High-precision multi-layer PCBA full-automatic production line intelligent scheduling optimization method

Abstract

The application relates to the technical field of overall factory control, in particular to a high-precision multi-layer PCBA full-automatic production line intelligent scheduling optimization method, which comprises the following steps: acquiring physical equipment data flow of a production line in real time, performing feature extraction, alignment and generating a production state data set through a data fusion model; then, running a data-driven model to predict key production indexes and implicit states, and simultaneously establishing a running mechanism model to simulate equipment physical behaviors; constructing a digital twin model, deeply fusing the prediction result of the data-driven model and the simulation result of the mechanism model by adopting a physical information neural network, calculating and updating the digital twin state of the production line; finally, applying a heuristic optimization algorithm to generate an optimized production scheduling scheme, and combining predefined adaptive control rules to generate real-time executable control instructions to adjust process parameters. The application can improve the scheduling efficiency of the PCBA production line and the self-adaptive ability to a dynamic environment.

Description

Technical Field

[0001] This invention relates to the field of industrial control technology, specifically to an intelligent scheduling and optimization method for a high-precision multilayer PCBA fully automated production line. Background Technology

[0002] Industrial Control Systems (ICS) are the cornerstone of modern industrial automation, widely used in critical infrastructure sectors such as manufacturing, energy, transportation, and water treatment. ICS integrate sensors, actuators, controllers (such as programmable logic controllers (PLCs), and communication networks, combining the functions of distributed control systems and monitoring and data acquisition systems to achieve real-time monitoring, data acquisition, and precise programmed control of industrial processes, aiming to ensure efficient, safe, and stable operation of production processes.

[0003] With the increasing integration of Industrial Internet and Internet of Things (IoT) technologies, the inherent limitations of traditional industrial control systems are becoming increasingly apparent when dealing with the complex scheduling requirements of high-precision multilayer PCBA fully automated production lines. High-precision multilayer PCBA production lines integrate a series of precise and interdependent processes, including SMT placement, reflow soldering, AOI inspection, through-hole assembly, and wave soldering. Each step places stringent requirements on control accuracy, production cycle time, and final quality. However, the PCBA production environment is rife with dynamic events and uncertainties, such as unexpected equipment failures, material supply delays, urgent order insertions, fluctuations in actual processing time, and yield variations. These disturbances often cause pre-programmed static production plans or control programs based on deterministic assumptions to quickly become ineffective. Consequently, these industrial control systems often lack sufficient dynamic adaptability, making it difficult to quickly and effectively adjust their preset control logic and scheduling sequences to adapt to new situations. This lag in response or insufficient adjustment capability directly affects the continuity and stability of production, as well as the ability to respond quickly to dynamic market demands.

[0004] To address this, a smart scheduling optimization method for a fully automated high-precision multilayer PCBA production line is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent scheduling optimization method for a high-precision, multi-layer PCBA fully automated production line, addressing the problems of poor adaptability to dynamic disturbances and difficulty in coordinating multi-objective optimization in existing PCBA production scheduling. First, real-time data streams from the physical equipment of the production line are acquired, features are extracted and aligned, and a production status dataset is generated through a data fusion model. Then, a data-driven model is run to predict key production indicators and implicit states, while a mechanism model is established to simulate the physical behavior of the equipment. A digital twin model is constructed, and a physical information neural network is used to deeply fuse the prediction results of the data-driven model with the simulation results of the mechanism model to calculate and update the digital twin state of the production line. Finally, a heuristic optimization algorithm is applied to generate an optimized production scheduling scheme, and predefined adaptive control rules are combined to generate real-time executable control commands to adjust process parameters. This invention can improve the scheduling efficiency of PCBA production lines and their adaptability to dynamic environments.

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

[0007] A method for intelligent scheduling and optimization of a high-precision multilayer PCBA fully automated production line includes:

[0008] Acquire physical device data streams, including structured and unstructured data;

[0009] Feature extraction is performed on the data stream of the physical equipment to generate multi-source features; feature alignment is performed based on timestamps and work order identifiers, and multi-source features of the same workpiece are fused using a data fusion model to output a production status dataset;

[0010] A data-driven model is trained and run based on the production status dataset to predict the comprehensive production status indicators and implicit production status of the production line equipment; a mechanism model is established and run to simulate the physical behavior of the production line equipment; and a digital twin model is constructed to calculate and update the digital twin status of the production line by weighted fusion of the comprehensive production status indicators, the implicit production status, and the physical behavior.

[0011] Using a heuristic optimization algorithm, a production scheduling scheme is generated based on the production status dataset, the digital twin status, and preset constraints; and predefined adaptive control rule logic is applied to generate executable control instructions.

[0012] Furthermore, the sources of the structured data and the unstructured data include:

[0013] The structured data includes equipment status and process parameter data from programmable logic controllers, physical parameter data from sensors, and production plan information from manufacturing execution systems; the unstructured data includes quality inspection data from automated optical inspection equipment.

[0014] Furthermore, generating the multi-source features includes:

[0015] For the unstructured data, a lightweight convolutional neural network is used to identify and quantify welding defect features;

[0016] For the structured data, signal processing methods are used to extract frequency domain features and time domain statistical features from the time-series physical parameters in the physical parameter data; principal component analysis is used to extract principal component features from the process parameter data and the non-time-series physical parameters in the physical parameter data; categorical data in the equipment status and production plan information are encoded to generate categorical features; the welding defect features are combined with the frequency domain features, the time domain statistical features, the principal component features, and the categorical features to form the multi-source features.

[0017] Furthermore, the process of outputting the production status dataset includes:

[0018] The multi-source features are standardized, and the timestamp and work order identifier are used as association keys. The multi-source features of the same workpiece within a predetermined time window are matched and aligned using a time window matching algorithm.

[0019] A weighted fusion algorithm is used to weight and combine the aligned multi-source features to generate a comprehensive production status index; the implicit production status is inferred based on the aligned multi-source features using a probabilistic graphical model.

[0020] The comprehensive production status index and the implicit production status are included as results in the output production status dataset.

[0021] Furthermore, establishing and operating the aforementioned mechanism model specifically includes:

[0022] For the selected production line equipment, an ordinary differential equation model based on simplified physical principles is established;

[0023] Obtain real-time parameters reflecting the current operating status of the equipment from the production status dataset, and use the real-time parameters as the initial conditions of the ordinary differential equation model;

[0024] Load the preset physical parameters related to the production line equipment; use a numerical calculation library to numerically solve the ordinary differential equation model, simulate and output the physical behavior.

[0025] Furthermore, by weightedly fusing the comprehensive production status indicators, the implicit production status, and the physical behavior, specifically including:

[0026] The comprehensive production status index and the implicit production status are initially weighted and combined to generate a first fusion index;

[0027] The physical behavior is feature extracted and converted into a second fusion index;

[0028] The first fusion index, the second fusion index, and the third fusion index selected from the multi-source features are then weighted and combined again to generate and update the digital twin state.

[0029] Furthermore, generating a production scheduling scheme using the heuristic optimization algorithm specifically includes:

[0030] A genetic algorithm is used as the heuristic optimization algorithm.

[0031] The production scheduling problem is encoded into individuals, where each individual represents a candidate production scheduling solution;

[0032] A fitness function is constructed to evaluate the merits of the candidate production scheduling schemes. Based on the multi-source features and the digital twin state, the fitness function is calculated by weighted combination of scheduling objectives, wherein the scheduling objectives include work order delivery delays, predicted production yield losses, and equipment energy consumption.

[0033] Under the premise of satisfying the preset constraints, the optimal individual with the best fitness value is determined from the population containing N individuals by iteratively performing selection, crossover and mutation operations, and the optimal individual is decoded into the production scheduling scheme.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] 1. This invention constructs a comprehensive and detailed production status dataset through systematic real-time acquisition of multi-source heterogeneous data (structured and unstructured), deep feature engineering, and a preliminary data fusion model. This effectively overcomes the problems of incomplete information and low quality caused by traditional single data sources or simple data aggregation. By performing targeted feature extraction on the original data, precise alignment of time and work order dimensions, and preliminary fusion based on weighted algorithms and probabilistic graphical models, it not only significantly improves the quality and relevance of the input data but also generates rich intermediate-layer information containing comprehensive production status indicators and implicit production status. This lays a solid, reliable, and comprehensive data foundation for subsequent construction of high-fidelity digital twins and driving intelligent decision-making.

[0036] 2. This invention constructs and utilizes a digital twin state that incorporates a data-driven model, a simplified mechanistic model, and is calculated and updated in real time through a specific fusion mechanism. This digital twin state is not a simple reproduction of the physical production line, but rather an organic combination of accurate prediction of complex production indicators and implicit states, and simulation of core physical behaviors. In particular, when using segmented dynamic weighted fusion, it ensures that the twin state, while fitting data patterns, also adheres to key physical constraints or dynamically adjusts its information emphasis based on real-time context. This digital twin state, integrating insights from multiple sources, provides heuristic scheduling algorithms with unprecedented depth of understanding of the current and future trends of the production line.

[0037] 3. This invention, based on production status datasets and digital twin states, achieves deep integration and closed-loop operation of heuristic scheduling optimization and adaptive control rule logic. The heuristic optimization algorithm utilizes these rich real-time and predictive inputs to dynamically and accurately quantify and evaluate its multi-objective fitness function, thereby generating a highly optimized production scheduling scheme that adapts to the current real-world operating conditions. Simultaneously, the adaptive control rule logic generates executable process adjustment instructions in real time based on these same inputs. This tight integration and closed-loop feedback between scheduling and control ultimately results in comprehensive benefits in improving production efficiency, ensuring product quality, enhancing production line resilience, and reducing operating costs. Attached Figure Description

[0038] Figure 1 This invention provides a flowchart illustrating an intelligent scheduling and optimization method for a fully automated high-precision multilayer PCBA production line.

[0039] Figure 2 This invention provides a schematic diagram of the structure for multi-source feature generation;

[0040] Figure 3 This is a schematic diagram of the weighted fusion process of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] Please see Figures 1 to 3 This invention provides an intelligent scheduling and optimization method for a high-precision multilayer PCBA fully automated production line. The technical solution is as follows:

[0043] Example 1:

[0044] To more specifically illustrate the technical details and application process of the intelligent scheduling optimization method for a high-precision multilayer PCBA fully automated production line disclosed in this invention, this embodiment of the invention takes the automated production process of the core processing board of an automotive engine control unit as an example, aiming to demonstrate how this method can be applied to actual production scenarios to improve the continuity and stability of production and the ability to respond quickly to dynamic market demands.

[0045] The core processing board is manufactured on a typical high-precision, multi-layer PCBA fully automated production line. This line is configured sequentially with key process units such as solder paste printing (SPM), solder paste inspection (SPI), high-speed surface mount technology (SMT), multi-zone reflow soldering, automated optical inspection (AOI), X-ray inspection (AXI), and in-circuit functional testing and programming (FCT). Each major process device on the line is equipped with standard industrial communication interfaces (e.g., supporting the OPC UA protocol), providing a foundation for real-time data acquisition and remote command issuance.

[0046] like Figure 1 As shown, a method for intelligent scheduling and optimization of a high-precision multilayer PCBA fully automated production line includes:

[0047] Acquire physical device data streams, including structured and unstructured data.

[0048] Furthermore, the sources of the structured data and the unstructured data include:

[0049] The structured data includes:

[0050] In this embodiment, the equipment status and process parameter data from the programmable logic controller (PLC) include: a deployed data acquisition software module acting as an OPC UA client, connected to the PLC controllers of various key devices (such as pick-and-place machines and reflow ovens) via an industrial Ethernet network. This software subscribes in real-time or polls predefined key data tags within the PLC at frequencies configured on demand (e.g., high-frequency data such as equipment status is collected at the hundreds of millisecond level, while lower-frequency data such as process parameter data is collected at the second level) to obtain equipment status and process parameter data such as machine operating mode codes, fault codes, actual temperature zones, and conveyor belt speeds. All collected data is appended with a precise timestamp and associated with the currently processed work order identifier and PCBA serial number.

[0051] Physical parameter data from sensors includes various industrial-grade sensors installed in key equipment locations (such as SMT placement heads) and workshop environments (such as near reflow ovens and specific areas within the workshop). These sensors include vibration sensors, high-precision temperature sensors, ambient temperature and humidity sensors, and smart meters or power sensors for monitoring equipment energy consumption. These sensors are connected to an edge data acquisition gateway via standard industrial buses (such as Modbus RTU / TCP) or digital / analog signal interfaces. The gateway reads the raw sensor data and performs preliminary processing (such as unit conversion and filtering). Subsequently, using a lightweight messaging protocol (such as MQTT), the data, with timestamps and source identifiers, is published to an internal message middleware within the factory. The data acquisition software module, acting as an MQTT client, subscribes to the relevant content to obtain real-time physical parameters. The data update frequency is dynamically set according to the specific sensors and monitoring requirements.

[0052] And production planning information from the Manufacturing Execution System (MES), including: secure interaction with the MES through standard web service interfaces (such as RESTful APIs), triggered by production rhythm or specific events (such as work order start / end), or querying the MES for detailed information on current and subsequent work orders at preset intervals (such as every minute), including work order number, product model, planned quantity, priority code, planned start and end time, and other production instructions and planning data. The structured data obtained is parsed and used to enrich and correlate production status data.

[0053] The unstructured data includes quality inspection data from automated optical inspection equipment, including: after the AOI equipment completes the inspection, in addition to generating a structured report containing defect details, it also saves optical images of the defect areas determined to contain defects according to the configuration. The image file name contains traceable information (such as PCBA serial number, inspection time, defect coordinates, etc.) and is stored in a designated network file server or image database. While processing the structured inspection report, the data acquisition service records the storage location information of the associated image files.

[0054] By integrating data from PLC, sensors, MES, and AOI, a comprehensive and multi-dimensional data foundation was built, providing more complete information for subsequent analysis and decision-making, and avoiding the one-sidedness caused by a single data source.

[0055] Furthermore, feature extraction is performed on the physical equipment data stream to generate multi-source features; feature alignment is performed based on timestamps and work order identifiers, and the multi-source features of the same workpiece are fused using a data fusion model to output a production status dataset.

[0056] Among them, the data fusion model is a set of core computational logic used to integrate aligned multi-source feature information to generate production status datasets.

[0057] Furthermore, such as Figure 2 As shown, generating the multi-source features using a data fusion model includes:

[0058] For optical images of defective regions in unstructured data, a lightweight convolutional neural network (CNN) is used to identify and quantify soldering defect features. During acquisition, a pre-trained CNN model, fine-tuned on PCBA defect images, is employed. First, the input optical images of defective regions are pre-processed, resizing to 224×224 pixels for size normalization, and pixel values ​​are also normalized. After preprocessing, the images are input into the CNN model. This model outputs the probability value of the image belonging to each predefined defect category, such as cold solder joints, solder bridges, tombstoning, insufficient solder, and component misalignment. These defect categories and their corresponding probability values ​​together constitute the soldering defect features.

[0059] For the structured data:

[0060] For the time-series physical parameters in the physical parameter data, signal processing methods are applied to extract frequency domain features and time domain statistical features. Taking the time series segment of the original vibration signal collected from the chip mounter vibration sensor and synchronized with the PCBA mounting operation as an example, when extracting time domain statistical features, NumPy and the SciPy.stats library are used to calculate key statistics such as the mean, standard deviation, root mean square, peak-to-peak value, kurtosis, skewness, and crest factor of the vibration signal within a specific time window. For frequency domain feature extraction, SciPy.fft.fft is used to perform a Fast Fourier Transform to obtain spectral information. Then, the main frequency components are identified and extracted from the spectrum, such as the three highest energy frequency points and their amplitudes. Simultaneously, spectral kurtosis, spectral entropy, and the energy proportion of each preset frequency band are calculated. Finally, this series of operations generates a set of numerical features, laying the foundation for subsequent analysis.

[0061] Principal component analysis (PCA) is applied to extract principal component features from process parameter data and non-time-series physical parameters. Process parameter data such as temperature in each temperature zone and conveyor belt speed from the PLC, as well as historical data on non-time-series physical parameters such as average ambient temperature, average humidity, and average equipment energy consumption from sensors, are collected and integrated into a dataset. After standardization, PCA is used to select the top k principal components retaining 95% of the variance, and this data is saved to form a PCA model. In the real-time phase, the currently acquired parameter values ​​are subjected to the same standardization process, and the saved PCA model is projected onto the principal component space, ultimately outputting k linearly independent numerical principal component features, thus achieving data dimensionality reduction.

[0062] Categorical data in equipment status and production plan information are encoded to generate categorical features. Equipment status includes equipment status codes provided by the PLC, such as "Running," "Idle," and "Fault Code A." Categorical data in production plan information includes order priorities from the MES, such as "High," "Medium," and "Low." One-hot encoding is applied to this type of categorical data, converting it into a set of binary vectors. For example, the order priority "High" would correspond to the encoding [0,0,1].

[0063] Finally, the welding defect features are combined with the frequency domain features, the time domain statistical features, the principal component features, and the category features to form a multi-source feature that includes data from different data sources and processing methods.

[0064] By applying appropriate feature extraction methods to different modal information such as images, sensor signals, equipment status, and production plans, and transforming them into numerical or uniformly coded categorical features, multimodal data fusion is achieved, laying the foundation for a comprehensive and accurate understanding of production status.

[0065] Furthermore, the process of outputting the production status dataset includes:

[0066] First, the multi-source features are standardized, and the timestamp and work order identifier are used as association keys. Then, the multi-source features of the same workpiece within a predetermined time window are matched and aligned using a time window matching algorithm.

[0067] The time window matching algorithm involves aggregating feature records that are close in time (e.g., differing by no more than a preset short time threshold, or belonging to the same logical processing unit, such as reflow soldering processes) and have the same work order and sequence number identifier. For features that are generated multiple times within the time window (such as high-frequency sensor features), aggregation (e.g., taking the average, maximum, or latest value) or preservation of their sequence can be performed.

[0068] Then, a weighted fusion algorithm is used to weight and combine the aligned multi-source features to generate the comprehensive production status index. This process includes: selecting a subset of relevant features for each index and assigning preset weights to these features based on expert experience to generate comprehensive production status indicators such as the comprehensive health score of key equipment and the quality risk index of the current process. Taking the generation of the "SMT placement quality risk index" as an example, it is used to evaluate the overall placement quality of the current PCBA in the SMT process. The selected feature subset mainly comes from the aligned and standardized multi-source features extracted in the previous steps that are closely related to the SMT process, and may include: statistical features reflecting solder paste printing quality (e.g., solder paste volume consistency index obtained by analyzing structured data); quantitative features reflecting the placement accuracy of key components (e.g., the average offset of the X, Y axes and rotation angle of high-precision components obtained by analyzing pick-and-place machine feedback data or AOI pre-inspection image features); vibration features reflecting the stability of pick-and-place machine operation (e.g., the time-domain RMS value or specific frequency band energy of the placement head vibration signal); and category features reflecting the current state of the pick-and-place machine (e.g., the coded value of whether the equipment is in a fault-free operating state). After normalizing the selected feature values, they are assigned corresponding weights based on expert experience and then summed using a weighted average.

[0069] Simultaneously, the implicit production state is inferred using a probabilistic graphical model based on the aligned multi-source features. One or more probabilistic graphical models (PGMs), such as Bayesian networks, are constructed to infer implicit production states that are difficult to measure directly. The network structure and conditional probability table of the PGM are obtained through offline training and learning based on domain knowledge and a large amount of historical data. Multi-source features are input as observational evidence into the trained PGM, and the posterior probability distribution of the target implicit state node is calculated using the PGM's inference algorithm. For example, inferring the "current weld joint internal microcrack risk level" or the "probability interval of unplanned equipment downtime in the next 1 hour". The output implicit production state is a probabilistic description of the implicit state, such as: "microcrack risk": {"high": 0.1, "medium": 0.3, "low": 0.6}, "downtime probability (1h)": 0.05.

[0070] Finally, the comprehensive production status indicators and the implicit production status are included as results in the output production status dataset. Each record is associated with a unique workpiece identifier (timestamp and work order number). This dataset can be published to a message queue (such as Kafka) in the form of a JSON object stream for subsequent intelligent production line scheduling.

[0071] Aligned multi-source features from multiple sensors, devices, and processes are weighted and combined to transform them into a few comprehensive indicators with clear business implications. This transformation process enables production managers and subsequent intelligent algorithms to more easily understand and utilize the precise status of the current workpiece. Compared with scattered raw features or simple pass / fail judgments, these indicators can more precisely reflect the degree of good or bad status.

[0072] A data-driven model is trained and run based on the production status dataset to predict the comprehensive production status indicators and implicit production status of the production line equipment; a mechanism model is established and run to simulate the physical behavior of the production line equipment; and a digital twin model is constructed to calculate and update the digital twin status of the production line by fusing the comprehensive production status indicators, the implicit production status, and the physical behavior through a physical information neural network.

[0073] Furthermore, establishing and operating the aforementioned mechanism model specifically includes:

[0074] For the selected production line equipment, an ordinary differential equation model based on simplified physical principles is established;

[0075] Obtain real-time parameters reflecting the current operating status of the equipment from the production status dataset, and use the real-time parameters as the initial conditions of the ordinary differential equation model;

[0076] Load the preset physical parameters related to the production line equipment; use a numerical calculation library to numerically solve the ordinary differential equation model, simulate and output the physical behavior.

[0077] Specifically, in this embodiment, a reflow oven is selected for modeling to simulate the average temperature change trend of the PCBA over time when passing through a specific temperature zone of the reflow oven. It is assumed that heat transfer within this temperature zone mainly consists of forced convection heating of the PCBA by the heating elements and heat dissipation from the PCBA to the environment. Complex radiative heat transfer and temperature gradients at different locations within the PCBA are ignored, and it is treated as a lumped parameter system. Based on a simplified energy balance, an ordinary differential equation of the following form is established to describe the average temperature of the PCBA. Rate of change:

[0078] ;

[0079] in, Indicates the average temperature of PCBA The derivative with respect to time t, The equivalent heat capacity of the PCBA. and These are the equivalent convective heat transfer coefficient and effective heat transfer area between the heating air and the PCBA, respectively. The effective temperature for heating air in the current temperature range. and These represent the equivalent heat dissipation coefficient and effective heat dissipation area between the PCBA and the unheated environment within the temperature range, respectively. This is the reference temperature for the environment within the temperature range. This is the heat generation power of the PCBA itself, which can be ignored and is zero.

[0080] When a target PCBA is about to enter or has just entered the modeled temperature zone, the latest data related to the PCBA and the temperature zone is retrieved from the production status dataset to obtain the initial temperature of the PCBA when entering the temperature zone. ,Will As the initial values ​​for solving the ordinary differential equation, numerical calculations return a time series representing the average temperature of the PCBA within the simulation time interval. The curve showing the change. For example, {"Timestamp": [0, 0.5, 1.0], "Simulated PCBA Average Temperature": [60.5, 68.2, 75.3]}. This result represents the simulated physical behavior.

[0081] The mechanistic model is built upon physical principles, enabling it to clearly elucidate the underlying physical causes when analyzing equipment behavior, thereby enhancing the interpretability of the digital twin model. In practical applications, using real-time parameters from the production status dataset, such as current temperature and speed, as initial conditions or time-varying inputs to ordinary differential equations allows the mechanistic model's simulation to start from the actual current state of the physical production line. This makes the model's simulation results in the short term more closely resemble reality, providing a more accurate physical behavior benchmark for subsequent fusion with physical information neural networks.

[0082] Furthermore, such as Figure 3 The diagram illustrates the structure of the weighted fusion process of the present invention. The weighted fusion of the comprehensive production state index, the implicit production state, and the physical behavior specifically includes:

[0083] The comprehensive production status index and the implicit production status are initially weighted and combined to generate a first fusion index;

[0084] The physical behavior is feature extracted and converted into a second fusion index;

[0085] The first fusion index, the second fusion index, and the third fusion index selected from the multi-source features are then weighted and combined again to generate and update the digital twin state.

[0086] Specifically, for example, a comprehensive production status indicator is predicted based on a data-driven model (such as an LSTM model), including the expected yield of the current batch and the current health score of key equipment A, while assessing the implicit production status, such as the risk level of microcracks appearing in workpiece a. These various predictive information from the data-driven model are then integrated to form the first fusion indicator.

[0087] On the other hand, the mechanistic model can simulate the peak temperature and maximum thermal stress of the BGA chip during the critical stage of reflow soldering. For the peak temperature, a process specification window is set, for example, a target range of 240-245℃, and then the degree of conformity or deviation between the simulated peak temperature and this specification window is calculated. For the maximum thermal stress, a safety threshold is set, for example, 100MPa, and then the safety margin is calculated. These converted values ​​are combined to form the second fusion index.

[0088] Furthermore, real-time features most relevant to the current evaluation objective are selected from multi-source features, such as the number of minor defects detected by upstream AOI and the standardized value of the current production line's real-time throughput. Finally, these features are weighted and combined to generate a comprehensive indicator reflecting the state of the digital twin. The digital twin state refers to a comprehensive set of virtual information that can be updated in real time and contains an accurate description of the current operating status, inherent risks, and future trends of the physical production line, formed by fusing key indicators predicted by data-driven models with implicit states and physical behaviors simulated by mechanistic models. It constitutes the decision-making basis for subsequent intelligent scheduling and adaptive control.

[0089] In addition, before the final weighted combination is executed, the dynamic weights required for this round of fusion can be dynamically updated based on the latest contextual factors (such as the complexity level of the current product model being processed) and expert experience.

[0090] Based on real-time changes and contextual factors in the production status dataset, the fusion weights from data-driven models, mechanistic models, and key real-time multi-source features are dynamically adjusted to accurately map the real production scenario under specific operating conditions. This not only comprehensively and accurately summarizes the real-time status and future trends of the production line but also provides high-quality input for subsequent heuristic scheduling optimization algorithms and adaptive control rule logic, ensuring the continuity and stability of production and the ability to respond quickly to dynamic market demands.

[0091] Using a heuristic optimization algorithm, a production scheduling scheme is generated based on the production status dataset, the digital twin status, and preset constraints; and predefined adaptive control rule logic is applied to generate executable control instructions.

[0092] Furthermore, generating a production scheduling scheme using the heuristic optimization algorithm specifically includes:

[0093] A genetic algorithm is used as the heuristic optimization algorithm.

[0094] The production scheduling problem is encoded as individuals, where each individual represents a candidate production scheduling scheme; here, the production scheduling problem is to find an optimal or near-optimal task-resource matching and sorting scheme.

[0095] A fitness function is constructed to evaluate the merits of the candidate production scheduling schemes. Based on the multi-source features and the digital twin state, the fitness function is calculated by weighted combination of scheduling objectives, wherein the scheduling objectives include work order delivery delays, predicted production yield losses, and equipment energy consumption.

[0096] Under the premise of satisfying the preset constraints, the optimal individual with the best fitness value is determined from the population containing N individuals by iteratively performing selection, crossover and mutation operations, and the optimal individual is decoded into the production scheduling scheme.

[0097] Specifically, an encoding method combining operation-based representation and machine allocation is adopted, and the individual encoding methods include:

[0098] The first part defines the global processing order of all operations in all work orders (e.g., [Op1.1, Op2.1, Op1.2, Op3.1, Op2.2], where Op1.j represents the j-th operation of work order i).

[0099] The second part specifies which specific equipment each process is assigned to.

[0100] For each candidate scheduling scheme represented by an individual, its scheduling objectives are calculated as follows:

[0101] The system obtains the planned delivery date and priority of each work order j from multi-source features, calculates the delay time of a single work order, and then assigns corresponding weights according to priority, summing them to calculate the final work order delivery delay. It also obtains the defect probability of work orders j processed on equipment k allocated to each process Op in the scheduling scheme from the digital twin state, directly obtaining the yield loss of a single process. This is then weighted and summed according to the number of work orders to obtain the predicted production yield loss. Finally, it obtains the predicted energy consumption rate per unit time for each equipment k in different operating states (e.g., processing state, idle standby state, preheating state). This predicted energy consumption rate is obtained by the digital twin state based on a comprehensive evaluation of equipment model, current equipment health status, and real-time load parameters from multi-source features. The predicted energy consumption rate per unit time is multiplied by the expected duration of each equipment k in various operating states to obtain the energy consumption of a single equipment. The energy consumption of all equipment is then summed to obtain the total equipment energy consumption.

[0102] In addition, the preset constraints include:

[0103] The first preset condition comes from the production status dataset, including whether the equipment is currently available and whether the materials are complete.

[0104] The second preset condition comes from the digital twin status, specifically whether the device's health level is higher than the safety threshold.

[0105] The fitness function evaluation relies on real-time production line data extracted from multi-source features and high-level cognitive information deeply integrated into the digital twin state, such as predicted processing time, yield rate, energy consumption rate, and equipment health status. This enables scheduling decisions to closely align with actual production conditions, accurately identify potential risks, and achieve dynamic optimization, thereby improving the flexibility and accuracy of the PCBA production process.

[0106] By deeply integrating real-time data perception, a production status dataset constructed based on multi-source features and data fusion models, and a digital twin state updated through weighted fusion calculations that combines data-driven model prediction (comprehensive production status indicators and implicit production status) and mechanism model simulation (physical behavior), combined with heuristic scheduling algorithms and adaptive control rule logic, this invention ultimately achieves optimized fully automated production. Based on a comprehensive and accurate assessment of the current and future states of the production line, this invention enables dynamic, precise, and globally optimized production scheduling, effectively addressing various disturbances. Furthermore, by anticipating quality risks and adjusting process parameters in real-time with a closed-loop mechanism, the scheduling efficiency of the PCBA production line and its adaptability to dynamic environments are further improved.

[0107] Example 2:

[0108] Based on Embodiment 1, to further verify the effectiveness of the present invention, this embodiment will further describe an intelligent scheduling optimization method for a high-precision multilayer PCBA fully automated production line, including:

[0109] Acquire physical device data streams, including structured and unstructured data;

[0110] Feature extraction is performed on the data stream of the physical equipment to generate multi-source features; feature alignment is performed based on timestamps and work order identifiers, and multi-source features of the same workpiece are fused using a data fusion model to output a production status dataset;

[0111] A data-driven model is trained and run based on the production status dataset to predict the comprehensive production status indicators and implicit production status of the production line equipment; a mechanism model is established and run to simulate the physical behavior of the production line equipment; and a digital twin model is constructed to calculate and update the digital twin status of the production line by weighted fusion of the comprehensive production status indicators, the implicit production status, and the physical behavior.

[0112] Using a heuristic optimization algorithm, a production scheduling scheme is generated based on the production status dataset, the digital twin status, and preset constraints; and predefined adaptive control rule logic is applied to generate executable control instructions.

[0113] The adaptive control rule logic continuously monitors real-time information from the production status dataset, such as current measured values ​​of process parameters, preliminary quality inspection feedback, and in-depth insights and comprehensive evaluation results provided by the digital twin status, such as the predicted comprehensive quality risk index of the current workpiece, the health score of key equipment, and the deviation between simulated physical behavior and ideal state. This information is then continuously compared with control objectives, process windows, or risk thresholds in the rule base. Once the monitored status meets the triggering conditions of a predefined rule, such as a key process parameter continuously deviating from the set center, a quality risk indicator exceeding the warning line, or an equipment health score indicating potential instability, the corresponding rule is activated. The activated rule, based on a preset adjustment strategy, automatically calculates the specific correction amount or new set value of the key process parameters to be adjusted, such as the reflow oven temperature zone, the SMT pick-and-place machine speed, and the solder paste printer squeegee pressure. The adjustment decision is then translated into a standard instruction format that the target production equipment controller can understand and execute, and sent to the corresponding equipment via the industrial communication network interface, achieving real-time, closed-loop, automated, and precise adjustment of the production process. This process enables the production line to ensure stable quality and efficient operation based on accurate assessments of current and future conditions, and to respond quickly to subtle changes and potential problems in the PCBA production line.

[0114] To verify the effectiveness of the method described in this invention, we simulated the automotive engine control unit core processing board production line described in Example 1. This production line covers key processes and equipment such as SPM, SPI, SMT, Reflow, AOI, AXI, and FCT. The simulation operation cycle was set to four consecutive weeks, and the following dynamic and uncertain factors were introduced: Regarding order arrival, the simulation included work order flows for various engine control unit models with different batch sizes, priorities (normal and expedited), and delivery period requirements; Regarding equipment failure, equipment failure events were randomly introduced based on the mean time between failures (MTBF) and mean time to repair (MTTR) of key equipment (e.g., SMT placement machine and reflow oven); Regarding process parameter drift, the simulation simulated slight drift of process parameters (such as oven temperature and placement accuracy) caused by natural aging of equipment or environmental factors.

[0115] For comparison, the following baseline scheduling and control strategies representing different technical levels were set up:

[0116] Baseline Model 1: In terms of scheduling, a simple "first-in, first-out" rule is used to determine the processing order of work orders in each process; in terms of control, all equipment process parameters (such as reflow soldering temperature profiles and pick-and-place machine speeds) adopt standard, offline fixed values ​​and are not adjusted in real time.

[0117] Baseline Model 2: During scheduling, a genetic algorithm similar to that of this invention is used for optimization. However, the targets such as processing time, yield, and energy consumption in its fitness function are calculated based on static historical average data or fixed estimates, and real-time "production status dataset" or "digital twin status" is not used for dynamic evaluation. The control method is the same as that of Baseline Model 1, using fixed process parameters.

[0118] Baseline Model 3: Scheduling is optimized based on the real-time "production status dataset" and "digital twin status" as described in this invention; while in terms of control, fixed process parameters are used, and the genetic algorithm and adaptive control logic described in this invention are not enabled.

[0119] The comparison results are shown in Table 1, obtained by running each of the above four strategies (including the method of this invention) 10 times in the set scenario and taking the average value.

[0120] Table 1 Comparison of Key Performance Indicators

[0121]

[0122] Table 2 Comparison of robustness to dynamic disturbance response

[0123]

[0124] As shown in Table 2, the performance comparison under dynamic and uncertain factors is presented. It can be seen that the present invention exhibits the best robustness in dealing with emergency order insertions and equipment failures, thus minimizing additional delays and output losses, and improving the scheduling efficiency of PCBA production lines and their adaptability to dynamic environments.

[0125] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent scheduling and optimization of a high-precision multilayer PCBA fully automated production line, characterized in that, include: Acquire physical device data streams, including structured and unstructured data; Feature extraction is performed on the data stream of the physical device to generate multi-source features; Feature alignment is performed based on timestamps and work order identifiers, and multi-source features of the same workpiece are fused using a data fusion model to output a production status dataset. The process of outputting the production status dataset includes: standardizing the multi-source features, using the timestamps and work order identifiers as association keys, and matching and aligning the multi-source features of the same workpiece within a predetermined time window using a time window matching algorithm; using a weighted fusion algorithm to weight and combine the aligned multi-source features to generate a comprehensive production status index; using a probabilistic graphical model to infer the implicit production status based on the aligned multi-source features; and including the comprehensive production status index and the implicit production status as results in the output production status dataset. A data-driven model is trained and run based on the production status dataset to predict comprehensive production status indicators and implicit production status of production line equipment; a mechanistic model is established and run to simulate the physical behavior of the production line equipment; a digital twin model is constructed to calculate and update the digital twin status of the production line by weighted fusion of the comprehensive production status indicators, the implicit production status, and the physical behavior. Specifically, establishing and running the mechanistic model includes: for selected production line equipment, establishing an ordinary differential equation model based on simplified physical principles; obtaining real-time parameters reflecting the current operating status of the equipment from the production status dataset and using these real-time parameters as initial conditions for the ordinary differential equation model; loading preset physical parameters related to the production line equipment; and using a numerical computation library to numerically solve the ordinary differential equation model, simulating and outputting the physical behavior. Using a heuristic optimization algorithm, a production scheduling scheme is generated based on the production status dataset, the digital twin status, and preset constraints; and predefined adaptive control rule logic is applied to generate executable control instructions.

2. The intelligent scheduling and optimization method for a high-precision multilayer PCBA fully automated production line according to claim 1, characterized in that, The sources of the structured data and the unstructured data include: The structured data includes equipment status and process parameter data from programmable logic controllers, physical parameter data from sensors, and production plan information from manufacturing execution systems; the unstructured data includes quality inspection data from automated optical inspection equipment.

3. The intelligent scheduling and optimization method for a high-precision multilayer PCBA fully automated production line according to claim 1, characterized in that, Generating the multi-source features includes: For the unstructured data, a lightweight convolutional neural network is used to identify and quantify welding defect features; For the structured data, signal processing methods are used to extract frequency domain features and time domain statistical features from the time-series physical parameters in the physical parameter data; principal component analysis is used to extract principal component features from the process parameter data and the non-time-series physical parameters in the physical parameter data; categorical data in the equipment status and production plan information are encoded to generate categorical features; the welding defect features are combined with the frequency domain features, the time domain statistical features, the principal component features, and the categorical features to form the multi-source features.

4. The intelligent scheduling and optimization method for a high-precision multilayer PCBA fully automated production line according to claim 1, characterized in that, The weighted fusion of the comprehensive production status indicators, the implicit production status, and the physical behavior specifically includes: The comprehensive production status index and the implicit production status are initially weighted and combined to generate a first fusion index; The physical behavior is feature extracted and converted into a second fusion index; The first fusion index, the second fusion index, and the third fusion index selected from the multi-source features are then weighted and combined again to generate and update the digital twin state.

5. The intelligent scheduling and optimization method for a high-precision multilayer PCBA fully automated production line according to claim 1, characterized in that, The specific steps involved in generating a production scheduling scheme using the heuristic optimization algorithm are as follows: A genetic algorithm is used as the heuristic optimization algorithm. The production scheduling problem is encoded into individuals, where each individual represents a candidate production scheduling solution; A fitness function is constructed to evaluate the merits of the candidate production scheduling schemes. Based on the multi-source features and the digital twin state, the fitness function is calculated by weighted combination of scheduling objectives, wherein the scheduling objectives include work order delivery delays, predicted production yield losses, and equipment energy consumption. Under the premise of satisfying the preset constraints, the optimal individual with the best fitness value is determined from the population containing N individuals by iteratively performing selection, crossover and mutation operations, and the optimal individual is decoded into the production scheduling scheme.

Images