Intelligent quality adaptive control system and method for pack production line

Abstract

The invention relates to an intelligent quality adaptive control system and method for a pack production line, and the system comprises a data collection module which collects various types of data of a plurality of key stations in the production line in real time, and forms an original production signal; the multi-source data fusion module receives, associates and processes the original production signal to form a fusion production state signal; the decision module receives the fused production state signal and generates a process parameter adjustment instruction signal for actively compensating incoming material fluctuation or correcting quality deviation in real time; the dynamic adjustment module receives a process parameter adjustment instruction signal and forms an execution feedback signal containing a parameter execution result; the optimization module receives the fusion production state signal, the process parameter adjustment instruction signal and the execution feedback signal, and continuously optimizes the machine learning model based on historical data and a feedback result. The problems that product quality consistency of a battery pack production line is poor, production depends on manual experience parameter adjustment, and quality abnormal response and root cause tracing are slow can be solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and industrial automation control technology, specifically to an intelligent quality adaptive control system and method for a pack production line. Background Technology

[0002] With the rapid development of new energy vehicles, energy storage, and other industries, the requirements for the quality, consistency, and production efficiency of battery packs, as the core power unit, are becoming increasingly stringent. Traditional battery pack production line quality control relies primarily on two levels: first, manually setting and periodically verifying key process parameters (such as welding current, tightening torque, and adhesive application amount), expecting them to operate within a fixed range; second, setting up offline or online inspection stages (such as visual inspection, electrical performance testing, and airtightness testing) at the end of production or after key workstations to screen and reject defective products. This model has significant drawbacks. First, facing inherent performance fluctuations in raw materials such as battery cells, performance degradation of production equipment over time, and dynamic interference factors such as changes in environmental temperature and humidity, fixed process parameter settings cannot achieve optimal adaptation, leading to "qualified but not excellent" products or potential defect risks, seriously affecting product consistency and long-term reliability. Second, quality inspection and process control are disconnected, representing a typical "post-production inspection" model. When quality anomalies are detected in the inspection stage, a considerable number of defective products have often already been produced, resulting in waste of materials and time, and a delayed response to anomalies. Furthermore, once batch quality problems occur, root cause analysis heavily relies on the personal experience of engineers, requiring manual backtracking of massive amounts of equipment logs, sensor data, and inspection reports. This process is time-consuming and labor-intensive, resulting in significant production stoppage losses, and the accuracy and efficiency of the analysis conclusions are difficult to guarantee. In recent years, although individual technologies such as digital twins, IoT sensing, and machine vision have been applied in the industrial field—for example, through the creation of 3D visualization models of production lines for monitoring, or the use of artificial intelligence algorithms to classify inspection images—these technologies often exist as "data silos," used only for presenting local conditions or judging single links, failing to form a deep closed loop with production control logic. That is, they can "see" problems and even "predict" some trends, but cannot autonomously and collaboratively "execute" precise adjustments to process parameters to prevent or correct problems, and lack the ability to self-iterate and optimize based on closed-loop feedback data. Therefore, the industry urgently needs an integrated system that can deeply integrate data from the entire process, achieve dynamic quality prediction and adaptive control, and automatically trace and optimize, in order to break through the bottlenecks of traditional production models in terms of quality, efficiency, and flexibility. Summary of the Invention

[0003] In view of the shortcomings of the prior art, the purpose of this invention is to provide an intelligent quality adaptive control system and method for a battery pack production line, which solves the problems of poor product quality consistency, reliance on manual experience for parameter adjustment, and slow response to quality anomalies and root cause tracing in battery pack production lines caused by fluctuations in incoming materials and changes in equipment status. This invention solves the problem by constructing a closed-loop control system and method integrating data perception, digital twin, intelligent decision-making, and execution feedback. The system collects multi-source data from the production line in real time and integrates it into a digital twin model to form a global production status view. The decision-making module analyzes this view based on a machine learning model, adjusting parameters in advance for feedforward compensation in response to fluctuations in incoming materials, or providing feedback correction for quality deviations detected in real time. The adjustment instructions are implemented by the execution module, and the results and final quality data are fed back to the optimization module. This module automatically traces the root cause when anomalies occur and continuously optimizes the decision-making model using historical feedback data, thereby enabling the system to have self-perception, self-decision-making, self-execution, and self-learning capabilities, achieving dynamic, accurate, and adaptive control of product quality.

[0004] This invention provides an intelligent quality adaptive control system and method for a pack production line, comprising: The data acquisition module collects various types of data from multiple key workstations on the production line in real time, forming raw production signals. The multi-source data fusion module receives and correlates raw production signals, synchronously mapping them to a digital twin model corresponding to the physical production line to form a fused production status signal. The decision module receives and integrates production status signals, calls the built-in machine learning model for analysis, and generates process parameter adjustment command signals for proactively compensating for fluctuations in incoming materials or correcting quality deviations in real time. The dynamic adjustment module receives process parameter adjustment command signals and converts them into control commands for specific execution equipment, driving the corresponding actuators to change the process parameters, while simultaneously generating an execution feedback signal containing the parameter execution results. The optimization module receives and integrates production status signals, process parameter adjustment command signals, and execution feedback signals. When a quality anomaly is detected, it automatically analyzes the root cause of the anomaly and generates a traceability report signal. At the same time, it continuously optimizes the machine learning model based on historical data and feedback results.

[0005] In one embodiment of the present invention, the data acquisition module collects various types of data in real time, including process parameter signals formed by physical quantities that directly reflect the production process status, visual and geometric detection signals formed by online scanning of product appearance and key dimensions based on machine vision or laser measurement technology, incoming material batch information signals formed by the current batch of raw material specifications and initial performance indicators input according to the production plan, and equipment health status signals formed by monitoring the operating status of key production equipment. These signals are packaged and preprocessed in real time according to a unified timestamp and a unique product identifier to form a standardized data package that can be directly parsed and used by subsequent modules, which together constitute the original production signal.

[0006] In one embodiment of the present invention, the multi-source data fusion module performs correlation processing on the original production signals by matching and aligning data from different sensors, workstations, and time points based on the temporal logic and spatial positional relationship of the products flowing on the production line, using a unique product identifier and a precise time synchronization mechanism. This ensures that all data can be accurately mapped to specific physical production units and individual products. Its synchronous mapping to the digital twin model means that the multi-dimensional data that has been correlated and aligned drives the corresponding entities in the virtual model in real time, enabling the virtual model to dynamically and faithfully reproduce the equipment status, material flow, and process at every moment in the physical production line, thereby generating a comprehensive, consistent, and spatiotemporally consistent fused production status signal.

[0007] In one embodiment of the present invention, the machine learning model built into the decision module is a quality prediction and decision model trained on historical production data. After receiving the integrated production status signal, the model can simultaneously perform two types of core analysis tasks: first, to predict the upcoming process results based on the current incoming material characteristics and equipment status; and second, to evaluate the status of the ongoing process based on the real-time feedback online detection results. Based on the analysis results, if the model predicts that the quality deviation may be caused by the fluctuation of incoming materials, it generates a feedforward adjustment instruction for active compensation; if it evaluates that the current process is generating a quality deviation, it generates a feedback adjustment instruction for real-time correction. The above two types of instructions together constitute the process parameter adjustment instruction signal.

[0008] In one embodiment of the present invention, the process by which the dynamic adjustment module converts the process parameter adjustment instruction signal into a control instruction for a specific execution device involves generating a low-level drive command that can be directly recognized and executed by a programmable logic controller, servo driver, or robot controller, based on the target station identifier, target process parameter type, and adjustment value contained in the instruction, combined with the control protocol and interface specification of the station actuator. While driving the actuator, the module also uses sensors to confirm the actual changes in parameters in real time, and encapsulates the confirmed actual parameter value, the status flag indicating whether the execution was successful, and the execution timestamp to form an execution feedback signal containing the parameter execution result.

[0009] In one embodiment of the present invention, the optimization module determines the quality anomaly based on continuous monitoring of a subset of online detection results in the fused production status signal. When the detection results continuously exceed the preset quality standard range or show a sudden change trend, the anomaly determination process is triggered. Its automatic analysis of the root cause of the anomaly means that the system automatically extracts all historical sequences of process parameters, raw material data, and state snapshots of each piece of equipment associated with the product during the entire production line flow process based on the product identifier and time point of the anomaly. Using causal inference algorithms or association rule mining technology, the system locates the combination of key factors most likely related to the current anomaly from multi-dimensional data, and integrates the analysis conclusions with supporting data to generate a traceability report signal.

[0010] In one embodiment of the present invention, the optimization module continuously optimizes the machine learning model based on historical data and feedback results. This means that the module establishes a closed-loop learning case library containing historical production process data, previous adjustment instructions, corresponding execution feedback, and final quality results. The optimization process uses the data in this case library to periodically or triggerically retrain and update the parameters of the machine learning model in the decision module. The goal of the update is to enable the adjustment instructions generated by the model to more accurately predict quality trends and more effectively correct deviations, thereby continuously improving the adaptive control accuracy and robustness of the system and forming a self-iterative and improving capability.

[0011] In one embodiment of the present invention, the system further includes a centralized human-machine interaction module, which is connected to the data acquisition module, the multi-source data fusion module, the decision-making module, the dynamic adjustment module, and the optimization module, respectively. This module is used to receive and visualize raw production signals, fused production status signals, process parameter adjustment command signals, execution feedback signals, and traceability report signals. At the same time, this module receives control strategy configuration information, quality target setting information, and model intervention commands input by operators, and converts this information into corresponding configuration signals and distributes them to each functional module, thereby realizing a working mode that combines automated system operation with manual monitoring and management.

[0012] In one embodiment of the present invention, the data acquisition module, multi-source data fusion module, decision-making module, dynamic adjustment module, and optimization module are not deployed in a completely centralized manner, but are deployed in a distributed manner according to the real-time requirements of data processing and computing load. The data acquisition module and dynamic adjustment module are deployed on edge computing nodes at the production line site to ensure the lowest possible latency in data acquisition and instruction execution, while the multi-source data fusion module, decision-making module, and optimization module are deployed on servers at the workshop or factory level to provide sufficient computing resources for complex data fusion, model inference, and big data analysis. The modules communicate and transmit instructions securely and reliably with each other through an industrial network.

[0013] The present invention also includes a pack production line quality adaptive control method, comprising: S1: Real-time acquisition of various types of data from multiple key workstations in the production line to form raw production signals; S2: Receive and associate raw production signals, synchronously map them to the digital twin model corresponding to the physical production line, and form a fused production status signal; S3: Collects and integrates production status signals, calls the built-in machine learning model for analysis, and generates process parameter adjustment command signals for proactively compensating for material fluctuations or correcting quality deviations in real time. S4: Receive process parameter adjustment instruction signals, convert them into control instructions for specific execution equipment, drive the corresponding actuators to change process parameters, and generate an execution feedback signal containing the parameter execution results. S5: Receives and integrates production status signals, process parameter adjustment command signals, and execution feedback signals. When a quality anomaly is detected, it automatically analyzes the root cause of the anomaly and generates a traceability report signal. At the same time, it continuously optimizes the machine learning model based on historical data and feedback results.

[0014] This invention provides an intelligent quality adaptive control system and method for a pack production line. It addresses the problem by constructing a closed-loop control system and method integrating data perception, digital twin, intelligent decision-making, and execution feedback. The system collects multi-source data from the production line in real time and integrates it into a digital twin model to form a global production status view. The decision-making module analyzes this view based on a machine learning model, adjusting parameters in advance to compensate for fluctuations in incoming materials or providing feedback correction for quality deviations detected in real time. Adjustment instructions are implemented by the execution module, and the results, along with the final quality data, are fed back to the optimization module. This module automatically traces the root cause when anomalies occur and continuously optimizes the decision-making model using historical feedback data. This enables the system to possess self-perception, self-decision-making, self-execution, and self-learning capabilities, achieving dynamic, precise, and adaptive control of product quality. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A system architecture diagram of an intelligent quality adaptive control system and method for a pack production line; Figure 2 This is a flowchart of an intelligent quality adaptive control method for a pack production line. Detailed Implementation

[0017] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0018] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0019] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0020] Please see Figure 1-2The image shows an intelligent quality adaptive control system and method for a pack production line according to the present invention. The intelligent quality adaptive control system and method for a pack production line according to the present invention includes: a data acquisition module, which collects various types of data from multiple key workstations in the production line in real time to form raw production signals; a multi-source data fusion module, which receives and correlates the raw production signals, synchronously mapping them to a digital twin model corresponding to the physical production line to form a fused production status signal; a decision module, which receives the fused production status signal, calls a built-in machine learning model for analysis, and generates process parameter adjustment instruction signals for proactively compensating for incoming material fluctuations or correcting quality deviations in real time; a dynamic adjustment module, which receives the process parameter adjustment instruction signals and converts them into control instructions for specific execution equipment, driving the corresponding actuators to change the process parameters, and simultaneously forming an execution feedback signal containing the parameter execution results; and an optimization module, which receives the fused production status signal, the process parameter adjustment instruction signal, and the execution feedback signal, and when a quality anomaly is determined, automatically analyzes the root cause of the anomaly and generates a traceability report signal, while continuously optimizing the machine learning model based on historical data and feedback results.

[0021] like Figure 1As shown, the intelligent quality adaptive control system and method for a pack production line provided by this invention constitutes a highly intelligent industrial control closed loop with self-correction and evolution capabilities in its core architecture and operating mechanism. This system completely changes the passive, lagging, and fragmented quality control process in traditional production lines. Through a series of tightly coupled technical modules, it achieves autonomous management of product quality throughout its entire lifecycle, from prediction and control to traceability and optimization. The foundation of the entire system lies in the deep perception and digital reconstruction of all production elements. Its decision-making center relies on an intelligent model that can learn from data and generate control strategies. Its ultimate value lies in applying virtual world decision commands to the physical production line without delay and with high fidelity, forming a continuous improvement cycle based on effect feedback. The following will elaborate on the first three modules of the system's core components to reveal its inherent technical logic and collaborative working method. First, the system's basic perception layer consists of modules that collect various types of data from multiple key workstations in the production line in real time to form raw production signals. This module's function extends far beyond traditional data collection; it acts as the "sensory nerve endings" of the entire system, and its design directly determines the breadth, depth, and timeliness of information upon which upper-level decisions can be based. The module's deployment covers the entire process from raw material input to final product output. The selection criteria for its "critical workstations" are based on an analysis of their impact on the final product quality, typically including core processes such as cell pretreatment, welding (e.g., laser welding, ultrasonic welding), screw tightening, adhesive sealing, electrical performance testing, and airtightness testing. At each such workstation, the module integrates and coordinates various types of sensing and data acquisition devices. These devices do not work in isolation but rather collaborate to capture data according to a unified time sequence and product identification. The collected "multi-type data" is a rich collection. Specifically, the first category is parameter data that directly drives or records the process. For example, at the welding station, this includes the current waveform, voltage value, pulse frequency, welding head pressure, and welding speed of the welding power supply; at the tightening station, it includes the real-time torque, rotation curve, and final torque value of the servo motor; at the gluing station, it includes the dispensing pressure of the glue pump, screw speed, and the width and height of the glue strip measured visually. These parameters are the direct "cause" of product quality, and their accuracy and stability are crucial. The second category is online inspection data, which reflects the immediate "effect" produced after the process is executed. This is mainly achieved through machine vision systems, laser profilometers, and high-precision sensors. For example, after welding, images of weld points or weld beads are acquired using high-speed cameras to analyze surface defects such as spatter, pits, and misalignments; geometric dimensions such as the flatness and electrode height difference of the battery module are obtained through laser scanning; and the circuit resistance after welding is measured using an online internal resistance tester.The third category is material and environmental background data, the most important of which is incoming material information accompanying the production batch flow. This includes cell batch numbers, initial open-circuit voltages, DC internal resistance classification information, and size grouping data obtained through the Manufacturing Execution System (MES) or barcode scanning. These are key inputs for production adjustment feedforward compensation. Additionally, it includes monitoring data on the health status of the production equipment itself, such as the current and vibration spectrum of servo motors, estimated wear lifespan of key moving parts, and readings from environmental temperature and humidity sensors. This data provides a basis for determining whether process fluctuations stem from equipment degradation. All these heterogeneous data streams undergo preliminary preprocessing within the acquisition module, including signal filtering to eliminate noise, data format standardization, and, most importantly, adding a millisecond-accurate timestamp and a unique identification code for the corresponding product (or battery module) to each data point. After this processing, the originally discrete and chaotic sensor readings are transformed and integrated into structured or semi-structured "raw production signals" with clear spatiotemporal attributes and semantic labels. This signal packet is no longer a bunch of incomprehensible numbers, but has become an information carrier that can clearly express "when, where, which product, what operation was performed, and what direct result was produced", laying a solid and high-quality foundation for subsequent high-level data fusion and intelligent analysis.

[0022] Secondly, the system's data hub and virtual image are achieved by the module that receives and correlates the raw production signals, synchronously mapping them to a digital twin model corresponding to the physical production line to form a fused production status signal. The core mission of this module is to solve the common "data silo" problem in industrial production and build a virtual space that can mirror and interact with the physical world in real time. Its workflow begins with deep correlation processing of the raw production signals from upstream. This "correlation processing" is not a simple data aggregation, but a complex spatiotemporal alignment and semantic fusion process. The module maintains a virtual production line timing logic model and spatial layout model. When a continuous stream of raw signals arrives, the module, like an efficient scheduler, automatically categorizes the current data from the welding station, the weld point image analysis results from the visual inspection station, and the internal resistance data of the battery cells used in the product into the same virtual product "file" based on the unique product identifier and timestamp carried by each signal, and arranges them according to the actual time sequence of the process. This ensures that when analyzing the welding quality of a product, all related antecedents (cell parameters, welding parameters) and consequences (visual inspection results) can be retrieved immediately, forming a complete causal data chain. Simultaneously, data from equipment status sensors is linked to the corresponding equipment model in the virtual production line based on the equipment number and workstation mapping. After completing the spatiotemporal association and alignment of the data, the next step is "synchronous mapping to the digital twin model." This digital twin model is not merely a 3D visualization interface, but a dynamic computational model deeply integrating physical laws, process knowledge, equipment attributes, and real-time data. This module "injects" the processed data stream into this virtual model in real time. For example, it uses real-time welding current and voltage data to drive the simulation calculation of the virtual welding process, predicting the heat-affected zone of the weld; it updates the adhesive strip size data measured by the vision system to the corresponding parts of the virtual battery pack model in real time; and it assigns the internal resistance value obtained from online testing as an attribute to the virtual product. Through this continuous, high-frequency data-driven approach, the digital twin model is no longer a static or periodically updated "snapshot," but rather a "living" mirror image that operates in parallel with the physical production line and is highly synchronized with its status. It not only shows whether equipment is running, but also whether its key process parameters are within optimal ranges; it not only shows which workstation a product has reached, but also displays the cumulative "health status" after processing at each workstation. Ultimately, the output of this module—the "fused production status signal"—is a highly integrated, context-rich, and unified data view that combines real-time performance with historical traceability. It includes both a global situational awareness of the physical production line at this moment (such as the status of all equipment, work-in-process distribution, and trends in comprehensive quality indicators) and in-depth insight into any specific product or process segment (such as retrieving the entire lifecycle data of any product).This signal provides downstream intelligent decision-making with unique, accurate, and comprehensive "battlefield intelligence," enabling decisions to be based on global transparency rather than local or partial information.

[0023] Furthermore, the core of the system's intelligent decision-making is the module that receives the fused production status signal, calls the built-in machine learning model for analysis, and generates process parameter adjustment command signals for proactively compensating for incoming material fluctuations or correcting quality deviations in real time. This module is the "brain" of the entire adaptive control system and method, and its intelligence level directly determines whether the system can achieve the leap from "perception" to "wise action." The module's work begins with receiving the aforementioned fused production status signal, which provides a panoramic input for decision-making. The module encapsulates one or more carefully designed machine learning models trained on a large amount of historical production data. These models are the "algorithm engine" for implementing intelligent decision-making. These models are not single-function; they are usually a set of models or a complex model with multi-task learning capabilities, capable of handling two types of core decision-making needs in parallel: feedforward compensation and feedback correction. When the module receives the fused signal, it first parses the incoming batch information and the product identification of the product about to enter the key process stage (such as welding). The model is then invoked, its task being to predict the upcoming process outcome based on the characteristics of the current batch of incoming materials (e.g., the average internal resistance of the cells in this batch is 5% higher than the standard value), combined with the current health status and historical performance data of the target workstation equipment. By learning the complex mapping relationship of "incoming material characteristics + process parameters -> quality results" from massive amounts of historical data, the model can infer that if the original standard welding parameters remain unchanged, the solder joint heat may be insufficient for this batch of cells with high internal resistance, leading to an increased risk of poor soldering. Based on this prediction, the model automatically generates a set of calculated new process parameter suggestions, such as increasing the welding current by a specific percentage. This instruction aims to "proactively compensate" for the inherent fluctuations in incoming materials, eliminating potential quality problems before they actually occur—this is feedforward control. On the other hand, the module is also simultaneously monitoring the online detection results contained in the fused signals in real time. For example, signals from the post-weld visual inspection station show that the solder joint spatter area of ​​several recent products has shown an increasing trend, although it has not yet exceeded the red alarm line, it already shows a clear deterioration tendency. The model will then be invoked again for real-time status assessment. It analyzes the real-time correlation between current process parameters, equipment status (such as the estimated cleanliness of the welding torch lens), and quality trends, determining that this degradation may stem from a gradual change (such as a slight decrease in shielding gas flow). The model then generates another set of adjustment commands, such as fine-tuning the pulse shape in the welding waveform or adjusting the cleaning gas pipeline, to "correct" the occurring quality deviation in real time—this falls under feedback control. Whether using feedforward or feedback decision-making, the model's analysis process is dynamic, nonlinear, and considers the coupling of multiple factors. Ultimately, the module encapsulates the model's decision output into standardized "process parameter adjustment command signals."This signal clearly indicates the target workstation (e.g., "laser welding station number three"), target parameters (e.g., "welding current"), adjustment direction and magnitude (e.g., "increase by 3.5%), and, if necessary, the adjustment priority or effective conditions. The significance of this signal lies in its transformation of data-driven insights into clear and actionable control guidelines, paving the way for ultimately changing the physical production process and achieving a crucial leap from cognitive intelligence to action intelligence.

[0024] Specifically, the module that receives the process parameter adjustment command signal, converts it into specific execution equipment control commands to drive the actuator, and generates execution feedback signals plays the role of the "executor" connecting the information domain and the physical domain. This module is the final and crucial bridge for intelligent decision-making to translate into actual productivity. Its workflow begins with receiving the adjustment command signal from the upstream decision module. This signal logically specifies "what to do," but it cannot yet be directly understood and executed by the programmable logic controller, servo driver, or robot controller on the production line. Therefore, one of the core tasks of this module is to perform "protocol conversion" and "command translation." Based on the unique identifier of the target workstation contained in the command signal, it locates the specific physical execution unit; then, based on the process parameter type (such as "welding current," "final tightening torque") and adjustment value specified in the command, combined with the specific communication protocol and data format specifications of the execution unit's control interface, it generates a set of compliant and accurate low-level drive commands. For example, for a command that requires increasing the welding current by three percent, this module will calculate the new current setpoint and encapsulate it into a data packet conforming to the Modbus TCP or Ethernet IP protocol format of the specific welding power controller. This process ensures that high-level intelligent decisions can be safely and reliably delivered to various heterogeneous devices in the industrial field. Next, the module sends the generated control commands to the target actuators via the industrial network. However, in industrial control, "issuing a command" does not equate to "perfect execution of the command." The actuator may fail to reach the expected state due to mechanical jamming, communication delays, or its own malfunction. Therefore, the module simultaneously activates a verification process for the execution results at the same time as or after issuing the command. It confirms whether the process parameters have changed as required by the command and what the actual values ​​are after the change by accessing or listening to feedback data from relevant sensors (such as the sampled value of the actual output current of the welding power source and the actual torque feedback of the torque gun). This verification information, along with the start and end times of the command execution and a status code indicating whether the execution was successful or not, or the type of anomaly, is encapsulated in real time by the module to form a detailed "execution feedback signal." This signal is crucial; it accurately records the gap between the decision intention and the actual implementation effect, providing first-hand evidence for subsequent evaluation of the decision's effectiveness and diagnosis of problems in the execution process, making the entire control loop verifiable and traceable.

[0025] In one embodiment of the invention, a module that receives multiple signals and automatically analyzes the root cause when a quality anomaly is determined, while continuously optimizing the machine learning model, serves as the intelligent central hub of the system, possessing the ability to "reflect" and "evolve." This module is like a combination of a quality analyst and an algorithm engineer with ever-growing experience. It continuously monitors upstream integrated production status signals, process parameter adjustment command signals, and execution feedback signals, forming a coherent data flow about the entire process of "system perception-decision-action." Its logic for "determining quality anomalies" does not simply rely on the exceeding of a single detection result, but adopts a more intelligent and forward-looking strategy. It comprehensively utilizes the ideas of statistical process control and pattern recognition technology. For example, it performs trend analysis on continuous measurements of key quality characteristics (such as weld pull-out force and sealing test leakage rate) to identify potential anomalies that, although within specification limits, are continuously drifting towards the boundary; or it uses a machine learning model to calculate the quality score under the current production status in real time, triggering an alert when the score rapidly declines or falls below a certain dynamic threshold. Once an anomaly is confirmed, the module immediately starts an automated root cause analysis engine. This process simulates the troubleshooting approach of a senior engineer, but its speed and data breadth far exceed those of human intervention. The engine uses the abnormal product and its production time as anchors to automatically trace back all the process steps the product has undergone since its launch from the historical data of the digital twin model. It doesn't simply list all the data; instead, it uses causal inference algorithms (such as causal graph-based inference and Granger causality tests) or efficient association rule mining techniques to find the most statistically significant combination of factors in massive amounts of data that is most correlated with the current anomaly pattern. These factors might include: the voltage fluctuation pattern of a welding machine during a specific time period, the distribution characteristics of a certain parameter of a batch of incoming battery cells, or the temporal correlation between a sudden change in an environmental parameter (such as humidity) and the occurrence of the anomaly. After analysis, the module generates a structured "traceability report signal," which not only points out the most likely root cause (e.g., "the welding voltage module at station 3 experienced intermittent instability during the abnormal time period"), but also includes key data curves and correlation indicators supporting this conclusion, greatly reducing fault diagnosis time. Simultaneously, another parallel and continuous task of this module is to optimize the machine learning model. It systematically builds and maintains a "closed-loop learning case library," where each record fully encompasses the input (production status), decision (adjustment instructions), action (execution feedback), and final result (product quality confirmation) of a control loop. The model optimization process is initiated periodically or based on specific triggering conditions (such as accumulating a sufficient number of new cases). It utilizes this ever-expanding case library to retrain or fine-tune the model in the decision-making module.The optimization goals are multi-dimensional: they could be to improve the model's prediction accuracy for quality deviations caused by incoming material fluctuations, to improve the effectiveness of the model in generating corrective instructions under specific equipment degradation modes, or to enhance the model's robustness under different production rhythms. Through this continuous learning based on its own practical data, the system can gradually adapt to the unique characteristics of the production line and even anticipate slow changes that have not yet been noticed by operators (such as equipment aging), thereby making its decisions more and more accurate and more "adept".

[0026] In one embodiment of the present invention, the system constitutes a complete adaptive control closed loop, which is the ultimate manifestation and value proposition of the collaborative work of all its technical modules. This closed loop is not a simple cycle, but a dynamic equilibrium system formed by the deep coupling of an intelligent agent with four functions: perception, decision-making, execution, and learning, and the physical production process. Its closed-loop process begins with capturing the most subtle changes in the physical world: the data acquisition module, like a keen sense, continuously converts multidimensional data from the production line into raw signals. Next, the multi-source data fusion module acts as the nervous system, comprehensively processing this sensory information and constructing a three-dimensional "production situation map" synchronized with reality in the brain (decision-making module). The decision-making module, as the core of the brain, reasons and judges based on this situation map and its understanding of historical experience (built-in models). When it anticipates that changes in incoming materials may lead to future quality risks, it issues feedforward compensation instructions in advance; when it identifies unfavorable trends in real-time quality indicators from the situation map, it immediately issues feedback correction instructions. These decision-making instructions are accurately transmitted and translated into specific actions of the equipment through the dynamic adjustment module, which acts as both a "motor nerve" and an "effector," directly intervening in the physical production process and altering the product's processing conditions. Subsequently, the adjustment module sends back feedback signals regarding the execution status of the instructions (success or failure, and actual results). Simultaneously, a new round of data acquisition begins, capturing the new results generated by the changes in process parameters and forming a new fusion state signal. The optimization module, like the part of the brain responsible for memory and experience summarization, examines each complete case of "perception-decision-execution-result" from a higher time scale. It not only deeply analyzes and traces the root causes of sudden quality anomalies, but more importantly, it silently archives all successful and unsuccessful experiences, using them to continuously train and optimize the model in the decision-making module, making future decisions more intelligent. This closed loop operates in real-time and continuously. This enables the system to proactively respond to rather than passively withstand various disturbances: when the characteristics of incoming material batches change, the system can adaptively adjust process parameters to compensate; when the performance of key equipment slowly declines, the system can offset or provide early warnings by fine-tuning other parameters; when environmental factors fluctuate, the system can find the optimal operating point under new conditions. The entire system thus evolves from a static automated device requiring constant manual intervention and parameter setting into a dynamic intelligent system capable of self-adjustment and self-optimization, aiming to maintain stable and optimal output quality.

[0027] like Figure 1As shown, the system also includes a module that connects all functional modules to achieve centralized information management and human interaction. This design ensures efficient collaboration between the intelligent system and human operators. This module serves as the unified user-facing window for the entire adaptive control system and method, undertaking the key functions of information aggregation, visualization, configuration management, and human-machine interaction. It connects in real-time with various functional modules through data interfaces, continuously receiving raw production signal snapshots from the data acquisition module, panoramic digital twin status images from the multi-source data fusion module, real-time adjustment instructions and interpretability analysis of their decision-making basis from the decision-making module, instruction execution status feedback from the dynamic adjustment module, and anomaly tracking reports and model optimization progress notifications from the optimization module. One of the core functions of this module is to transform this massive, multi-dimensional, and high-frequency industrial data into an information format that human operators, process engineers, and managers can intuitively understand and quickly grasp. This is typically achieved through a high-definition graphical human-machine interface (HMI), such as displaying real-time animations of a 3D virtual model of the production line, dynamically labeling the equipment status and process parameters of each workstation with colors and values; drawing trend control charts and quality indicator dashboards for key quality characteristics; displaying intelligent adjustment instructions that are being executed or queued in the form of lists or flowcharts; and presenting the analytical conclusions of anomaly traceability reports using clear cause-and-effect diagrams or timeline views. Beyond "presentation," this module's more important function is receiving "input." It provides users with a channel for supervising, guiding, and providing advanced intervention to the intelligent system. Process engineers can use this interface to input or modify the corresponding quality standard targets and tolerance ranges for producing different product models; equipment maintenance personnel can view potential health problems alerted by the optimization module and enter maintenance records; in special circumstances, experienced operators can veto or correct automatically generated adjustment instructions or directly issue manual control commands. All this manually input information is converted by the module into standard configuration signals or intervention instructions and safely and reliably distributed to the corresponding functional modules, such as updating rule constraints in the decision-making module or sending emergency stop commands to the dynamic adjustment module. This design achieves a perfect combination of automation and human experience: in most routine and predictable scenarios, the system operates autonomously and efficiently, ensuring quality and efficiency; when encountering extreme anomalies, entirely new operating conditions, or situations requiring strategic judgment based on deep human knowledge, human experts can easily intervene and take control of the overall situation. The existence of this module transforms advanced intelligent control systems and methods from "black boxes" into transparent, reliable, and collaborative productivity partners, greatly enhancing the system's practicality and acceptability.

[0028] like Figure 2As shown, S1: Real-time acquisition of various types of data from multiple key workstations in the production line to form raw production signals; S2: Receiving and processing the raw production signals, synchronously mapping them to the digital twin model corresponding to the physical production line to form a fused production status signal; S3: Receiving the fused production status signal, calling the built-in machine learning model for analysis, and generating process parameter adjustment instruction signals for proactively compensating for incoming material fluctuations or correcting quality deviations in real time; S4: Receiving the process parameter adjustment instruction signals and converting them into control instructions for specific execution equipment, driving the corresponding actuators to change the process parameters, and simultaneously forming an execution feedback signal containing the parameter execution results; S5: Receiving the fused production status signal, process parameter adjustment instruction signals, and execution feedback signals, when a quality anomaly is determined, automatically analyzing the root cause of the anomaly and generating a traceability report signal, while continuously optimizing the machine learning model based on historical data and feedback results.

[0029] Specifically, the unit within the method that connects all functional modules to achieve centralized information management and human interaction is a key hub ensuring seamless integration of intelligent methods with human experience and management will. This unit is not a simple data display screen, but a comprehensive interactive platform integrating data relay, visualization rendering, configuration management, command routing, and audit logs. It establishes bidirectional and stable communication links with data acquisition, data fusion, intelligent decision-making, dynamic execution, and self-optimization units through standardized industrial data interfaces or service buses. The data flowing from each functional unit to this interactive platform is massive and multidimensional: raw production signals provide the lowest-level real-time data stream, like unedited live video; fused production status signals provide a deeply processed global situational map with rich semantic tags, like a live broadcast with data annotations and analytical explanations; process parameter adjustment command signals and the interpretable information behind them, such as decision confidence and model basis, reveal the method's "thinking process"; execution feedback signals report the "echoes" of the commands in the physical world; and traceability reports and model optimization progress signals convey the results of the method's self-diagnosis and learning. One of the core missions of this interactive platform is to transform the deluge of machine-friendly but human-obscure data into highly intuitive, clearly structured visual information that supports rapid decision-making. This is typically achieved through a multi-view collaborative graphical interface: a 3D panoramic view based on digital twins, displaying real-time equipment status, material flow, and process parameter fluctuations in a virtual-real overlay manner; a series of customizable monitoring dashboards and trend charts focusing on key quality indicators, overall equipment efficiency, and energy consumption; a dynamically updated instruction queue and execution status list, transparently displaying the autonomous control activities of the method; and a message center integrating alarms, traceability reports, and knowledge push notifications. More importantly, the platform provides a powerful human-machine collaboration channel. It allows process experts to inject domain knowledge into the method through a user-friendly configuration interface, such as defining process routes for different product models, setting or fine-tuning the weights and tolerance ranges of quality objectives, and establishing contingency plans for specific incoming material anomalies. When the method encounters extremely rare operating conditions or when management decisions require intervention, authorized personnel can use the platform for in-depth intervention, such as temporarily switching control modes, approving or rejecting major parameter adjustment proposals put forward by the method, or even directly issuing manual operation instructions. These human inputs are transformed into structured configuration signals or higher-priority control commands, securely injected into the method loop, ensuring that human intelligence can drive and guide automated intelligence at critical moments. Furthermore, the platform also undertakes complete operational auditing and knowledge accumulation functions, recording the complete operational chain, decision context, and results of all automated decisions and human interventions, forming a traceable and reproducible digital archive.The existence of this unit completely breaks down the barrier that intelligent control methods are often regarded as "black boxes," and builds a transparent, trustworthy, and collaborative human-machine co-governance environment. This enables cutting-edge intelligent technologies to be understood, trusted, and efficiently utilized by management and technical personnel on the production site, greatly improving the practical value and acceptability of the methods.

[0030] Furthermore, the distributed and collaborative deployment of the method's functional units based on real-time data processing requirements and computational load is the technical guarantee for its ability to cope with the stringent requirements of complex industrial environments. This deployment strategy profoundly embodies the computing philosophy of "edge-cloud" collaboration, aiming to place the appropriate computing tasks in the most suitable locations for execution, thereby globally optimizing response speed, network load, reliability, and cost. Specifically, the two units that directly interface with the physical world and have extremely stringent real-time requirements—the unit responsible for data acquisition and the unit responsible for dynamic execution—are preferentially deployed in edge computing nodes or high-performance industrial controllers near the production line. This deployment allows the acquisition, preliminary filtering, and timestamping of raw production signals to be completed within microseconds to milliseconds, capturing subtle changes in production moments with almost no delay. Simultaneously, the process of converting process parameter adjustment instructions into equipment drive commands and executing them is also completed locally at extremely high speed, ensuring the accuracy and timeliness of control actions, which is crucial for the quality stability of high-speed continuous processes such as welding and gluing. The deployment of edge nodes effectively isolates the impact of workshop network fluctuations on the core control loop, enhancing the method's local autonomy and robustness. On the other hand, units with huge computing resource requirements but less stringent real-time requirements—responsible for multi-source data fusion, intelligent decision-making, and self-optimization—are deployed on workshop-level or factory-level servers or private cloud platforms. These units aggregate data from various production lines and edge nodes, providing ample computing power for building enterprise-level digital twins, conducting large-scale historical data correlation analysis, and running complex machine learning model training and inference. For example, tasks such as real-time rendering and physical simulation of high-fidelity digital twin models, retraining deep learning models based on massive case libraries, and cross-batch, cross-product quality root cause analysis all operate efficiently at this level. The distributed units are connected via highly reliable, low-latency industrial networks (such as Time-Sensitive Networking (TSN) and Industrial Ethernet) and secure communication protocols (such as OPC UA), ensuring the secure, orderly, and efficient flow of data and control commands between the edge and the center. This architecture not only optimizes the allocation of computing resources but also facilitates method expansion and maintenance: new production lines or workstations can be easily connected by adding edge nodes; algorithm model upgrades can be centralized on the server side, eliminating the need for tedious updates to every field device; and the reliability of the method is enhanced by the distributed functionality, making it less likely for a failure of a local edge node to paralyze the entire method. This distributed collaborative deployment model enables the method to flexibly adapt to different scale application scenarios, from small production lines to large digital factories.

[0031] Specifically, all technical units work together to form a complete, dynamically evolving intelligent adaptive control closed loop, which is the ultimate form and core value of the technical solution of this invention. This closed loop is not a simple circular process, but a symbiotic method formed by the deep integration of an "industrial intelligent agent" with perception, cognition, decision-making, action, and learning capabilities with the physical production line. Its operation manifests as a never-ending, self-reinforcing intelligent cycle. The starting point of the cycle is the method's continuous perception of the most authentic state of the physical world: the sensor network throughout the production line is like the method's senses, converting welding sparks, tightening sounds, the form of colloids, and subtle changes in electrical properties into bit streams in the data world, forming the most primitive cognitive material. Immediately afterwards, this material is rapidly sent to the data fusion unit, which is like the method's cerebral cortex, responsible for spatiotemporal alignment, correlation integration, and meaning construction of multi-source heterogeneous information, generating a highly faithful "digital mirror" that breathes in sync with the physical production line. This global, structured situation map is the foundation for the method to carry out all advanced cognitive activities. Based on this situational map, the intelligent decision-making unit begins operation. Like the prefrontal cortex of the method's brain, it is responsible for reasoning, prediction, and planning. It invokes an embedded, trained empirical model (machine learning model) to assess the current production state: on the one hand, it proactively predicts impending quality risks (such as identifying minor deviations in material parameters that might cause fluctuations in downstream processes) and generates preventative feedforward compensation instructions; on the other hand, it diagnoses subtle deviations in real time (such as capturing unfavorable trends in quality characteristics from online inspection data) and generates corrective feedback adjustment instructions. These decision instructions are then transmitted to the dynamic execution unit, which acts like the method's spinal cord and motor neurons, responsible for translating abstract strategic intentions into concrete, executable mechanical actions. It precisely controls the power of the welding torch, the torque of tightening the torch, and the opening of the glue-applying valve, directly altering the "cause" of the production process. After the action, the method immediately collects the "effect" of the action through the perception layer, forming execution feedback, while simultaneously initiating a new round of perception to observe how the world has changed due to its actions. The optimization unit, much like the hippocampus and neocortex responsible for memory, reflection, and evolution, stands at a higher temporal dimension, calmly examining each complete case of "perception-cognition-decision-action-result." For successful cases, it extracts experience and incorporates it into the model; for abnormal or failed cases, it initiates deep root cause analysis, not only quickly locating the source of the problem to support immediate handling, but also transforming the lessons learned into nutrients to make the method smarter, used for iterative optimization of the decision-making model. Thus, the entire closed loop achieves a complete value sublimation from data to information, to knowledge, and then to intelligent action. It transforms the production line from a static machine requiring constant manual adjustments into an intelligent living entity capable of proactively adapting to raw material fluctuations, automatically compensating for equipment performance drift, dynamically optimizing process parameters, and accumulating wisdom through continuous operation, becoming increasingly precise with use.The ultimate goal of this closed loop is to achieve a fundamental leap in production quality from "meeting standards" to "continuously stable and optimal," providing core technological support for modern manufacturing with high complexity and high consistency requirements.

[0032] This invention discloses an intelligent quality adaptive control system and method for a pack production line. It addresses the problem by constructing a closed-loop control system and method integrating data perception, digital twin, intelligent decision-making, and execution feedback. The system collects multi-source data from the production line in real time and integrates it into a digital twin model to form a global production status view. The decision-making module analyzes this view based on a machine learning model, adjusting parameters in advance to compensate for fluctuations in incoming materials or providing feedback correction for quality deviations detected in real time. Adjustment instructions are implemented by the execution module, and the results, along with the final quality data, are fed back to the optimization module. This module automatically traces the root cause when anomalies occur and continuously optimizes the decision-making model using historical feedback data. This enables the system to possess self-perception, self-decision-making, self-execution, and self-learning capabilities, achieving dynamic, precise, and adaptive control of product quality.

[0033] Therefore, the intelligent quality adaptive control system and method for battery pack production lines of the present invention solves the problems of poor product quality consistency caused by fluctuations in incoming materials and changes in equipment status, reliance on manual experience for parameter adjustment, and slow response to quality anomalies and root cause tracing in battery pack production lines.

[0034] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. An intelligent quality adaptive control system for a pack production line, characterized in that, include: The data acquisition module collects various types of data from multiple key workstations on the production line in real time to form raw production signals. A multi-source data fusion module receives and associates the original production signal, synchronously mapping it to a digital twin model corresponding to the physical production line to form a fused production status signal. The decision module receives the fused production status signal, calls the built-in machine learning model for analysis, and generates process parameter adjustment instruction signals for actively compensating for incoming material fluctuations or correcting quality deviations in real time. The dynamic adjustment module receives the process parameter adjustment instruction signal and converts it into a control instruction for the specific execution device, drives the corresponding actuator to change the process parameters, and generates an execution feedback signal containing the parameter execution result. The optimization module receives the integrated production status signal, process parameter adjustment instruction signal, and execution feedback signal. When a quality anomaly is determined, it automatically analyzes the root cause of the anomaly and generates a traceability report signal. At the same time, it continuously optimizes the machine learning model based on historical data and feedback results.

2. The intelligent quality adaptive control system for a pack production line according to claim 1, characterized in that, The data acquisition module collects various types of data in real time, including process parameter signals formed by physical quantities that directly reflect the production process status, visual and geometric detection signals formed by online scanning of product appearance and key dimensions based on machine vision or laser measurement technology, incoming material batch information signals formed by the current batch of raw material specifications and initial performance indicators input according to the production plan, and equipment health status signals formed by monitoring the operating status of key production equipment. These signals are packaged and preprocessed in real time according to a unified timestamp and a unique product identifier to form a standardized data package that can be directly parsed and used by subsequent modules, which together constitute the original production signal.

3. The intelligent quality adaptive control system for a pack production line according to claim 1, characterized in that, The multi-source data fusion module performs correlation processing on the original production signals. This means that, based on the temporal logic and spatial positional relationship of the products flowing on the production line, data from different sensors, different workstations, and different time points are matched and aligned through unique product identifiers and a precise time synchronization mechanism. This ensures that all data can accurately correspond to specific physical production units and individual products. Its synchronous mapping to the digital twin model means that the correlated and aligned multi-dimensional data drives the corresponding entities in the virtual model in real time. This enables the virtual model to dynamically and faithfully reproduce the equipment status, material flow, and process at every moment in the physical production line, thereby generating a comprehensive, consistent, and spatiotemporally consistent fused production status signal.

4. The intelligent quality adaptive control system for a pack production line according to claim 1, characterized in that, The machine learning model built into the decision module is a quality prediction and decision model trained on historical production data. After receiving and integrating production status signals, the model can simultaneously perform two types of core analysis tasks: first, to make advance predictions of the upcoming process results based on the current incoming material characteristics and equipment status; and second, to evaluate the status of the ongoing process based on real-time feedback of online detection results. Based on the analysis results, if the model predicts that fluctuations in incoming materials may lead to quality deviations, it generates a feedforward adjustment command for proactive compensation. If it assesses that the current process is causing quality deviations, it generates a feedback adjustment command for real-time correction. The two types of commands together constitute the process parameter adjustment command signal.

5. The intelligent quality adaptive control system for a pack production line according to claim 1, characterized in that, The process by which the dynamic adjustment module converts the process parameter adjustment instruction signal into a control instruction for a specific execution device is to generate a low-level drive command that can be directly recognized and executed by a programmable logic controller, servo driver, or robot controller, based on the target station identifier, target process parameter type, and adjustment value contained in the instruction, combined with the control protocol and interface specification of the station actuator. While driving the actuator, this module uses sensors to confirm the actual changes in parameters in real time, and encapsulates the confirmed actual parameter values, the status flag indicating whether the execution was successful or not, and the execution timestamp to form the execution feedback signal containing the parameter execution result.

6. The intelligent quality adaptive control system for a pack production line according to claim 1, characterized in that, The optimization module determines quality anomalies based on continuous monitoring of a subset of online detection results in the integrated production status signal. When the detection results continuously exceed the preset quality standard range or show a sudden change trend, the anomaly determination process is triggered. The automatic analysis of the root cause of anomalies refers to the system automatically extracting all historical sequences of process parameters, raw material data, and state snapshots of each piece of equipment associated with the product during its entire production line flow, based on the product identifier and time point of the anomaly. Using causal inference algorithms or association rule mining techniques, the system locates the combination of key factors most likely related to the anomaly from multi-dimensional data, and integrates the analysis conclusions with supporting data to generate the traceability report signal.

7. The intelligent quality adaptive control system for a pack production line according to claim 1, characterized in that, The optimization module continuously optimizes the machine learning model based on historical data and feedback results. This means that the module establishes a closed-loop learning case library containing historical production process data, previous adjustment instructions, corresponding execution feedback, and final quality results. The optimization process uses the data in this case library to periodically or triggerically retrain and update the parameters of the machine learning model in the decision module. The goal of the update is to enable the adjustment instructions generated by the model to more accurately predict quality trends and more effectively correct deviations, thereby continuously improving the adaptive control accuracy and robustness of the system and forming a self-iterative and improving capability.

8. The intelligent quality adaptive control system for a pack production line according to claim 1, characterized in that, The system also includes a centralized human-machine interaction module, which is connected to the data acquisition module, multi-source data fusion module, decision-making module, dynamic adjustment module, and optimization module. This module receives and visualizes the original production signals, fused production status signals, process parameter adjustment command signals, execution feedback signals, and traceability report signals. Simultaneously, this module receives control strategy configuration information, quality target setting information, and model intervention commands input by operators, and converts this information into corresponding configuration signals to distribute to each functional module, thereby realizing a working mode that combines automated system operation with manual monitoring and management.

9. The intelligent quality adaptive control system for a pack production line according to claim 1, characterized in that, The data acquisition module, multi-source data fusion module, decision-making module, dynamic adjustment module, and optimization module are not deployed in a centralized manner, but rather in a distributed manner based on the real-time requirements of data processing and the computing load. The data acquisition module and dynamic adjustment module are deployed on edge computing nodes at the production line site to ensure minimal latency in data acquisition and instruction execution, while the multi-source data fusion module, decision-making module, and optimization module are deployed on servers at the workshop or factory level to provide sufficient computing resources for complex data fusion, model inference, and big data analysis. The modules communicate and transmit instructions securely and reliably with each other through an industrial network.

10. A method for an intelligent quality adaptive control system for a pack production line according to any one of claims 1-9, comprising: S1: Real-time acquisition of various types of data from multiple key workstations in the production line to form raw production signals; S2: Receive and associate the original production signal, and synchronously map it to the digital twin model corresponding to the physical production line to form a fused production status signal; S3: Receive the integrated production status signal, call the built-in machine learning model for analysis, and generate process parameter adjustment instruction signals for actively compensating for incoming material fluctuations or correcting quality deviations in real time. S4: Receive the process parameter adjustment instruction signal, convert it into a control instruction for the specific execution device, drive the corresponding actuator to change the process parameters, and at the same time generate an execution feedback signal containing the parameter execution result; S5: Receive the integrated production status signal, process parameter adjustment instruction signal and execution feedback signal. When a quality abnormality is determined, automatically analyze the root cause of the abnormality and generate a traceability report signal. At the same time, continuously optimize the machine learning model based on historical data and feedback results.

Images