at the same time, the unit monitors the current, temperature and other parameters of the motor during operation, and when an abnormality (such as current exceeding the rated value, temperature being too high) is detected, the protection mechanism (such as pausing movement, alarm prompt) is automatically triggered; finally, the servo control unit feeds back the adjusted actual position data (X / Y axis coordinates, rotation angle) to the deviation correction execution module, confirms that the displacement and angle adjustment is in place, and provides a position reference for subsequent suction nozzle adjustment and other steps.

[0046] The suction nozzle adjustment unit is used to adjust the height and pressure of the suction nozzle according to the correction parameters. Specifically, first, the suction nozzle adjustment unit receives correction parameters from the strategy output unit, which contain suction nozzle height adjustment amount (such as ±0.5 mm) and pressure adjustment value (such as ±0.1 N), which are determined based on component type (such as micro-chip requiring lower pressure) and PCB board state (such as surface flatness); then, the unit converts the height adjustment amount into drive signals of the servo motor (such as 200 pulses per 0.1 mm), controls the Z-axis servo motor to move the suction nozzle up and down, and at the same time, real-time monitoring of the actual height of the suction nozzle is realized through the laser displacement sensor, which is compared with the target height, and the deviation is corrected through closed-loop control (control accuracy ±0.01 mm); next, the pressure adjustment value is converted into control signals of the gas pressure regulating valve (such as 0-5V voltage corresponding to 0-1N pressure), the output pressure of the gas source is adjusted, and at the same time, the actual pressure when the suction nozzle contacts the component is collected through the pressure sensor and fed back to the unit for dynamic calibration, ensuring that the pressure is stable within the target value ±0.02N range; at the same time, the unit links the timing control of the chip mounter, pre-adjusts the height and pressure when the suction nozzle approaches the component, and completes the final calibration at the moment of contact, avoiding component damage or unstable adsorption; finally, the suction nozzle adjustment unit feeds back the adjusted height and pressure actual values to the deviation correction execution module, confirming that the suction nozzle parameters adapt to the current mounting requirements.

[0047] The parameter calibration unit is used to periodically verify the motion accuracy of the chip mounter. Specifically, first, the parameter calibration unit loads a preset motion accuracy calibration template when starting, which contains standard calibration parameters of each motion axis (X-axis, Y-axis, rotation axis) and suction nozzle positioning of the chip mounter (such as X-axis one-way motion accuracy allowed error ±2μm, rotation axis angle deviation allowed value ±0.02°); Then, the unit triggers the calibration process according to the preset period (such as every 2000 pieces of PCB produced or every 8 hours), or starts immediately when receiving a manual calibration instruction; Next, the verification is performed through a high-precision calibration jig (such as a standard plate with a precise scale, a laser interferometer): the chip mounter is controlled to move according to the preset path (such as X-axis reciprocating movement of 100mm, Y-axis movement of 50mm, rotation axis rotation of 360°), and the actual motion trajectory data is collected synchronously through the three-dimensional vision unit or the laser ranging sensor; Then, the collected actual trajectory data is compared with the theoretical trajectory parameters, the position deviation of each axis (such as X-axis actual movement of 99.995mm, deviation of-5μm), the repeat positioning error (the maximum difference of multiple movements to the same position) and the angle deviation are calculated, and a quantitative accuracy verification report is generated; At the same time, the motion accuracy compensation parameters are automatically generated according to the deviation value (such as setting +5μm compensation value for X-axis deviation-5μm), and the motion control parameter library of the chip mounter is updated; Finally, the parameter calibration unit feeds back the calibration results (including deviation data and compensation value) to the deviation correction execution module and the AI decision optimization module, to ensure that the subsequent mounting motion accuracy is maintained within the preset threshold.

[0048] In one embodiment, specifically, the process simulation verification module includes a thermal deformation model unit, a simulation calculation unit and a suggestion generation unit; The thermal deformation model unit is configured to construct a thermal expansion coefficient model of the PCB material. Specifically, first, the thermal deformation model unit loads a preset material property database when starting, which contains the basic thermal expansion coefficient (CTE) reference values of common PCB substrates (such as FR-4, aluminum substrate), copper foil, solder mask (solder mask), and other materials (for example, the CTE of FR-4 is usually 14-20 ppm / ℃); then, the unit obtains the detailed design file (including the layer structure, material type and thickness of each layer) and material parameters (such as the measured CTE data of each material provided by the supplier) of the current PCB from the smart data platform module through the data interface; next, a thermal expansion coefficient model is constructed based on the laminate theory: considering the thickness proportion, directional anisotropy (such as the CTE difference of FR-4 in-plane and out-of-plane) and interlayer bonding force of each layer of material, the equivalent thermal expansion coefficient of the whole PCB in different temperature intervals (such as room temperature to 260℃ reflow soldering temperature) is calculated by weighted calculation, the model needs to reflect the nonlinear relationship between temperature and expansion coefficient (the CTE of some materials will change at high temperature); then, the model is calibrated, the calculated equivalent CTE is compared with the CTE back calculated from the historical thermal deformation data of similar PCBs, and the material interaction coefficient in the model is corrected to ensure that the model error is ≤5%; finally, the thermal deformation model unit outputs the optimized PCB material thermal expansion coefficient model, including the equivalent CTE values corresponding to each temperature point and the model parameters, providing basic physical model support for the simulation calculation unit of the process simulation verification module.

[0049] The simulation calculation unit is configured to perform thermal deformation simulation based on a finite element analysis method. Specifically, first, the simulation calculation unit receives the PCB material thermal expansion coefficient model from the thermal deformation model unit, and simultaneously obtains the current PCB three-dimensional design model (including component layout, pad position), reflow soldering process temperature curve data (such as the temperature rising process from preheating to 260°C), and material attribute parameters (layer thickness, elastic modulus, etc.) from the intelligent data center module; then, a thermal deformation simulation model is constructed based on the finite element analysis (FEA) method: the PCB is divided into a fine grid (such as a component area grid size of 0.1 mm, and a non-component area of 0.5 mm), the thermal physical properties of each material are defined (combined with the thermal expansion coefficient model), the boundary conditions are set (such as the temperature field distribution in the reflow soldering furnace and the PCB fixed constraint position); then, an explicit dynamic algorithm is used for thermal deformation simulation calculation to simulate the dynamic deformation of the PCB in the whole reflow soldering temperature change process: the displacement of each node at different temperature points (such as 100°C, 200°C, and 260°C) is calculated, the thermal deformation offset of the pad position (X / Y axis direction stretching amount) is analyzed, and the thermal deformation cloud chart and the deformation curve of the key position are generated; then, the simulation results are compared with the actual thermal deformation data (such as the actual offset of the component after soldering) of the post-furnace AOI detection module, the simulation error (such as controlled within ±10%) is calculated, the grid division density and the material parameter correction model are optimized; finally, the simulation calculation unit outputs the simulation report containing the PCB thermal deformation amount at each temperature stage, provides thermal deformation data support for the parameter optimization unit of the process simulation verification module, and assists in predicting the component offset trend caused by thermal deformation.

[0050] The suggestion generation unit is configured to generate temperature curve optimization parameters and patch pressure compensation values according to the simulation results. Specifically, first, the suggestion generation unit receives the thermal deformation simulation results (including the deformation amount of the PCB at each temperature stage, the offset data of the key pad position) from the simulation calculation unit, and at the same time, obtains the actual production data (such as historical temperature curve parameters, patch pressure values and associated production yield, offset defect rate) of the corresponding batch from the intelligent data platform module; then, the unit extracts key features from the simulation results: identifies the temperature interval corresponding to the thermal deformation peak (such as 220-250℃ with the maximum deformation), the deformation difference of different component regions (such as BGA chip region, 0201 component region), and the correlation between deformation and temperature change rate; then, the temperature curve optimization parameters are generated: based on the thermal deformation sensitive interval, adjust the key parameters of the reflow soldering temperature curve, such as reduce the heating rate in the preheating stage (from 3℃ / s to 2℃ / s) to reduce the thermal stress accumulation, shorten the duration of the peak temperature (260℃) (from 30s to 20s) to reduce the high temperature deformation, while ensuring the melting requirements of soldering; for the patch pressure compensation value, combined with the thermal deformation characteristics of different components (such as large size components are prone to assembly offset due to PCB deformation), generate differentiated compensation parameters according to component types: for components in areas with large thermal deformation (such as the edge of the PCB), increase the patch pressure compensation value (such as from 0.3N to 0.35N) to enhance the stability of assembly; for small components (such as 0201), set a lower compensation value (such as 0.25N) to avoid component damage; then, the unit corrects the preliminary parameters generated by comparing the optimal parameters under the same conditions in the history production (such as when the simulation deformation amount deviates from the actual deformation by more than 10%, introduce a correction coefficient), and verifies that the parameters are within the operating constraints of the equipment (such as the temperature curve adjustment does not exceed the temperature control capability of the reflow soldering furnace); finally, the suggestion generation unit outputs the final temperature curve optimization parameters (including heating rate, constant temperature time, peak temperature) and patch pressure compensation value (divided according to component types and PCB regions), which are transmitted to the reflow soldering furnace control module and the suction nozzle adjustment unit of the patch machine respectively, providing decision basis for dynamic optimization of process parameters.

[0051] In summary, the SMT patch offset correction system associated with AOI provided by the embodiment of the present application acquires pre-assembly offset data through the pre-furnace AOI detection module, acquires actual offset data after welding through the post-furnace AOI detection module, combines environmental parameters such as temperature and vibration collected by the multi-dimensional perception module, fuses and constructs a knowledge graph through the intelligent data platform module, and generates a standardized data set; then, the AI decision optimization module generates correction parameters based on the collaborative strategy of deep learning and reinforcement learning, adjusts the position of the patch machine and the parameters of the suction nozzle through the offset correction execution module to realize dynamic correction, and the process simulation verification module generates optimization suggestions by means of thermal deformation simulation and feeds back to the AI module to form a closed loop, so as to realize offset root tracing, dynamic compensation of environmental factors and adaptability improvement in different scenarios, and comprehensively improve the SMT patch precision and production yield.

[0052] Figure 2 This is a flowchart illustrating an SMT placement correction method linked with AOI according to an embodiment of the present invention. It should be noted that if substantially the same result is achieved, the method of this application is not necessarily identical. Figure 2 The illustrated process sequence is limited. For example... Figure 2 As shown: A method for SMT placement correction in conjunction with AOI, comprising the following steps: Step 1: Obtain pre-mount offset data, actual offset data after soldering, and environmental parameter data, and transmit them to the intelligent data platform module; Step 2: The intelligent data platform module receives pre-mount offset data, actual offset data, and environmental parameter data, establishes a unified data model, constructs a knowledge graph, and generates a standardized dataset. Step 3: The AI ​​decision optimization module receives the standardized dataset, trains the offset prediction model through the deep learning unit, and dynamically iterates and optimizes the parameter generation logic by combining the reinforcement learning unit. The policy output unit then outputs the corrected parameters. Step 4: The correction execution module receives the correction parameters, adjusts the X / Y axis displacement and rotation angle of the pick-and-place machine through the servo control unit, adjusts the nozzle height and pressure through the nozzle adjustment unit, and periodically verifies the motion accuracy through the parameter calibration unit. Step 5: The process simulation verification module constructs a thermal expansion coefficient model of PCB materials through the thermal deformation model unit. The simulation calculation unit performs thermal deformation simulation based on finite element analysis. The suggestion generation unit generates temperature curve optimization parameters and chip pressure compensation values ​​based on the simulation results, which are fed back to the AI ​​decision optimization module to form a closed loop.

[0053] In one embodiment, specifically in step three, when the AI ​​decision optimization module generates the correction parameters, it employs a collaborative strategy of deep learning and reinforcement learning, as follows: When there are ≤5 production batches, the weight of deep learning units accounts for 70% and the weight of reinforcement learning units accounts for 30%. When the production batch size exceeds 5 batches, the weight of the reinforcement learning unit is dynamically increased to 60%, and the dynamic iterative optimization adapts to the mounting characteristics of different component types.

[0054] In summary, the SMT patch deviation correction method linked with AOI provided by the embodiment of the present application acquires the pre-pasting deviation data, the actual deviation data after welding and the environmental parameter data and transmits them to the intelligent data hub module; the intelligent data hub module establishes a unified data model, constructs a knowledge graph and generates a standardized data set; the AI decision optimization module receives the data set, generates correction parameters by using a deep learning and reinforcement learning collaborative strategy (when the production batch is less than or equal to 5 batches, the deep learning accounts for 70% and the reinforcement learning accounts for 30%; when the production batch is more than 5 batches, the reinforcement learning weight is increased to 60%); the deviation correction execution module adjusts the patch machine displacement and the suction nozzle parameters according to the correction parameters and calibrates the precision regularly; the process simulation verification module generates a temperature curve and a pressure compensation value by thermal deformation simulation and feeds them back to the AI module to form a closed loop, realizes the full-process dynamic deviation correction and process parameter optimization, and improves the SMT patch precision and scene adaptability.

[0055] For other details of the technical solutions of each step in the SMT patch deviation correction method linked with AOI, refer to the description of the SMT patch deviation correction system linked with AOI in the above embodiment.

[0056] It should be noted that each embodiment in the present specification adopts a progressive description manner, and each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to. For method embodiments, since they are basically similar to system embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the system embodiment.

[0057] The basic principles of the present disclosure are described above in combination with specific embodiments, but it should be pointed out that the advantages, advantages, effects and the like mentioned in the present disclosure are only examples and are not limiting, and these advantages, advantages, effects and the like cannot be considered as the must-have of each embodiment of the present disclosure. In addition, the above specific details are only for the purpose of example and understanding, and are not limiting, and the above details do not limit the present disclosure to the must-use of the above specific details to realize.

[0058] In this disclosure, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. The block diagram of the devices, apparatus, equipment, systems referred to in this disclosure is merely illustrative and not intended to imply the necessity or arrangement of the connections, arrangement, configuration as shown in the block diagram. As will be appreciated by those skilled in the art, the devices, apparatus, equipment, systems can be connected, arranged, configured in any way. The words "comprising," "containing," "including," "having," and the like, are to be construed open-ended, meaning "including but not limited to," and are to be taken in their broadest context. The words "or" and "and" as used herein, mean "and / or," and are to be taken in their broadest context, unless the context clearly indicates otherwise. The word "comprising" as used herein, means "comprising but not limited to," and is to be taken in its broadest context.

[0059] Also, as used in this disclosure, "or" as used in the context of items A and B as "at least one of A or B" is intended to mean A or B or both A and B. Further, the phrase "example of" as used in this disclosure is intended to mean "example of, but not limited to, and should be taken in its broadest context.

[0060] It is also important to note that the systems and methods of the disclosure can be embodied in a variety of forms without departing from the teachings thereof. Such forms include components, steps, locations, and / or a combination thereof adapted in various ways as equivalents within the scope of the disclosure.

[0061] Various changes, modifications and alterations in the teachings and techniques described herein can be made without departing from the teachings of the technology defined by the appended claims. In addition, the scope of the following claims is not limited to the specific embodiments described herein. Rather, the scope of the following claims is intended to be coextensive with the scope of the claims as defined by the language of the claims and equivalents thereof. Accordingly, the disclosure is not to be seen as limited by the foregoing description but is only limited by the scope of the appended claims.

[0062] The above description of the disclosed aspects is given for clearness of understanding only. Various modifications can be made to the aspects described herein, and the general principles taught herein can be applied to other aspects without departing from the scope of the disclosure. Accordingly, the disclosure is not intended to be limited to the aspects described herein but is to be accorded the widest scope consistent with the scope of the claims as defined by the language of the claims and equivalents thereof.

[0063] The foregoing description has been presented for the purposes of illustration and description. Furthermore, the description is not intended to limit the embodiments of the disclosure to the forms disclosed herein. Although the various example aspects and embodiments have been described herein with regard to particular aspects and embodiments, those skilled in the art will recognize that certain modifications, changes, substitutions, additions and sub-combinations can be made without departing from the spirit of the disclosure.

Claims

1. An SMT patch deviation correction system linked with AOI, characterized in that, The furnace front AOI detection module, the furnace rear AOI detection module, the multi-dimensional perception module, the intelligent data middle station module, the AI decision optimization module, the deviation correction execution module and the process simulation verification module are comprised. The furnace front AOI detection module is used for acquiring pre-pasting offset data and transmitting the pre-pasting offset data to the intelligent data middle station module. The furnace rear AOI detection module is used for acquiring actual offset data after welding and transmitting the actual offset data to the intelligent data middle station module. The multi-dimensional perception module is used for collecting environmental parameter data in real time and transmitting the environmental parameter data to the intelligent data middle station module. The intelligent data middle station module is used for receiving the pre-pasting offset data, the actual offset data and the environmental parameter data, establishing a unified data model and constructing a knowledge graph, and generating a standardized data set. The AI decision optimization module is used for receiving the standardized data set of the intelligent data middle station module, generating correction parameters based on deep learning and reinforcement learning, and sending the correction parameters to the deviation correction execution module. The deviation correction execution module is used for receiving the correction parameters of the AI decision optimization module and adjusting position parameters and attitude parameters of a chip mounter according to the correction parameters. The process simulation verification module is used for generating optimization suggestions through thermal deformation simulation and feeding back the optimization suggestions to the AI decision optimization module.

2. The SMT patch deviation correction system linked with AOI according to claim 1, characterized in that: The intelligent data middle station module comprises a protocol conversion unit, a knowledge graph construction unit and a data storage unit. The protocol conversion unit is used for converting detection data in different formats into a unified format. The knowledge graph construction unit is used for associating mapping relationships of component models, process parameters and offset data. The data storage unit is used for storing a standardized data set and historical production data.

3. The SMT patch deviation correction system linked with AOI according to claim 1, characterized in that: The furnace front AOI detection module comprises a camera unit, a multi-spectral light source unit and an image preprocessing unit. The camera unit is used for collecting image data of pre-pasting components. The multi-spectral light source unit is used for providing red and white band light sources, which are respectively adapted to tin paste detection and component identification. The image preprocessing unit is used for performing feature enhancement processing on collected images.

4. The SMT patch deviation correction system linked with AOI according to claim 1, characterized in that: The furnace rear AOI detection module comprises a three-dimensional vision unit, a defect identification unit and a data output unit. The three-dimensional vision unit is used for scanning three-dimensional coordinate data of components after welding. The defect identification unit is used for identifying welding defects based on a preset feature library. The data output unit is used for transmitting actual offset data and defect information after integration.

5. The SMT patch deviation correction system linked with AOI according to claim 1, characterized in that: The multi-dimensional perception module comprises a temperature sensing unit, a pressure sensing unit and a vibration sensing unit. The temperature sensing unit is used for detecting real-time temperatures of a chip mounter motion shaft and a reflow soldering furnace. The pressure sensing unit is used for monitoring pressure values of a suction nozzle in contact with components. The vibration sensing unit is used for collecting vibration frequency and amplitude data when equipment is running.

6. The SMT patch deviation correction system linked with AOI according to claim 1, characterized in that: The AI decision optimization module comprises a deep learning unit, a reinforcement learning unit and a strategy output unit. The deep learning unit is used for training an offset prediction model based on historical data. The reinforcement learning unit is used for generating correction parameter generation logic through dynamic iteration optimization. The strategy output unit is used for converting correction parameters into executable control signals.

7. The SMT patch deviation correction system linked with AOI according to claim 1, characterized in that: The deviation correction execution module comprises a servo control unit, a suction nozzle adjustment unit and a parameter calibration unit; The servo control unit is configured to drive the motor to adjust the X / Y axis displacement and rotation angle of the chip mounter. The suction nozzle adjustment unit is configured to adjust the height and pressure of the suction nozzle according to the correction parameter. The parameter calibration unit is configured to periodically check the motion accuracy of the chip mounter.

8. The SMT patch deviation correction system linked with AOI according to claim 1, characterized in that: The process simulation verification module comprises a thermal deformation model unit, a simulation calculation unit and a suggestion generation unit; The thermal deformation model unit is configured to construct a thermal expansion coefficient model of the PCB material. The simulation calculation unit is configured to perform thermal deformation simulation based on the finite element analysis method. The suggestion generation unit is configured to generate temperature curve optimization parameters and patch pressure compensation values according to the simulation results.

9. A method for AOI-linked SMT patch rectification, used for executing the AOI-linked SMT patch rectification system of any one of claims 1-8, characterized in that, The method comprises the following steps: Step one, obtaining pre-mounting offset data, actual offset data after soldering and environmental parameter data, and transmitting them to the intelligent data platform module; Step two, the intelligent data platform module receives the pre-mounting offset data, actual offset data and environmental parameter data, establishes a unified data model and constructs a knowledge graph, and generates a standardized data set; Step three, the AI decision optimization module receives the standardized data set, trains the offset prediction model through the deep learning unit, dynamically iteratively optimizes the correction parameter generation logic through the reinforcement learning unit, and outputs the correction parameter through the strategy output unit; Step four, the deviation correction execution module receives the correction parameter, adjusts the X / Y axis displacement and rotation angle of the chip mounter through the servo control unit, adjusts the height and pressure of the suction nozzle through the suction nozzle adjustment unit, and periodically checks the motion accuracy through the parameter calibration unit; Step five, the process simulation verification module constructs a thermal expansion coefficient model of the PCB material through the thermal deformation model unit, performs thermal deformation simulation based on the finite element analysis through the simulation calculation unit, generates temperature curve optimization parameters and patch pressure compensation values according to the simulation results through the suggestion generation unit, and feeds back to the AI decision optimization module to form a closed loop.

10. The SMT patch deviation correction method linked with AOI according to claim 9, characterized in that: In step three, when the AI decision optimization module generates the correction parameter, a deep learning and reinforcement learning collaborative strategy is adopted, which is as follows: When the production batch is ≤5 batches, the weight proportion of the deep learning unit is 70% and the weight proportion of the reinforcement learning unit is 30%; When the production batch is >5 batches, the weight of the reinforcement learning unit is dynamically increased to 60%, which is adapted to the mounting characteristics of different components through dynamic iterative optimization.

Images