A printed circuit board production process optimization system based on digital twinning

Abstract

The application discloses a kind of based on digital twinning printed circuit board production process optimization system, specifically related to industrial digital twinning field, including production data acquisition module, process data governance module, digital twinning construction module, virtual-real synchronous module, process state evaluation module, quality prediction module, process optimization module, production execution cooperation module and closed-loop update module;Through acquisition equipment, process, material, environment, quality and production task data, construct and real-time update digital twinning model, evaluate current process state, and predict the quality result under different parameter combinations;Under the constraint of equipment capacity, process specification, quality requirement and production plan, generate candidate process scheme, determine target scheme and issue execution through virtual simulation and comprehensive evaluation, while continuously correcting digital twinning model and optimization strategy according to actual production result, realize the dynamic closed-loop optimization of printed circuit board production process.

Description

Technical Field

[0001] This invention relates to the field of industrial digital twin technology, and more specifically, to a printed circuit board manufacturing process optimization system based on digital twins. Background Technology

[0002] Printed circuit board (PCB) production involves multiple interconnected processes such as material preparation, lamination, drilling, electroplating, etching, solder masking, and testing. Changes in upstream process parameters can accumulate in subsequent stages, ultimately leading to quality issues such as linewidth deviations, abnormal via copper, interlayer misalignment, warping, and open / short circuits. With the development of high-density interconnects, multilayer circuits, and high-frequency / high-speed PCBs, the requirements for linewidth, line spacing, via diameter, alignment accuracy, and copper thickness uniformity are constantly increasing. Traditional methods of parameter control relying on fixed process formulas and manual experience are no longer sufficient to adapt to process fluctuations under different products, equipment conditions, and production environments.

[0003] While existing enterprises have deployed manufacturing execution systems, equipment control systems, and various testing equipment, enabling them to collect large amounts of production data, the data formats, granularity, and labeling methods are inconsistent across different systems. There is a lack of effective correlation between product, process, equipment, parameter, and quality data. Production data is primarily used for recording and post-event traceability, making it difficult to support real-time process optimization. Furthermore, traditional trial production and step-by-step parameter adjustment methods require multiple rounds of production verification, resulting in high material consumption, low equipment utilization, and batch quality risks.

[0004] Digital twin technology can create a model in virtual space that corresponds to the actual production process and continuously update it using real-time data, providing conditions for virtual verification of process parameters. However, existing technologies mainly focus on equipment visualization and status monitoring, lacking mechanisms for multi-factor correlation, virtual parameter verification, and feedback of optimization results in printed circuit board (PCB) production. Therefore, it is necessary to construct a PCB manufacturing process optimization system based on digital twins and combined with actual quality feedback to achieve closed-loop optimization of process parameters. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a digital twin-based printed circuit board manufacturing process optimization system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The production data acquisition module is used to acquire equipment operation data, process parameter data, material data, production environment data, quality inspection data, and production task data during the printed circuit board production process.

[0008] The process data governance module is used to perform data cleaning, time alignment, identity association and standardization on the production data, and to establish data association relationships between production objects, process steps, production equipment and quality results.

[0009] The digital twin construction module is used to establish a digital twin model corresponding to the actual production process based on the actual production resources, process routes, process parameters, and quality correlation of printed circuit boards.

[0010] The virtual-real synchronization module is used to update the equipment status, production object status and process execution status in the digital twin model based on the production data obtained in real time from the production site, so that the digital twin model maintains a state mapping with the actual production process.

[0011] The process status assessment module is used to determine the deviation between the current production status and the target process status based on the digital twin model, and to determine the process parameters that affect product quality and need to be optimized.

[0012] The quality prediction module is used to predict the corresponding product quality results based on the characteristics of the production object, the current process status, and different combinations of candidate process parameters.

[0013] The process optimization module is used to generate candidate process schemes based on the predicted product quality results, under the constraints of equipment operating capacity, process specifications, product quality requirements and production tasks. The digital twin model is used to perform virtual simulation and comprehensive evaluation of the candidate process schemes, and the target process scheme is determined based on the evaluation results.

[0014] The production execution coordination module is used to send the process control parameters corresponding to the target process scheme to the corresponding production equipment or production management system, and to obtain the actual production results after the target process scheme is executed.

[0015] The closed-loop update module is used to update the model parameters of the digital twin model and the optimization strategy of the process optimization module based on the difference between the actual production results and the prediction results of the digital twin model, so as to form a closed-loop optimization of the printed circuit board manufacturing process.

[0016] Preferably, the production data acquisition module establishes data interfaces with the production equipment control system, manufacturing execution system, quality inspection system, and production management system, respectively. In the time alignment processing of the production data, for high-frequency continuously acquired equipment status data, a sliding window averaging method is used to aggregate and reduce the frequency within the process execution time period; for low-frequency discrete records, if there are missing values, a linear interpolation method is used to fill in the missing values ​​based on the values ​​of adjacent time points; and a unified production identity is established according to production batch, production board, and process execution record, so that production data from different sources are associated with the corresponding printed circuit board production objects.

[0017] Preferably, the digital twin model includes a production resource twin model, a process twin model, and a quality correlation twin model; the production resource twin model is used to describe the equipment capacity, operating status, and adjustable process parameters of the production equipment; the process twin model is used to describe the process sequence, process parameter transmission relationship, and production object status change relationship between different processes in the printed circuit board production process; the quality correlation twin model is used to describe the correlation between process parameters, production status, and quality inspection results.

[0018] Preferably, the virtual-real synchronization module acquires the actual production status according to production events and a preset synchronization cycle; the calculation process of the predicted state in the digital twin model is as follows: first, a set of basic differential equations is established based on the physicochemical process mechanism to obtain the mechanism prediction state; then, the current production status and the process parameters to be controlled are input into the error compensation neural network to output the prediction residual of the mechanism model; finally, the mechanism prediction state and the prediction residual are superimposed to obtain the final prediction state; the virtual-real synchronization module compares the actual production status with the prediction state in the digital twin model; when the state difference between the two exceeds a preset synchronization threshold, the corresponding state variable in the digital twin model is corrected according to the actual production status.

[0019] Preferably, the process status assessment module determines the process status characteristics based on the current product type, substrate properties, production batch, historical process parameters, equipment operating status, and current quality inspection results, and determines the process parameters to be optimized and the corresponding parameter adjustment range based on the correlation between each process parameter and quality indicator.

[0020] Preferably, the quality prediction model is a neural network model based on a multilayer perceptron. This neural network model includes an input layer, a hidden layer using the ReLU activation function, and an output layer. It uses historical production samples as the training set and uses the mean squared error as the loss function to iteratively update the weight parameters through the Adam optimization algorithm. The quality prediction module inputs the product characteristics, equipment status, process status, and candidate process parameters of the printed circuit board to be produced into the quality prediction model to obtain the product qualification probability, predicted quality index value, and quality anomaly risk under the corresponding candidate process parameters.

[0021] Preferably, the process optimization module generates multiple candidate process schemes based on the allowable adjustment range of the process parameters to be optimized, and performs virtual production in the digital twin model respectively. The module then performs a comprehensive evaluation based on the predicted quality results, production cycle time, equipment load, material consumption, and energy consumption corresponding to each candidate process scheme.

[0022] Preferably, the process optimization module determines the optimization weights of different evaluation indicators according to product quality priority, production efficiency priority, resource consumption priority, and equipment stability priority, and obtains in real time the remaining delivery time of the current production order and the standard theoretical production man-hours evaluated in the digital twin model for the order. Based on the obtained product type label and the different interval thresholds of the ratio of the remaining delivery time to the standard theoretical production man-hours, the module dynamically assigns preset numerical weights to different evaluation indicators to determine the target process scheme that meets production constraints and has the best comprehensive evaluation results from multiple candidate process schemes.

[0023] Preferably, the production execution coordination module continuously acquires actual process parameters and quality inspection data during the execution of the target process scheme; when the actual process state deviates from the target process scheme or the predicted quality anomaly risk reaches the preset risk condition, the process optimization module is triggered to re-execute the process scheme optimization and adjust the process parameters of subsequent production processes according to the re-determined target process scheme.

[0024] Preferably, the closed-loop update module forms a process optimization sample from the target process scheme, the prediction results of the digital twin model, the actual process execution data, and the actual quality inspection results, and iteratively updates the process correlation in the quality prediction model and the digital twin model, as well as the process optimization strategies corresponding to different operating conditions, based on the newly added process optimization sample.

[0025] The technical effects and advantages of this invention are as follows:

[0026] This invention establishes a digital twin relationship between the actual and virtual production processes of printed circuit boards (PCBs), unifying production equipment, processes, parameters, and quality status within a digital space to achieve collaborative analysis of equipment, process, and quality data. Before actual adjustments to production equipment parameters, the digital twin model is used to conduct virtual production verification of multiple candidate process schemes, eliminating those with high quality risks or exceeding equipment capabilities, thereby reducing the number of actual trial runs and mitigating the risks of on-site process adjustments. Simultaneously, the process parameters to be optimized are dynamically determined based on the actual equipment status, product type, and quality inspection results, leading to optimized processes. The system can adapt to production conditions of different product batches, different board materials, and different equipment states. During the optimization process, it comprehensively considers product quality, production cycle time, equipment load, material consumption, and energy consumption, so that the optimization results take into account both product yield and production efficiency. It continuously corrects the digital twin model and quality prediction model through actual production results, so that the model can adapt to equipment aging, changes in chemical state, material changes, and changes in production environment, thereby improving the accuracy of long-term process prediction. When new quality risks or deviations in process state occur during production, the digital twin simulation and process parameter optimization are re-triggered to achieve dynamic closed-loop adjustment of printed circuit board production process parameters. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0028] Figure 2 This is a schematic diagram of the closed-loop optimization process for the production process of the present invention. Detailed Implementation

[0029] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0030] Example 1

[0031] This embodiment takes the production process of a multilayer printed circuit board as an example. The printed circuit board production line is set up with the following production steps in sequence: inner layer pattern making, lamination, drilling, hole wall treatment, copper plating, electroplating, outer layer pattern transfer, etching, automatic optical inspection, solder mask, surface treatment, forming, and electrical performance testing.

[0032] The production data acquisition module connects to the data interfaces of the Manufacturing Execution System (MES) and each production device. For each production batch entering the production line, corresponding batch identity information is established; for products requiring board-level tracking, board-level identity information is further established. When a production object enters any production process, its current identity, production equipment, production process, process start time, and process parameters are bound together.

[0033] For example, during the electroplating process, data such as electroplating current density, electroplating time, bath temperature, copper ion concentration, sulfuric acid concentration, additive status, circulation flow rate, production board area, and equipment load status can be collected. After production is completed, copper thickness measurement results and hole copper thickness detection results are obtained and linked to the corresponding production records. Through this method, each quality result can be traced back to the actual equipment and process parameters used in the production of the product.

[0034] Because different production equipment uses different data formats and acquisition cycles, data governance is performed by the process data governance module before digital twin modeling. First, equipment data, manufacturing execution system data, and quality system data are converted to a unified time base. In the time alignment process, for production data with different sampling frequencies, the system uses the process-level timestamps recorded by the manufacturing execution system as the base time axis. For high-frequency, continuously acquired equipment status data (such as tank temperature and spray pressure data acquired once per second), a sliding window averaging method is used to aggregate and reduce the frequency within the process execution time period. For low-frequency, discretely recorded data (such as manual inspection records), if missing values ​​exist within the same process, linear interpolation is used to fill in the missing values ​​based on the values ​​of adjacent time points, thereby unifying the multi-source heterogeneous data to the same time granularity.

[0035] For continuously acquired data, the average, maximum, minimum, standard deviation, and rate of change are extracted within the corresponding process execution time range. For discrete process parameters, the corresponding parameters are recorded according to the process start event and parameter modification event. Then, outlier handling is performed on the production data. Data that clearly exceeds the physical range of the equipment is marked as invalid data; for short-term missing data, whether interpolation or direct marking as missing is determined based on the preceding and following production states.

[0036] After data governance is completed, the production data is logically linked to "product type - production batch - production board - production process - production equipment - process parameters - equipment status - quality result", thereby obtaining a unified production dataset for digital twin modeling and quality prediction.

[0037] A digital twin model can be composed of a production resource twin model, a process twin model, and a quality correlation twin model. The production resource twin model stores the operating capabilities of each production piece of equipment. For example, for electroplating equipment, allowable current density ranges, allowable processing time ranges, operating temperature ranges, and operating chemical solution ranges can be defined; for etching equipment, allowable ranges for conveyor speed, spray pressure, etching solution temperature, and etching solution concentration can be defined.

[0038] The process twin model establishes the state transition relationships of production objects according to the actual production route. For example, after a production object undergoes a drilling process, its digital state adds information on the hole diameter and hole wall state; after electroplating, it further updates the hole copper thickness and surface copper thickness; after etching, it updates the line width and line edge state.

[0039] The current production state of a digital twin model can be represented as a set of states. The process parameters to be controlled are expressed as follows: The digital twin model calculates the predicted state of the production object after the corresponding process based on the current state X and process parameters U. In this embodiment, the construction process of the state evolution function F is as follows: First, a set of basic differential equations (e.g., the electroplating copper thickness growth equation based on Faraday's law) is established based on the physicochemical process mechanism to obtain the mechanism prediction state; however, due to the "edge effect" of electric field concentration at the edge of the board and the "flow field inhomogeneity" caused by the circulation of the bath solution during the actual electroplating process, the pure mechanism model often has a large prediction deviation at the edge of the board and deep in the micropores. Therefore, the current state X (especially the physical position coordinates of the board on the rack and the flow rate of the bath solution) and the process parameter U are then input into an error compensation neural network trained based on historical data to output the prediction residual of the mechanism model; finally, the predicted state is obtained. The mechanistic prediction state is the superposition of the predicted residual. Actual quality feedback is used to periodically update the weights of the error compensation neural network, thereby forming a relationship F that dynamically evolves with the equipment state. Therefore, before actual production, different process parameters in U can be changed to obtain different predicted production results in the digital space.

[0040] The actual production equipment continuously sends real-time data to the virtual-real synchronization module. The system can employ a combination of periodic synchronization and event synchronization. Periodic synchronization is used to acquire continuous change data such as equipment temperature, pressure, current, and liquid status; event synchronization is used to acquire discrete events such as process start, process end, equipment alarm, parameter modification, and quality inspection completion.

[0041] For any given state parameter, the predicted state of the digital twin model is compared with the actual production state. In actual production, considering the strong acid and alkaline environment of the PCB workshop, which can easily lead to sensor drift or data loss, the system has a built-in abnormal degradation mechanism: when the difference between the actual detected temperature and the predicted temperature exceeds a preset reasonable physical limit (e.g., a sudden change exceeding 10℃) or a sensor disconnection occurs, the virtual-real synchronization module will temporarily suspend actual data synchronization and use the forward mechanism prediction value of the digital twin model for process control, triggering an equipment maintenance alarm. However, within the normal error range (e.g., predicted 25.0℃, actual 26.1℃), and when the difference exceeds a preset state synchronization threshold, the current state of the digital twin model is corrected based on the actual detection result. But the original predicted value of the digital twin model is retained during the state correction process, allowing for subsequent calculation of the model prediction error.

[0042] The process status assessment module first obtains the target quality standard corresponding to the current production batch. Taking outer layer circuit etching as an example, it can obtain the target circuit width and allowable deviation range. The system further obtains the conveyor speed, etching solution temperature, etching solution concentration, spray pressure, and previous circuit pattern status corresponding to the current etching equipment. After inputting the above parameters into the digital twin model, it obtains the circuit width prediction result corresponding to the current process status.

[0043] When the predicted line width approaches the quality allowable boundary, a quality risk can be identified in the current production process, even if no non-conforming results have yet been produced. The process status assessment module further calculates the impact of changes in different process parameters on changes in quality indicators. For a given process parameter... By changing this parameter by a certain margin while keeping other parameters constant, the change in quality indicators can be obtained in the digital twin model, thereby determining the quality sensitivity of the process parameters. Process parameters with higher quality sensitivity are prioritized for optimization.

[0044] The quality prediction module uses historical production data to establish a mapping relationship between product quality and process conditions. Input features may include product board thickness, copper foil thickness, design value of line width, line spacing, equipment operating status, current process parameters, quality status of previous processes, and environmental conditions; outputs may include product pass probability, predicted value of line width, predicted value of copper thickness, predicted value of via copper thickness, predicted value of interlayer offset, warpage risk, open circuit risk, and short circuit risk.

[0045] In a preferred embodiment of the present invention, the quality prediction model employs a neural network model based on a multilayer perceptron (MLP). Specifically, the neural network includes an input layer, two to three hidden layers, and an output layer. The number of neurons in the input layer is consistent with the dimension of the input features. For example, when the input features include product board thickness, copper foil thickness, line width design value, line spacing, equipment operating status, and current process parameters, the number of nodes in the input layer is set accordingly. The hidden layers use the ReLU activation function to capture the nonlinear mapping relationship between process parameters and quality indicators. The number of nodes in the output layer corresponds to the number of predicted quality indicators. For example, if the predicted values ​​for line width and copper thickness are to be predicted, then the number of nodes in the output layer is 2. During model training, historical production datasets are used as training samples, mean squared error (MSE) is used as the loss function, and the weight parameters are iteratively updated using the Adam optimization algorithm until the loss function converges, thereby obtaining the trained quality prediction model. For example, in the process of electroplating process optimization, after inputting current density, processing time, solution state, production board area, and historical equipment status into the quality prediction model, the predicted value of copper thickness under the corresponding parameter combination can be obtained. For each candidate combination of process parameters, a corresponding quality prediction result can be obtained.

[0046] Once the system determines the process parameters to be optimized, it first reads the corresponding process specifications. For example, for a certain electroplating process, the allowable range of current density corresponds to the first parameter range, and the processing time corresponds to the second parameter range. The system then generates multiple candidate process parameter combinations within the allowable parameter ranges.

[0047] Candidate parameter combinations can be generated using methods such as equal-interval parameter combinations, local sampling near historically optimized process parameters, random sampling, Bayesian optimization, evolutionary optimization, or a combination of these methods. For each generated candidate process scheme, a hard constraint check is first performed. If a candidate scheme exceeds equipment capacity, process specifications, or safe operating limits, it is directly deleted and not included in the subsequent virtual production process.

[0048] For candidate process schemes that pass the constraint checks, virtual production is carried out in the digital twin model. Assume M candidate process schemes are generated. Each candidate process solution includes a set of process parameters to be adjusted. The digital twin model performs corresponding production state evolution calculations to obtain product quality prediction results, production time, equipment load, material consumption, energy consumption, and equipment operation stability.

[0049] To comprehensively compare different process schemes, a comprehensive evaluation function can be established. Where Q represents the quality risk assessment quantity, T represents the production time assessment quantity, E represents the energy and material consumption assessment quantity, R represents the equipment load assessment quantity, and S represents the process deviation assessment quantity. to These represent the weights of the corresponding evaluation indicators. The system dynamically calculates and allocates weights based on actual production tasks and delivery status. Specifically, the system obtains the remaining delivery time of the current production order in real time. and the standard theoretical production hours for evaluating the order in the digital twin model. And obtain the product type label.

[0050] When the product type label is "high-reliability printed circuit board" (such as boards for aerospace and medical devices), the system adopts an absolute quality priority strategy with fixed weight allocation: .

[0051] When the product type is a standard printed circuit board, the system will consider the lead time adequacy ( Dynamic adjustments will be made.

[0052] like This indicates that the delivery time is sufficient, and the system adopts a balanced quality-biased strategy, automatically assigning values. ;

[0053] like The system adopts an efficiency balancing strategy and automatically assigns values. ;

[0054] like This indicates a risk of delay. The system adopts an efficiency-first strategy and automatically assigns values. .

[0055] The target process scheme is determined by comparing the comprehensive evaluation results of each candidate process scheme.

[0056] Example 2

[0057] In one implementation, a production batch enters the outer layer etching process. Based on previous automated optical inspection data, it is found that the overall size of the current circuit pattern differs from the historical average. The process status assessment module determines that if the original etching parameters are continued, some fine circuit widths may fall below the quality control range. Therefore, the system identifies etching transfer speed and spray pressure as parameters to be optimized.

[0058] Subsequently, multiple parameter combinations are generated within permissible limits. Different parameter combinations are fed into a digital twin model for virtual etching, and the quality prediction module outputs the corresponding line width and short-circuit / open-circuit risks for each scheme. The process optimization module eliminates schemes with excessive line width and comprehensively compares production efficiency and equipment operating status among the remaining schemes to ultimately obtain the target conveying speed and target spray pressure.

[0059] The production execution collaboration module sends the corresponding target parameters to the production management system. After obtaining execution permission, the production management system sends parameter setting instructions to the equipment or generates corresponding process parameter adjustment tasks for production personnel to confirm and execute. Therefore, this invention can achieve both automatic parameter adjustment of production equipment and parameter adjustment after manual confirmation.

[0060] After the target process is actually executed, the automated optical inspection equipment obtains the actual line width of the corresponding product. Assume the digital twin model predicts the line width as follows: The actual measurement result is Then the prediction error of this model can be calculated. .

[0061] The closed-loop update module saves the characteristics of the production object, the process parameters before optimization, the target process parameters, the equipment status, the digital twin prediction results, the actual production results, and the actual quality results for the current production. This data forms a new production sample. When a set number of new production samples are accumulated, or when prediction errors continuously exceed preset error conditions, the digital twin model is updated. Model updates may include correcting the mapping coefficients between process parameters and quality indicators, updating equipment performance status, adjusting quality prediction model parameters, and adjusting the range of process optimization parameters for different product types. This allows the digital twin model to adapt to performance changes that occur after long-term use of production equipment.

[0062] Quality issues in printed circuit boards are not necessarily caused by a single process step. For example, an abnormal width of an outer layer circuit may be affected by multiple production steps simultaneously, including exposure, development, and etching. Therefore, in a further implementation, a digital twin model establishes cross-process influence relationships.

[0063] When a final quality anomaly occurs, the system traces the process status of different steps forward along the process route, determining the correlation between each preceding process parameter and the current quality indicator based on historical production data. If it is confirmed that the current quality anomaly is mainly caused by the preceding process status, the current process parameters do not need to be directly modified; instead, the corresponding preceding process parameters are adjusted for the next production batch.

[0064] For example, if the etched line width is consistently too small, system analysis reveals that the etching equipment parameters remain stable. Further analysis shows that the exposure energy has been consistently above the allowable range. A digital twin model simulates the overall production results after reducing the exposure energy. When the simulation results show that the final line width stability can be improved, the exposure energy adjustment scheme is used as a candidate optimization scheme for the next production batch, thereby achieving collaborative optimization among multiple production processes.

[0065] During production, if equipment malfunctions, the original target process may no longer be applicable to the current production environment. For example, the state of the plating solution in the electroplating equipment may change significantly. The virtual-real synchronization module first synchronizes this change to the digital twin model, which then re-predicts the copper thickness corresponding to the existing process parameters based on the new solution state.

[0066] If the predicted copper thickness exceeds the allowable range, the process status assessment module triggers re-optimization. During the re-optimization process, the previous equipment status is no longer used; instead, the current actual chemical solution status is used as the new initial condition to regenerate process parameters, thereby ensuring that the process optimization results always correspond to the current actual production conditions.

[0067] For each successfully executed process optimization, the corresponding data is saved as a historical process optimization case. A historical process optimization case should include at least the product category, product structural characteristics, abnormal quality indicators, the process in which the abnormality occurred, equipment status, original process parameters, adjusted process parameters, digital twin prediction results, and actual quality improvement results.

[0068] When similar products and process states appear subsequently, the process optimization module prioritizes searching historical similar cases and uses historically successful parameters as initial candidate parameters, followed by virtual verification through the current digital twin model. Therefore, the system can transform historical manual process experience and automated system optimization experience into reusable digital process knowledge.

[0069] In practical applications, the entire process optimization process can be carried out as follows: First, obtain the current production task and product process route; second, obtain the real-time status of the production equipment and the current product status; then, synchronize the above status to the digital twin model; the digital twin model predicts the product quality under the current process conditions.

[0070] If the predicted product quality meets production requirements and there are no obvious quality risks, then production continues with the existing production parameters. If the predicted results do not meet quality requirements or reach the preset risk conditions, then the process parameters to be optimized are determined, and multiple candidate process schemes are generated within the allowable range of process parameters. Digital twin virtual production is performed on each candidate process scheme, and candidate process schemes that do not meet equipment constraints, process constraints, and quality constraints are excluded.

[0071] By comparing the quality, production cycle time, equipment load, and resource consumption of other candidate process schemes, the target process scheme is determined, and the target process parameters are applied to actual production. After production is completed, the actual quality inspection results are obtained, and the actual quality results are compared with the digital twin prediction results. The digital twin model is then updated based on the comparison results. This process is continuously repeated with different production batches, achieving dynamic closed-loop optimization of the printed circuit board manufacturing process.

[0072] This implementation method, compared to production control based solely on historical fixed process formulas, allows for real-time changes to process parameters according to the current actual production status. Virtual verification is then performed using a digital twin model before the process parameters are actually executed, thereby reducing the trial-and-error process on the production floor.

[0073] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A printed circuit board manufacturing process optimization system based on digital twins, characterized in that, include: The production data acquisition module is used to acquire equipment operation data, process parameter data, material data, production environment data, quality inspection data, and production task data during the printed circuit board production process. The process data governance module is used to perform data cleaning, time alignment, identity association and standardization on the production data, and to establish data association relationships between production objects, process steps, production equipment and quality results. The digital twin construction module is used to establish a digital twin model corresponding to the actual production process based on the actual production resources, process routes, process parameters, and quality correlation of printed circuit boards. The virtual-real synchronization module is used to update the equipment status, production object status and process execution status in the digital twin model based on the production data obtained in real time from the production site, so that the digital twin model maintains a state mapping with the actual production process. The process status assessment module is used to determine the deviation between the current production status and the target process status based on the digital twin model, and to determine the process parameters that affect product quality and need to be optimized. The quality prediction module is used to predict the corresponding product quality results based on the characteristics of the production object, the current process status, and different combinations of candidate process parameters. The process optimization module is used to generate candidate process schemes based on the predicted product quality results, under the constraints of equipment operating capacity, process specifications, product quality requirements and production tasks, to perform virtual simulation and comprehensive evaluation of the candidate process schemes through the digital twin model, and to determine the target process scheme based on the evaluation results. The production execution coordination module is used to send the process control parameters corresponding to the target process scheme to the corresponding production equipment or production management system, and to obtain the actual production results after the target process scheme is executed. The closed-loop update module is used to update the model parameters of the digital twin model and the optimization strategy of the process optimization module based on the difference between the actual production results and the prediction results of the digital twin model, so as to form a closed-loop optimization of the printed circuit board manufacturing process.

2. The printed circuit board manufacturing process optimization system based on digital twins according to claim 1, characterized in that: The production data acquisition module establishes data interfaces with the production equipment control system, manufacturing execution system, quality inspection system, and production management system. In the time alignment processing of the production data, for high-frequency continuously acquired equipment status data, a sliding window averaging method is used to aggregate and reduce the frequency within the process execution time period; for low-frequency discrete records, if there are missing values, a linear interpolation method is used to complete them based on the values ​​of adjacent time points; and a unified production identity is established according to production batch, production board, and process execution record, so that production data from different sources are associated with the corresponding printed circuit board production objects.

3. The printed circuit board manufacturing process optimization system based on digital twins according to claim 1, characterized in that: The digital twin model includes a production resource twin model, a process twin model, and a quality correlation twin model. The production resource twin model describes the equipment capacity, operating status, and adjustable process parameters of the production equipment. The process twin model describes the process sequence, process parameter transmission relationship, and production object status change relationship between different processes in the printed circuit board production process. The quality correlation twin model describes the correlation between process parameters, production status, and quality inspection results.

4. The printed circuit board manufacturing process optimization system based on digital twins according to claim 1, characterized in that: The virtual-real synchronization module acquires the actual production status according to production events and a preset synchronization cycle. The calculation process of the predicted status in the digital twin model is as follows: first, a set of basic differential equations is established based on the physicochemical process mechanism to obtain the mechanism prediction status; then, the current production status and the process parameters to be controlled are input into the error compensation neural network to output the prediction residual of the mechanism model; finally, the mechanism prediction status and the prediction residual are superimposed to obtain the final prediction status. The virtual-real synchronization module compares the actual production status with the prediction status in the digital twin model. When the difference between the two states exceeds a preset synchronization threshold, the corresponding state variables in the digital twin model are corrected according to the actual production status.

5. The printed circuit board manufacturing process optimization system based on digital twins according to claim 1, characterized in that: The process status assessment module determines the process status characteristics based on the current product type, substrate properties, production batch, historical process parameters, equipment operating status, and current quality inspection results, and determines the process parameters to be optimized and the corresponding parameter adjustment range based on the correlation between each process parameter and quality indicator.

6. The printed circuit board manufacturing process optimization system based on digital twins according to claim 1, characterized in that: The quality prediction model is a neural network model based on a multilayer perceptron. This neural network model includes an input layer, a hidden layer using the ReLU activation function, and an output layer. It uses historical production samples as the training set and uses the mean squared error as the loss function to iteratively update the weight parameters through the Adam optimization algorithm. The quality prediction module inputs the product characteristics, equipment status, process status, and candidate process parameters of the printed circuit board to be produced into the quality prediction model to obtain the product qualification probability, predicted quality index values, and quality anomaly risk under the corresponding candidate process parameters.

7. The printed circuit board manufacturing process optimization system based on digital twins according to claim 1, characterized in that: The process optimization module generates multiple candidate process schemes based on the allowable adjustment range of the process parameters to be optimized, and performs virtual production in the digital twin model for each scheme. The module then performs a comprehensive evaluation based on the predicted quality results, production cycle time, equipment load, material consumption, and energy consumption of each candidate process scheme.

8. The printed circuit board manufacturing process optimization system based on digital twins according to claim 7, characterized in that: The process optimization module determines the optimization weights of different evaluation indicators according to product quality priority, production efficiency priority, resource consumption priority, and equipment stability priority. It also obtains the remaining delivery time of the current production order and the standard theoretical production man-hours evaluated in the digital twin model in real time. Based on the obtained product type label and the different interval thresholds of the ratio of the remaining delivery time to the standard theoretical production man-hours, it dynamically assigns preset numerical weights to different evaluation indicators to determine the target process scheme that meets production constraints and has the best comprehensive evaluation results from multiple candidate process schemes.

9. The printed circuit board manufacturing process optimization system based on digital twins according to claim 1, characterized in that: The production execution collaboration module continuously acquires actual process parameters and quality inspection data during the execution of the target process scheme; When the actual process state deviates from the target process plan or the predicted quality anomaly risk reaches the preset risk condition, the process optimization module is triggered to re-execute the process plan optimization and adjust the process parameters of subsequent production processes according to the redefined target process plan.

10. The printed circuit board manufacturing process optimization system based on digital twins according to claim 1, characterized in that: The closed-loop update module forms a process optimization sample from the target process scheme, the prediction results of the digital twin model, the actual process execution data, and the actual quality inspection results. Based on the newly added process optimization sample, it iteratively updates the process correlation in the quality prediction model and the digital twin model, as well as the process optimization strategies corresponding to different operating conditions.

Images