PCBA assembly process optimization method, device, equipment and storage medium

Abstract

This application provides a PCBA assembly process optimization method, apparatus, equipment, and storage medium. The method includes: building an assembly knowledge base, which includes: static knowledge of the assembly system, historical assembly optimization cases, and expert diagnostic rules; acquiring dynamic operating data from the assembly site, performing correlation analysis based on the dynamic operating data and the content of the assembly knowledge base, automatically identifying optimizable bottlenecks, and generating contextual optimization proposals; responding to the optimization proposals, creating a shared context integrating relevant information from the assembly knowledge base and real-time assembly status, invoking multiple intelligent agents to collaboratively perform process optimization analysis within the shared context, and generating an optimized assembly process flow; executing the optimized assembly process flow, and updating historical assembly optimization cases and process baselines in the assembly knowledge base based on the deviation between the execution results and expected indicators.

Description

Technical Field

[0001] This application relates to the field of PCBA technology, and in particular to a PCBA assembly process optimization method, apparatus, equipment and storage medium. Background Technology

[0002] PCBA assembly process optimization is a crucial means of improving production line efficiency in the electronics manufacturing industry. Currently, there are numerous time bottlenecks in the assembly process that can be optimized. Engineers typically need to manually review production reports, equipment operation logs, and operation videos from the Manufacturing Execution System (MES), reviewing each workstation and batch individually to pinpoint cycle time anomalies, and then manually calculate optimization potential and revise process documents. This process heavily relies on the engineer's personal experience, and because assembly site data is scattered across multiple heterogeneous systems, there is a lack of structured correlation between the data and optimization knowledge. This means that bottleneck identification, root cause analysis, optimization plan development, and effect verification must be carried out manually and sequentially, severely limiting the efficiency and accuracy of process optimization. Summary of the Invention

[0003] This application provides a PCBA assembly process optimization method, apparatus, equipment, and storage medium to improve the efficiency and accuracy of PCBA assembly process optimization.

[0004] In a first aspect, embodiments of this application provide a PCBA assembly process optimization method, the method comprising: Establish an assembly knowledge base, which includes: static knowledge of the assembly system, historical assembly optimization cases, and expert diagnostic rules; Acquire dynamic operational data from the assembly site, perform correlation analysis based on the dynamic operational data and the content of the assembly knowledge base, automatically identify optimizable bottlenecks, and generate contextual optimization proposals. In response to the optimization proposal, a shared context is created that integrates relevant information from the assembly knowledge base and real-time assembly status. Multiple agents are invoked to collaboratively perform process optimization analysis within the shared context to generate an optimized assembly process flow. The optimized assembly process is executed, and the historical assembly optimization cases and process baselines in the assembly knowledge base are updated based on the deviation between the execution results and the expected indicators.

[0005] Secondly, embodiments of this application provide a PCBA assembly process optimization apparatus, which includes: The knowledge construction module is used to build an assembly knowledge base, which includes: static knowledge of the assembly system, historical assembly optimization cases, and expert diagnostic rules. The optimization analysis module is used to acquire dynamic operation data from the assembly site, perform correlation analysis based on the dynamic operation data and the content of the assembly knowledge base, automatically identify optimizable bottlenecks, and generate contextual optimization proposals. The process optimization module is used to respond to the optimization proposal, create a shared context that integrates relevant information from the assembly knowledge base and real-time assembly status, call multiple agents to collaboratively perform process optimization analysis in the shared context, and generate an optimized assembly process flow. The baseline solidification module is used to execute the optimized assembly process flow and update the historical assembly optimization cases and process baselines in the assembly knowledge base according to the deviation between the execution results and the expected indicators.

[0006] Thirdly, embodiments of this application provide a production device, which includes a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program and, when executing the computer program, implement the PCBA assembly process optimization method as described in any of the embodiments of this application.

[0007] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to implement the PCBA assembly process optimization method as described in any of the embodiments of this application.

[0008] This application provides a PCBA assembly process optimization method, which includes: building an assembly knowledge base, which includes: static knowledge of the assembly system, historical assembly optimization cases, and expert diagnostic rules; acquiring dynamic operation data from the assembly site, performing correlation analysis based on the dynamic operation data and the content of the assembly knowledge base, automatically identifying optimizable bottlenecks, and generating contextual optimization proposals; responding to the optimization proposals, creating a shared context integrating relevant information from the assembly knowledge base and real-time assembly status, calling multiple intelligent agents to collaboratively perform process optimization analysis in the shared context, and generating an optimized assembly process flow; executing the optimized assembly process flow, and updating historical assembly optimization cases and process baselines in the assembly knowledge base based on the deviation between the execution results and expected indicators. In the above method, by constructing an assembly knowledge base that integrates static knowledge, historical cases, and diagnostic rules, the originally scattered experience and specifications are precipitated into a structured knowledge base. When dynamic operation data of the assembly site is acquired, it is analyzed in context with the knowledge base to achieve automatic bottleneck identification, replacing manual screening. After the bottleneck is identified, a shared context integrating knowledge base information and real-time status is created. Multiple agents collaborate to complete process optimization analysis and generate new process flows. The deviation of the execution results is used to update the cases and baselines in the knowledge base, so that the knowledge base can continuously evolve and improve the accuracy of subsequent optimization, thereby improving the efficiency and accuracy of PCBA assembly process optimization. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A schematic block diagram of a PCBA assembly process optimization system provided in this application embodiment; Figure 2 A schematic flowchart illustrating a PCBA assembly process optimization method provided in this application embodiment; Figure 3 A schematic flowchart illustrating the first collaborative method of intelligent agents provided in this application embodiment; Figure 4 A schematic flowchart illustrating the second type of intelligent agent collaboration method provided in this application embodiment; Figure 5 This is a schematic block diagram of a PCBA assembly process optimization device provided in an embodiment of this application. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described below with reference to the accompanying drawings.

[0012] The terms "first" and "second," etc., used in the specification, claims, and drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0013] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0014] It should be understood that in this application, "at least one (item)" means one or more, "more than one" means two or more, "at least two (items)" means two or three or more, and "and / or" is used to describe the relationship between related objects, indicating that there can be three relationships. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the related objects before and after are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0015] Please see Figure 1 , Figure 1 This is a schematic block diagram of a PCBA assembly process optimization system provided in an embodiment of this application. Figure 1 As shown, the PCBA assembly process optimization system includes: an assembly knowledge base, an intelligent agent server, and a process display terminal.

[0016] The assembly knowledge base is used to store static knowledge of the assembly system, historical assembly optimization cases, and expert diagnostic rules.

[0017] The intelligent agent server is used to acquire dynamic operation data from the assembly site, perform correlation analysis based on the dynamic operation data and the content of the assembly knowledge base to automatically generate context optimization proposals, create shared contexts and call multiple intelligent agents to collaboratively generate optimized assembly process flows, and update cases and baselines in the assembly knowledge base according to the deviation between process execution results and expected indicators.

[0018] The process demonstration terminal is used to show users contextual optimization proposals and optimized assembly process flows, and to receive user-inputted optimization strategy feedback.

[0019] Please see Figure 2 , Figure 2 This is a schematic flowchart illustrating a PCBA assembly process optimization method provided in an embodiment of this application. Figure 2 As shown, the specific steps of this PCBA assembly process optimization method include: S1-S4.

[0020] S1. Build an assembly knowledge base, which includes: static knowledge of the assembly system, historical assembly optimization cases, and expert diagnostic rules.

[0021] For example, an assembly knowledge base is constructed to optimize PCBA assembly processes. This knowledge base gathers three types of information: static knowledge of the assembly system, historical assembly optimization cases, and expert diagnostic rules. The static knowledge of the assembly system originates from the structured modeling of all elements of the production line. It covers the bill of materials at the product level and characteristic parameters such as package type, polarity, and temperature resistance of each component. At the process level, the standard operating procedures record standard working hours, operation sequence, and tool and fixture numbers at the step level. At the equipment level, it includes the model, capability, and maintenance cycle of pick-and-place machines, reflow ovens, wave soldering equipment, and automated optical inspection equipment. It also includes process constraint rules based on specifications such as IPC-A-610, such as acceptable soldering standards, operating requirements for electrostatic sensitive devices, and upper limits for the heating rate of high-temperature sensitive components. Building upon the static knowledge system, the company's accumulated assembly optimization project documents, production line anomaly analysis reports, and process change records are transformed into historical assembly optimization cases using a unified template. Each case systematically records four dimensions: the phenomenon description includes the cycle time data, defect rate, specific workstation and step name, the troubleshooting path records the detection methods used by engineers and the step-by-step elimination process, the key root cause is identified as the deep-seated reasons such as material pin oxidation leading to additional washing actions or unstable feeder feeding causing frequent material rejection, and the implementation effect includes the comparison of working hours before and after optimization and the change in first-pass yield. Simultaneously, the implicit diagnostic experience formed by senior process engineers in long-term practice is refined into expert diagnostic rules, expressed in a condition-triggered logic, such as "when the average actual working hours of the same workstation for three consecutive shifts exceed 130% of the standard working hours and involves the feeder, prioritize checking the feeder feed accuracy and material packaging consistency," or "when AOI detects bridging defects concentrated on specific pins of a certain batch of components after reflow soldering, check the coplanarity of the pins and the pad design dimensions of that batch." The above three types of knowledge are respectively stored in the system's static knowledge base, historical case base, and expert rule base, and cross-database association indexes are established based on unified product codes, workstation codes, and equipment codes to ensure that related information can be quickly retrieved according to context in subsequent steps.

[0022] S2. Acquire dynamic operation data from the assembly site, perform correlation analysis based on the dynamic operation data and the content of the assembly knowledge base, automatically identify optimizable bottlenecks, and generate contextual optimization proposals.

[0023] For example, on the production line side, dynamic operational data of the PCBA assembly site is continuously acquired through a data acquisition gateway or by utilizing the external data interfaces of the Manufacturing Execution System, equipment controllers, and vision inspection systems. The acquired data includes the actual cycle time of each workstation, accurate to the step-by-step operation cycle; comprehensive efficiency indicators of core equipment such as pick-and-place machines and insertion machines, along with their utilization rate, performance efficiency, and quality yield breakdown items; time-series records of material rejection rate, nozzle blockage alarms, and feeder jamming in equipment alarm and abnormal event logs; defect types, defect locations, and corresponding component reference numbers from automatic optical inspection and online testing outputs; and operator hand movement videos captured by workstation cameras. After acquiring the above dynamic operational data, a monitoring agent is invoked to perform automated correlation analysis with the assembly knowledge base. This analysis extracts cycle time, equipment status, and quality indicators from the dynamic operational data and compares them item by item with the standard working hours, equipment capability baseline, and process specification qualification criteria in the system's static knowledge base. When any indicator deviates from the baseline by more than a preset threshold, the workstation or equipment is marked as an anomaly to be confirmed. Extract a full-dimensional contextual snapshot within the time window of the anomaly, including the product model and batch produced during that period, the batch number of the materials used, operator information, equipment parameter settings, and the cycle time performance of adjacent workstations. Perform pattern matching between the contextual snapshot and trigger conditions in the expert rule base. Simultaneously, use this snapshot as a search criterion to retrieve existing cases from the historical case base with similarity exceeding a threshold. Once an expert rule is matched or a historical case is successfully matched, the anomaly is confirmed as an optimizable bottleneck, and a contextual optimization proposal is automatically generated. This proposal records the basic information of the bottleneck workstation or equipment, the specific value and magnitude of the current deviation indicator, a summary of the matched expert rule or a summary of the matched historical case, and a preliminary estimate of the potential optimization space based on historical case performance data. This replaces the inefficient process of engineers manually reviewing reports and conducting workstation-by-workstation debriefings.

[0024] S3. In response to the optimization proposal, create a shared context that integrates relevant information from the assembly knowledge base and real-time assembly status, and call multiple agents to collaboratively perform process optimization analysis in the shared context to generate an optimized assembly process flow.

[0025] For example, once a context optimization proposal is confirmed and triggered, the orchestration agent instantiates a shared context environment. This environment serves as an information space that multiple agents can access concurrently or sequentially, internally maintaining a unified data structure and state records. The orchestration agent performs context initialization, using the product model and workstation code involved in the optimization proposal as indexes. It retrieves relevant process constraints and historical case details from the assembly knowledge base. The process constraints include hard regulations such as welding temperature zone settings, electrostatic protection requirements, and upper limits of insertion force that must be followed for the workstation steps. Simultaneously, it extracts the real-time assembly status corresponding to the optimization proposal from monitoring data, covering the current actual cycle sequence, equipment parameter setting snapshots, and operation video frame sequences. The retrieved knowledge base information and real-time assembly status are then injected into the shared context environment. After completing context initialization, the orchestration agent calls upon the action analysis agent, line balancing agent, and compliance verification agent to enter the shared context environment to collaboratively perform process optimization analysis. The motion analysis agent reads the operation video frame sequence and standard time, analyzes the operator's hand movements using a predetermined motion time standard method, identifies optimization points such as missing parallel operations, redundant movement, and excessively long material picking paths, and outputs the revised motion sequence and its predicted time. The line balancing agent reads the cycle time data of each station on the current production line and equipment capacity constraints, receives the revised motion sequence output by the motion analysis agent, recalculates the station task allocation, merges or splits the steps with reduced predicted time with other adjacent steps, and calculates the adjusted overall line balance rate. The compliance verification agent monitors the output of the motion analysis agent and the line balancing agent in real time based on the process specification constraints in the shared context, verifies whether they violate hard constraints such as the upper limit of component pin bending angle and specific welding temperature curves, and writes constraint alarm information to the shared context when violations are found, triggering the agent that generated the violation output to perform a rollback adjustment. In a shared context, multiple agents continuously read, write, and interact with information to gradually converge and generate an optimized assembly process that simultaneously meets the cycle time optimization objective and process compliance requirements. Specifically, this process includes a work step re-division table, standard working hours for each work step, action specification descriptions for key operations, and equipment parameter settings.

[0026] S4. Execute the optimized assembly process flow, and update the historical assembly optimization cases and process baselines in the assembly knowledge base based on the deviation between the execution results and the expected indicators.

[0027] For example, the optimized assembly process flow is distributed to the production line manufacturing execution system to replace the original process documents, executed in the actual production environment, and the execution result data is continuously collected within a preset observation period. The execution result data covers two dimensions: actual working time and quality data. The actual working time data is the statistical value of the actual time consumed by each step after several batches of continuous production at each workstation. The quality data includes the first-pass yield of automatic optical inspection, the yield of online testing, and the number of defects in visual inspection or functional testing for the corresponding batch. The learning agent compares the execution result data with the expected indicators in the optimized assembly process flow. The expected indicators include the predicted working time output by the action analysis agent and the expected quality indicators confirmed by the compliance verification agent. The comparison process calculates the deviation rate between the actual working time and the predicted working time, as well as the difference between the actual quality indicators and the expected quality indicators. When the deviation rate or difference falls within a preset acceptable range, the learning agent generates a new historical assembly optimization case using a unified case template, writes it into the historical case library, and updates the corresponding process baseline in the knowledge base based on verified actual working hour data, using an exponentially weighted moving average or Bayesian update method, making the benchmark standard working hours closer to actual production capacity. When the deviation rate or difference exceeds the preset acceptable range, the learning agent marks the optimization as unsatisfactory, generates a deviation analysis report, and records the deviation exceeding the limit, the deviation value, and possible related factors extracted from the execution result data. This report is also written into the historical case library as a case with the "unsatisfactory effect" label to expand the abnormal pattern coverage of the knowledge base. Thus, each optimization execution and feedback is transformed into an incremental update of the assembly knowledge base, continuously enriching historical cases and constantly bringing the process baseline closer to the true optimal value, forming a self-evolving assembly context closed loop.

[0028] This application provides a PCBA assembly process optimization method, which includes: building an assembly knowledge base, which includes: static knowledge of the assembly system, historical assembly optimization cases, and expert diagnostic rules; acquiring dynamic operation data from the assembly site, performing correlation analysis based on the dynamic operation data and the content of the assembly knowledge base, automatically identifying optimizable bottlenecks, and generating contextual optimization proposals; responding to the optimization proposals, creating a shared context integrating relevant information from the assembly knowledge base and real-time assembly status, calling multiple intelligent agents to collaboratively perform process optimization analysis in the shared context, and generating an optimized assembly process flow; executing the optimized assembly process flow, and updating historical assembly optimization cases and process baselines in the assembly knowledge base based on the deviation between the execution results and expected indicators. In the above method, by constructing an assembly knowledge base that integrates static knowledge, historical cases, and diagnostic rules, the originally scattered experience and specifications are precipitated into a structured knowledge base. When dynamic operation data of the assembly site is acquired, it is analyzed in context with the knowledge base to achieve automatic bottleneck identification, replacing manual screening. After the bottleneck is identified, a shared context integrating knowledge base information and real-time status is created. Multiple agents collaborate to complete process optimization analysis and generate new process flows. The deviation of the execution results is used to update the cases and baselines in the knowledge base, so that the knowledge base can continuously evolve and improve the accuracy of subsequent optimization, thereby improving the efficiency and accuracy of PCBA assembly process optimization.

[0029] To more clearly illustrate the technical solution of this application, the technical solution of this application will be described below through specific embodiments. It should be noted that the specific embodiments are used to expand the description of the technical solution of this application, and are not intended to limit this application.

[0030] In some embodiments, correlation analysis is performed based on dynamic operating data and the content of the assembly knowledge base to automatically identify optimizable bottlenecks and generate contextual optimization proposals. This includes: invoking at least one first agent to perform contextual pattern matching between dynamic operating data and process baselines and diagnostic rules in the assembly knowledge base, identifying abnormal operating conditions that deviate from the process baseline, extracting equipment status, operation sequences, and material information associated with the abnormal operating conditions, and combining them to generate contextual optimization proposals. A shared context integrating relevant information from the assembly knowledge base and real-time assembly status is created, and multiple agents are invoked to collaboratively perform process optimization analysis within the shared context to generate an optimized assembly process flow. This includes: instantiating a shared context environment, injecting the optimization proposal, constraints and historical cases retrieved from the assembly knowledge base, and real-time assembly status into the shared context environment; invoking multiple second agents to execute at least one type of process optimization task within the shared context environment, and generating the optimized assembly process flow through information interaction within the shared context environment.

[0031] For example, in the process of automatically identifying optimizable bottlenecks and generating contextual optimization proposals through correlation analysis based on dynamic operating data and the content of the assembly knowledge base, at least one first agent deployed on the production line continuously consumes step time data from the manufacturing execution system, operating status logs from the equipment controller, and operation video streams from the vision inspection system. The first agent has a built-in process baseline loaded from the system's static knowledge base. This process baseline specifically includes the theoretical time range of each workstation under standard operating conditions, the stable operating range of the equipment's overall efficiency, and the qualified benchmark values ​​of quality indicators such as rejection rate and first-pass yield. The first agent will use the actual cycle time, equipment overall efficiency decomposition value, and defect rate from the acquired dynamic operating data. Detailed real-time indicators are compared item by item with the process baseline. During the comparison process, conditional trigger logic from the expert rule base is introduced simultaneously. When the number of continuous feeding miss alarms at a certain placement station exceeds three within fifteen minutes and the actual cycle time of that station deviates from the theoretical cycle time by more than 20%, an abnormal working condition is triggered. After the abnormal working condition is marked, the first intelligent agent expands a preset time window forward and backward from the time of occurrence of the abnormal working condition. It extracts all the relevant data within the window from the dynamic operation data, including the operation video frame sequence of that time period, equipment parameter setting records, batch codes of the materials used, cycle time fluctuations of adjacent stations before and after, and the unique identifier of the products in production during that time period. The first intelligent agent compares the extracted relevant data with the abnormal condition. The working conditions are packaged together to form a structured context snapshot, and logically combined with the summaries of diagnostic rules already matched in the expert rule base and similar cases retrieved from the historical case base to output a context optimization proposal. This proposal carries an abnormal workstation identifier, a quantitative comparison of the current value of the deviation indicator with the baseline value, a summary of the matched diagnostic rules, an overview of the root causes and effects of the matched similar historical cases, and a value of the expected optimization space calculated based on the average optimization magnitude of historical cases. In the process of creating a shared context integrating relevant information from the assembly knowledge base and real-time assembly status, and calling multiple intelligent agents to collaboratively perform process optimization analysis to generate the optimized assembly process flow, the orchestration agent instantiates a process with read-write consistency guarantees. The shared context environment synchronously injects the aforementioned context optimization proposal, all hard constraints and soft suggestion constraints retrieved from the constraint subset of the assembly knowledge base by product model and workstation index, the case details of the top few cases with similarity to the current optimization proposal extracted from the historical case library, and the assembly snapshot containing operation video frame sequences and real-time equipment status obtained from the first agent into the shared context environment. The orchestration agent schedules multiple second agents to execute their assigned process optimization tasks in parallel or sequentially within the shared context environment. The multiple second agents gradually converge through continuous reading, writing and interaction of information in the shared context environment, generating an optimized assembly process flow that simultaneously meets the cycle time optimization target and process compliance requirements.

[0032] In some embodiments, the plurality of second intelligent agents include a motion analysis intelligent agent, a line balancing intelligent agent, and a compliance verification intelligent agent; the motion analysis intelligent agent is configured to identify optimizable actions in an operation sequence and output predicted working hours based on operation videos and standard working hours; the line balancing intelligent agent is configured to adjust workstation task allocation and calculate balance rate based on production line cycle time and equipment capacity constraints; and the compliance verification intelligent agent is configured to verify the compliance of the generated assembly process flow based on assembly process specifications.

[0033] For example, the multiple second intelligent agents specifically include an action analysis agent, a line balancing agent, and a compliance verification agent. The action analysis agent reads the operation video frame sequence and standard time benchmark in the shared context, and decomposes the operator's action units such as material picking, positioning, insertion, and inspection item by item using a predetermined action time standard method. It identifies non-value-added actions such as single-handed waiting, repetitive picking and placing, and invalid movement, and eliminates or compresses the above non-value-added actions by rearranging the action sequence. It outputs an action sequence containing the revised standard time of each action unit and the overall predicted time. The line balancing agent reads the current cycle time distribution data of each workstation in the entire line, the processing capacity envelope of each device, and the revised action sequence output by the action analysis agent in the shared context. It constructs a workstation task reorganization model with the optimization goal of minimizing the cycle time of the bottleneck workstation. It performs a feasibility assessment of merging the steps with the reduced predicted time after revision with adjacent steps, or performs secondary splitting of excessively long steps, and outputs the adjusted workstation task allocation scheme and The corresponding overall line balancing rate calculation results; the compliance verification agent reads the constraints in the assembly process specification library in the shared context, and conducts a compliance review on each item of the revised action sequence output by the action analysis agent and the workstation task allocation scheme output by the line balancing agent. The review items cover whether the component insertion force exceeds the specification limit, whether the exposure time of the heat-sensitive device before reflow soldering exceeds the allowable window, whether the operation of the electrostatic sensitive device retains the specified electrostatic discharge action, and whether the soldering temperature curve conforms to the temperature zone setting of the product process specification. During the review process, the compliance verification agent writes any violations found into the shared context in the form of structured alarm information. The alarm information clearly marks the location of the violation, the specific specification clause violated, and the suggested correction direction. This triggers the corresponding second agent to read the alarm information in the shared context and adjust its output. After multiple rounds of reading, writing, and interaction, the three agents collaboratively converge in the shared context environment to generate a compliant and optimized assembly process flow.

[0034] In some embodiments, the assembly knowledge base includes a system static knowledge base, a historical case base, and an expert rule base. The method further includes selectively enabling at least one of the system static knowledge base, the historical case base, and the expert rule base, based on the current optimization task type, to provide contextual information.

[0035] For example, the assembly knowledge base is logically divided into three independent storage units: a system static knowledge base, a historical case base, and an expert rule base. These three units are linked across databases using a unified product code, workstation code, and equipment code. The system static knowledge base stores the component list, standard process flow, and standard time baseline for each process step for each product, using the product code as the primary key. The historical case base stores complete files of past optimization cases using workstation codes and exception types as a combined index. Each case includes four dimensions: phenomenon description, troubleshooting path, key root cause, and implementation effect. The expert rule base stores conditional diagnostic rules derived from the experience of senior process engineers, using trigger conditions as the key. In practice, the monitoring agent selectively activates at least one of the three knowledge bases mentioned above to provide contextual information based on the classification of the currently labeled optimizable bottleneck: when the bottleneck manifests as an abnormal cycle time at a single workstation, the monitoring agent activates the standard time baseline for that workstation in the system's static knowledge base and a subset of diagnostic rules matching the equipment type at that workstation in the expert rule base; when the bottleneck manifests as a quality defect across workstations, the monitoring agent activates the full-process constraints of related products in the system's static knowledge base and a set of cases with the same defect pattern in the historical case base; when the bottleneck manifests as a systemic decline in overall equipment efficiency, the monitoring agent prioritizes activating diagnostic rules related to equipment maintenance in the expert rule base and associates them with the equipment capability envelope in the system's static knowledge base and similar equipment failure cases in the historical case base. This on-demand selective activation method reduces the interference of irrelevant knowledge on correlation analysis and improves the accuracy and efficiency of context matching.

[0036] In some embodiments, dynamic operational data is acquired in real time from at least one of the manufacturing execution system, equipment controller, and vision inspection system via a model context protocol. The dynamic operational data includes at least one of the following: actual cycle time of the workstation, overall equipment efficiency, material rejection rate, operation cycle time, and operation video.

[0037] For example, computing nodes deployed at the edge of the production line establish data channels with the Manufacturing Execution System (MES), the equipment programmable logic controller (PLC), and the vision inspection system through the standard interface of the Model Context Protocol (MTP) to acquire dynamic operational data of the PCBA assembly site in real time. From the MES, they acquire the actual operation cycle time, work-in-process quantity, and reported abnormal events for each product at each workstation. From the equipment controller, they acquire real-time placement speed, ejection statistics, nozzle usage counts, and feeder status of the pick-and-place machine; the actual temperature and chain speed of each zone of the reflow soldering equipment; the flux spraying amount and solder wave height of the wave soldering equipment; and the inspection cycle time and first-pass yield of the automated optical inspection equipment. From the vision inspection system, they acquire continuous video frame data from each operating station, with the video frame data carrying timestamps and station identifiers, supporting traceability by product serial number. This dynamic operational data is aggregated to the edge computing nodes in a unified semantic format via the MTP, allowing the monitoring agent to consume it instantly. The data path is standardized and has low latency, ensuring the real-time performance and accuracy of bottleneck identification.

[0038] In some embodiments, based on the deviation between the execution result and the expected indicators, the historical assembly optimization cases and process baselines in the assembly knowledge base are updated, including: obtaining the actual working hours and quality data of the optimized assembly process flow, comparing them with the predicted working hours and expected quality indicators of the optimized assembly process flow, and obtaining the deviation value; when the deviation value exceeds a preset threshold, a new case containing the current optimization process, deviation analysis and root cause is generated, and the new case is written into the historical assembly optimization case library, while the process baseline of the corresponding step is updated based on the actual working hours.

[0039] For example, in the process of updating historical assembly optimization cases and process baselines in the assembly knowledge base based on the deviation between the execution results and expected indicators, when the optimized assembly process reaches the preset observation period in actual operation on the production line, the learning agent retrieves the actual working hours records of each relevant workstation within that period from the manufacturing execution system, and retrieves the automatic optical inspection pass rate, online test yield, functional test defect rate, and visual inspection defect distribution data of the corresponding batch of products from the quality management system to form an execution result dataset. The learning agent compares the actual working hours in the execution result dataset with the predicted working hours output by the action analysis agent in the optimization stage step by step to calculate the relative deviation rate. At the same time, it calculates the difference between the actual quality indicators and the expected quality indicators confirmed by the compliance verification agent in the optimization stage. The learning agent is internally configured with preset acceptable deviation thresholds: when the time deviation rate and quality index difference of each step fall within the threshold range, the optimization is deemed effective. The learning agent calls the writing interface of the assembly knowledge base to encapsulate the entire chain information of this optimization, from bottleneck identification, proposal generation, collaborative analysis to execution verification, into a new case according to the template of historical assembly optimization cases and stores it in the historical case library. The original standard time baseline of this step in the system's static knowledge base is replaced with the exponentially weighted moving average of the actual time, completing the adaptive calibration of the process baseline. When any deviation exceeds the threshold, the learning agent determines that the optimization has not met expectations and generates a deviation analysis report. The report lists the step number of the deviation exceeding the limit, the specific value and direction of the deviation, the possible interference factors extracted from the execution result data, and the conclusions of the exclusionary analysis. This deviation analysis report, along with the entire chain information of this optimization, is encapsulated into a negative sample case with the label of "not meeting expectations" and written into the historical case library. The accumulation of negative sample cases makes the historical case library more complete in covering abnormal patterns, providing a reverse reference for case matching and risk prediction in subsequent optimizations.

[0040] In some embodiments, invoking multiple agents to collaboratively perform process optimization analysis in a shared context to generate an optimized assembly process flow includes: during the collaborative process of multiple agents, at least one process constraint defined or triggered by at least one first agent is passed to at least one second agent through the shared context, so that the second agent can make decisions based on the received process constraint in its process optimization analysis and generate an optimization result that conforms to the process constraint.

[0041] In one embodiment, the first agent is a compliance verification agent, and the second agent is a motion analysis agent or a line balancing agent. For example... Figure 3 As shown, the collaborative process of multiple agents includes: S31-S32.

[0042] S31. The compliance verification agent defines at least one hard process constraint in the shared context and writes the hard process constraint into the shared context in the form of a structured constraint label.

[0043] S32. After the action analysis agent or line balancing agent reads the constraint label in the shared context, it treats the constraint as an inviolable decision boundary in its process optimization analysis and generates optimization results only in the feasible space outside the constraint boundary.

[0044] For example, in the process of multiple agents collaborating on process optimization analysis within a shared context, the compliance verification agent, acting as the first agent, retrieves all hard process constraints associated with the product and workstation from the process specification subset of the assembly knowledge base after reading the product model and workstation information in the shared context. These hard process constraints include, but are not limited to, the upper limit of the reflow soldering heating rate for thermistors, the retention requirement for the electrostatic discharge action of electrostatic sensitive devices, the allowable range of bending angles for specific component pins, and the lower limit threshold of flux spraying amount during wave soldering. The compliance verification agent encapsulates each of these hard process constraints into structured constraint tags. Each constraint tag includes a constraint number, constraint type, applicable work step range, quantified boundary values ​​of constraint parameters, and violation consequence level. These constraint tags are then injected into the global constraint area of ​​the shared context environment through the write interface of the shared context. As one of the second agents, the action analysis agent actively reads constraint labels from the global constraint region when performing operation sequence optimization in the shared context. During the arrangement of operations using a predetermined action time standard method, it identifies operation steps governed by constraint labels and uses the action retention requirements or time limits specified in the constraint labels as non-deletable and non-compressible decision boundaries. It only applies action simplification and rearrangement optimization to non-constrained steps, and the output revised action sequence automatically inherits and fully retains all operation elements marked by constraint labels. The line balancing agent, another second agent, also reads constraint labels from the global constraint region when performing workstation task reorganization in the shared context. During the construction of the workstation task reorganization model, steps with constraint labels are set as fixed nodes that cannot be split or merged into other workstations. The solver searches for a workstation allocation scheme that minimizes the bottleneck workstation's cycle time within the feasible region defined by the constraint labels, and the output workstation task allocation scheme strictly satisfies the boundary conditions of all constraint labels. Through the injection and reading of the aforementioned constraint labels in the shared context, the optimization behavior of the second agent is regulated by the process constraints defined by the first agent, generating optimization results that conform to all process constraints.

[0045] In another embodiment, the first agent is a line-balancing agent, and the second agent is an action analysis agent. For example... Figure 4 As shown, the collaborative process of multiple agents includes: S33-S34.

[0046] S33. The line balancing agent triggers a beat constraint condition based on the whole line beat target in the shared context, and writes the beat constraint condition into the shared context in the form of the target beat value.

[0047] S34. After reading the target beat value in the shared context, the action analysis agent makes action compression decisions in its action sequence optimization to meet the target beat value and generate an optimized action sequence that meets the beat constraint.

[0048] For example, in the process of multiple agents collaborating on process optimization analysis within a shared context, the line balancing agent, acting as the first agent, reads the actual cycle time distribution data of each workstation on the entire production line and the delivery cycle time requirements of product orders within the shared context. With the optimization direction of eliminating production line bottlenecks, it calculates the target cycle time value required for each workstation to achieve overall line balance. The line balancing agent writes the calculated target cycle time value for each workstation into the shared context in the form of structured cycle time constraints. Each cycle time constraint includes a workstation identifier, a target cycle time upper limit, and a corresponding priority weight. The action analysis agent, acting as the second agent, reads the operation video frame sequence and standard time baseline within the shared context, while simultaneously reading the target cycle time value written by the line balancing agent for its analyzed workstation. During the process of decomposing operational actions item by item using a predetermined action time standard method, the action analysis agent uses the target beat value as the upper limit of the total time for action sequence optimization. When performing optimization operations such as action merging, order rearrangement, or action elimination, it continuously checks whether the total predicted time of the current revised action sequence has converged to within the target beat value. If the total predicted time still exceeds the target beat value, it continues to apply progressive compression to non-core actions. When the total predicted time meets the target beat value, it stops further compression to retain the necessary operational margin. The output revised action sequence conforms to the beat constraints defined by the line balancing agent while satisfying operational integrity. After completing the optimization, the action analysis agent writes the revised action sequence and its predicted time back to the shared context. The line balancing agent re-checks the overall line balance state based on the predicted time. If bottleneck drift still exists, it updates the target beat value and triggers the action analysis agent to perform the next round of constraint-driven iterative optimization. Through this constraint-driven multi-round collaboration, multiple agents generate optimization results that simultaneously meet the requirements of overall line beat balance and action-level executability.

[0049] Please see Figure 5 , Figure 5 This is a schematic block diagram of a PCBA assembly process optimization apparatus 200 provided in an embodiment of this application. The PCBA assembly process optimization apparatus 200 is used to execute the aforementioned PCBA assembly process optimization method. The PCBA assembly process optimization apparatus 200 can be configured in a server.

[0050] The server can be a standalone server, a server cluster, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0051] like Figure 5 As shown, the PCBA assembly process optimization device 200 includes: a knowledge construction module 201, an optimization analysis module 202, a process optimization module 203, and a baseline solidification module 204.

[0052] The knowledge construction module 201 is used to build an assembly knowledge base, which includes: static knowledge of the assembly system, historical assembly optimization cases, and expert diagnostic rules.

[0053] The optimization analysis module 202 is used to acquire dynamic operation data from the assembly site, perform correlation analysis based on the dynamic operation data and the content of the assembly knowledge base, automatically identify optimizable bottlenecks, and generate contextual optimization proposals.

[0054] The process optimization module 203 is used to respond to optimization proposals, create a shared context that integrates relevant information from the assembly knowledge base and real-time assembly status, call multiple agents to collaboratively perform process optimization analysis in the shared context, and generate an optimized assembly process flow.

[0055] The baseline solidification module 204 is used to execute the optimized assembly process flow and update the historical assembly optimization cases and process baselines in the assembly knowledge base based on the deviation between the execution results and the expected indicators.

[0056] This application provides a production device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and, when executing the computer program, implement the PCBA assembly process optimization method as described in any of the embodiments of this application.

[0057] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it causes the processor to implement a PCBA assembly process optimization method as described in any of the embodiments of this application.

[0058] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A PCBA assembly process optimization method, characterized in that, The method includes: Establish an assembly knowledge base, which includes: static knowledge of the assembly system, historical assembly optimization cases, and expert diagnostic rules; Acquire dynamic operational data from the assembly site, perform correlation analysis based on the dynamic operational data and the content of the assembly knowledge base, automatically identify optimizable bottlenecks, and generate contextual optimization proposals. In response to the optimization proposal, a shared context is created that integrates relevant information from the assembly knowledge base and real-time assembly status. Multiple agents are invoked to collaboratively perform process optimization analysis within the shared context to generate an optimized assembly process flow. The optimized assembly process is executed, and the historical assembly optimization cases and process baselines in the assembly knowledge base are updated based on the deviation between the execution results and the expected indicators.

2. The PCBA assembly process optimization method according to claim 1, characterized in that, The process of performing correlation analysis based on the dynamic operating data and the content of the assembly knowledge base to automatically identify optimizable bottlenecks and generate contextual optimization proposals includes: At least one first intelligent agent is invoked to perform context pattern matching between the dynamic operation data and the process baseline and diagnostic rules in the assembly knowledge base, identify abnormal operating conditions that deviate from the process baseline, extract the equipment status, operation sequence and material information associated with the abnormal operating conditions, and combine them to generate the context optimization proposal. The process of creating a shared context integrating relevant information from the assembly knowledge base and real-time assembly status, and then invoking multiple agents to collaboratively perform process optimization analysis within that shared context to generate an optimized assembly process flow includes: Instantiate a shared context environment, and simultaneously inject the optimization proposal, constraints and historical cases retrieved from the assembly knowledge base, and the real-time assembly status into the shared context environment; invoke multiple second agents to execute at least one type of process optimization task in the shared context environment, and generate an optimized assembly process flow through information interaction in the shared context environment.

3. The PCBA assembly process optimization method according to claim 2, characterized in that, The plurality of second intelligent agents include an action analysis intelligent agent, a line balancing intelligent agent, and a compliance verification intelligent agent; The action analysis agent is configured to identify optimizable actions in an operation sequence and output predicted working hours based on operation videos and standard working hours. The line balancing agent is configured to adjust the workstation task allocation and calculate the balance rate based on production line cycle time and equipment capacity constraints. The compliance verification agent is configured to verify the compliance of the generated assembly process flow based on the assembly process specifications.

4. The PCBA assembly process optimization method of claim 1, wherein, The assembly knowledge base includes a system static knowledge base, a historical case base, and an expert rule base. The method also includes: Depending on the current optimization task type, at least one of the system static knowledge base, the historical case base, and the expert rule base may be selectively enabled to provide contextual information.

5. The PCBA assembly process optimization method of claim 1, wherein, The dynamic operation data is acquired in real time from at least one of the manufacturing execution system, equipment controller, and vision inspection system through the model context protocol. The dynamic operation data includes at least one of the following: actual cycle time of the workstation, overall equipment efficiency, material rejection rate, operation cycle time, and operation video.

6. The PCBA assembly process optimization method of claim 1, wherein, The step of updating the historical assembly optimization cases and process baselines in the assembly knowledge base based on the deviation between the execution results and the expected indicators includes: The actual working hours and quality data of the optimized assembly process are obtained and compared with the predicted working hours and expected quality indicators of the optimized assembly process to obtain the deviation value. When the deviation value exceeds the preset threshold, a new case is generated that includes the current optimization process, deviation analysis and root cause, and the new case is written into the historical assembly optimization case library. At the same time, the process baseline of the corresponding step is updated based on the actual working hours.

7. The PCBA assembly process optimization method of claim 1, wherein, The process of invoking multiple intelligent agents to collaboratively perform process optimization analysis within the shared context and generate an optimized assembly process flow includes: During the collaborative process of the multiple agents, at least one process constraint defined or triggered by at least one first agent is passed to at least one second agent through the shared context, so that the second agent can make decisions based on the received process constraint in its process optimization analysis and generate optimization results that conform to the process constraint.

8. A PCBA assembly process optimization apparatus, characterized in that, The PCBA assembly process optimization device is used to execute the PCBA assembly process optimization method as described in any one of claims 1-7, and the PCBA assembly process optimization device includes: The knowledge construction module is used to build an assembly knowledge base, which includes: static knowledge of the assembly system, historical assembly optimization cases, and expert diagnostic rules. The optimization analysis module is used to acquire dynamic operation data from the assembly site, perform correlation analysis based on the dynamic operation data and the content of the assembly knowledge base, automatically identify optimizable bottlenecks, and generate contextual optimization proposals. The process optimization module is used to respond to the optimization proposal, create a shared context that integrates relevant information from the assembly knowledge base and real-time assembly status, call multiple agents to collaboratively perform process optimization analysis in the shared context, and generate an optimized assembly process flow. The baseline solidification module is used to execute the optimized assembly process flow and update the historical assembly optimization cases and process baselines in the assembly knowledge base according to the deviation between the execution results and the expected indicators.

9. A production apparatus characterized by comprising: The production equipment includes a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program and, in executing the computer program, implement the PCBA assembly process optimization method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to implement the PCBA assembly process optimization method as described in any one of claims 1 to 7.

Images