Intelligent baking production line control method and system based on multi-process cooperation

Abstract

The application relates to the field of baking production line control, in particular to an intelligent baking production line control method and system based on multi-process cooperation. The method comprises the following steps: acquiring multi-source real-time data of a production line, based on which, a multi-component state evolution model of multi-component structure baked goods is constructed and synchronized with digital twinning, and a real-time digital twinning data set is generated; acquiring baking production order data, based on which, a golden window period of multi-component state synchronization is calculated, and multi-component state reverse synchronization planning is carried out with the window period as a constraint, and a synchronous process planning instruction set is generated; based on this, urgency weighted scheduling and resource conflict pre-arbitration are carried out, and an asynchronous cooperative scheduling instruction set is generated; based on this, multi-component state readiness joint judgment and synchronization triggering are carried out, and an intelligent control report of the baking production line is generated. In the baking production line control process, the application realizes accurate synchronization and flexible scheduling of asynchronous processes of multi-component baked goods.

Description

Technical Field

[0001] This application relates to the field of baking production line control, and in particular to an intelligent baking production line control method and system based on multi-process collaboration. Background Technology

[0002] In the modern food industry, especially in the field of high-end baking, intelligent production lines are the core carriers for achieving large-scale customization, ensuring product quality consistency, and improving production efficiency. The precision and coordination capabilities of their control systems directly determine the flavor, taste, and output stability of the final product, and are the key technological foundation for promoting the transformation of the baking industry towards flexible manufacturing and digital upgrading.

[0003] However, existing baking production line control methods lack a mechanism for real-time evolution modeling and twin synchronization of the physical state of each independent component when dealing with structured products with multiple components, multiple processes and complex process paths. This not only makes it difficult to accurately plan and coordinate the optimal process convergence window between components, but also makes it easy for production rhythm to be disrupted, products to be scrapped or quality to decline in a dynamic production environment due to resource competition or state mismatch. This restricts the production line's potential to cope with high mixed orders and refined management. Summary of the Invention

[0004] This application provides a control method and system for an intelligent baking production line based on multi-process collaboration to solve the above-mentioned technical problems.

[0005] In a first aspect, this application provides a control method for an intelligent baking production line based on multi-process collaboration, the method comprising: The system acquires multi-source real-time data from the production line. Based on this data, it constructs a multi-component state evolution model and synchronizes it with a digital twin for multi-component baked goods, generating a real-time digital twin dataset. It also acquires baking production order data, calculates the optimal window for multi-component state synchronization based on the real-time digital twin dataset and the baking production order data, and performs reverse synchronization planning of multi-component states constrained by this window, generating a synchronized process planning instruction set. Based on the real-time digital twin dataset and the synchronized process planning instruction set, it performs urgency-weighted scheduling and resource conflict pre-arbitration, generating an asynchronous collaborative scheduling instruction set. Finally, based on the real-time digital twin dataset and the asynchronous collaborative scheduling instruction set, it performs joint judgment and synchronization triggering of multi-component state readiness, generating a smart control report for the baking production line.

[0006] Through the above technical solutions, precise synchronization and flexible scheduling of asynchronous processes in multi-component baking are achieved, significantly improving assembly efficiency and resource utilization. Through state-driven intelligent control, the high consistency and stability of the final product quality are ensured. At the same time, the system enhances the production line's ability to respond quickly to order changes and emergency orders, and promotes the overall advancement of baking production towards intelligence and self-adaptation.

[0007] Optionally, the generation process of the real-time digital twin dataset includes: based on the multi-source real-time data of the production line, collecting core physical state data that determines the final availability of each independent component constituting the multi-component structured baked goods in parallel; analyzing the evolution of the core physical state of each independent component over time under different process treatments, constructing a state evolution prediction model for each independent component category; and dynamically comparing and calibrating the multi-source real-time data of the production line with the state evolution prediction model data to achieve synchronization between the physical production line and the virtual model, thereby generating the real-time digital twin dataset containing the real-time state of components, state prediction trajectories, and process resource states.

[0008] Optionally, the core physical state data includes: for the pastry or bread dough layer components that constitute the main body of the product, collecting their temperature and texture state that characterizes crispness or softness; for the filling or sauce layer components, collecting their temperature and viscosity state that characterizes flow or shaping ability; and for decorative components, collecting their temperature and hardness or shaping state that characterizes structural stability.

[0009] Optionally, the construction and calibration process of the state evolution prediction model includes: analyzing the historical state data of each independent component under specific heating, cooling, stirring, or settling processes, summarizing the dynamic change rules of the core physical state of the independent component from the initial value to the process target value, and establishing an initial state evolution prediction model for each category of independent components; inputting the real-time collected core physical state data into the initial state evolution prediction model during the current production process to obtain a predicted state sequence; and dynamically adjusting the response parameters of the initial state evolution prediction model in reverse by continuously comparing the deviation between the predicted state sequence and the actual state sequence, so that the predicted trajectory output by the model can adaptively fit the real evolution process under the current raw material characteristics, equipment operating conditions, and environmental conditions, thus completing the online calibration of the model.

[0010] Optionally, the process of generating the synchronous process planning instruction set includes: extracting the target product type from the baking production order data, and retrieving the target state window requirements of each component at the final assembly time of the target product from the baking knowledge base, wherein the target state window defines the range of core physical state parameters suitable for assembly of each component; based on the state evolution prediction model, forward deducing the shortest and longest process processing time required for each component to reach its respective target state window from the current state; with all components reaching the target state as a constraint, calculating the maximum common overlap time interval of all forward deduction time intervals, and solving this interval as the golden window period for this production; taking the starting point of the golden window period as the synchronization trigger time point, and using the state evolution prediction model, backward deducing the precise state trajectory and process start sequence that each component must follow to achieve synchronization from the synchronization trigger time point, and generating the synchronous process planning instruction set.

[0011] Optionally, calculating the maximum common overlap time interval of all forward extrapolation time intervals includes: when there is a direct overlap in the process processing time intervals of each component, directly solving the direct overlap portion as the golden window period; when there is no direct overlap in the process processing time intervals, applying several process parameter fine-tuning instructions to the state evolution prediction model, the process parameter fine-tuning instructions being used to adjust the process processing intensity or path of several components within the allowable quality deviation range, so as to change the time interval required for them to reach the target state window; recalculating the overlap portion based on the fine-tuned process processing time intervals, and solving the first successfully obtained direct overlap portion, or the optimal overlap portion obtained after a finite number of fine-tunings, as the golden window period for this production.

[0012] Optionally, the generation process of the asynchronous collaborative scheduling instruction set includes: when multiple process paths in the synchronous process planning instruction set compete for the same key resource, based on the baking production order data, order urgency is introduced as an arbitration weight; for orders with urgency higher than a preset urgency threshold, the process scheme that can shorten the total time to reach the target state is selected first in its competing path, and a higher resource preemption priority is given, generating a resource preemption scheduling scheme; for orders with urgency lower than or equal to the preset urgency threshold, the process scheme with the most stable state evolution and the lowest quality fluctuation risk is selected first in its competing path, and concessions are made in resource allocation, generating a robust avoidance scheduling scheme; the asynchronous collaborative scheduling instruction set is generated by combining the resource preemption scheduling scheme and the robust avoidance scheduling scheme.

[0013] Optionally, the process of generating the intelligent control report for the baking production line includes: integrating the actual production data after executing the asynchronous collaborative scheduling instruction set, quantitatively evaluating the dynamic deviation between the actual evolution trajectory of each order component's status and the corresponding predicted trajectory in the synchronous process planning instruction set; based on the dynamic deviation, calculating the overall synchronization efficiency index and resource conflict resolution success rate of this production cycle, and identifying the key process steps that cause the greatest deviation; and structurally encapsulating the overall synchronization efficiency index, the resource conflict resolution success rate, and the identification information of the key process steps to generate the intelligent control report for the baking production line.

[0014] Optionally, the method further includes: responding to a real-time customized adjustment instruction received after the calculation of the golden window period for the baking production order, the real-time customized adjustment instruction including modification of the target state window requirement for at least one component; based on the modified target state window requirement, partially updating the state target parameters of the affected component in the state evolution prediction model; re-executing the calculation of the forward inference and the maximum common overlap time interval with the updated state evolution prediction model, dynamically calculating and refreshing the golden window period and the synchronous process planning instruction set for this production; performing conflict detection and re-coordination on the refreshed synchronous process planning instruction set and the asynchronous collaborative scheduling instruction set, generating and issuing updated production control instructions, so that the production line can adapt to customized changes without interrupting the overall process.

[0015] Secondly, this application provides a control method for an intelligent baking production line based on multi-process collaboration, the system comprising: The digital twin module is used to acquire multi-source real-time data from the production line. Based on this data, it constructs a multi-component state evolution model for multi-component baked goods and synchronizes it with the digital twin, generating a real-time digital twin dataset. The synchronization instruction module is used to acquire baking production order data. Based on the real-time digital twin dataset and the baking production order data, it calculates the golden window period for multi-component state synchronization and performs reverse synchronization planning of multi-component states constrained by this window period, generating a synchronized process planning instruction set. The asynchronous instruction module is used to perform urgency-weighted scheduling and resource conflict pre-arbitration based on the real-time digital twin dataset and the synchronized process planning instruction set, generating an asynchronous collaborative scheduling instruction set. The control report module is used to perform joint judgment and synchronization triggering of multi-component state readiness based on the real-time digital twin dataset and the asynchronous collaborative scheduling instruction set, generating an intelligent control report for the baking production line. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of an application scenario provided in an embodiment of this application; Figure 2 A flowchart illustrating a smart baking production line control method based on multi-process collaboration, provided as an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an intelligent baking production line control system based on multi-process collaboration, provided as an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0019] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0020] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0021] Existing baking production line control methods lack a mechanism for real-time evolution modeling and twin synchronization of the physical state of each independent component when dealing with structured products with multiple components, multiple processes, and complex process paths. This not only makes it difficult to accurately plan and coordinate the optimal process convergence window between components, but also makes it easy for production rhythm to be disrupted, products to be scrapped, or quality to decline in a dynamic production environment due to resource competition or state mismatch. This restricts the production line's potential to cope with high mixed orders and refined management.

[0022] Based on this, this application provides a control method and system for an intelligent baking production line based on multi-process collaboration. First, multi-source data from the production line is collected in real time through a sensor network to construct a digital twin model reflecting the physical state evolution of multiple components such as pastry and fillings. Based on order requirements and the twin data, the time required for each component to meet standards is forward-engineered, and the optimal synchronous assembly "golden window" is calculated in reverse, generating a precise reverse process instruction set with this window as the endpoint. Faced with competition for resources from multiple orders, urgency weights are introduced for conflict pre-arbitration, forming a differentiated scheduling scheme. Finally, a joint judgment is made on whether the real-time status of all components simultaneously and stably meets the standards, triggering synchronous assembly at the optimal time, generating a quantitative performance report to drive iterative optimization, and generating and outputting the report to baking staff. This solution achieves precise synchronization and flexible scheduling of asynchronous processes for multi-component baking products, significantly improving assembly efficiency and resource utilization. Through state-driven intelligent control, it ensures a high degree of consistency and stability in the final product quality. Simultaneously, the system enhances the production line's agile response capability to order changes and emergency order insertions, comprehensively promoting the advancement of baking production towards intelligence and adaptability.

[0023] Figure 1 This is a schematic diagram illustrating an application scenario provided by this application. In the control process of a baking production line, the method provided in this application achieves precise synchronization and flexible scheduling of asynchronous processes for multi-component baked goods, enhancing the production line's agile response capability to order changes and emergency order insertions.

[0024] Specifically, the method of this application is applied to any server that communicates with the physical equipment on the production line and the enterprise order management system, and obtains real-time digital twin datasets provided by the physical equipment on the production line and baking production order data provided by the enterprise order management system through the server.

[0025] For specific implementation details, please refer to the following examples.

[0026] Figure 2 This is a flowchart illustrating a control method for an intelligent baking production line based on multi-process collaboration, provided as an embodiment of this application. The method of this embodiment can be applied to the server in the above scenario. For example... Figure 2 As shown, the method includes: S201. Obtain multi-source real-time data from the production line. Based on the multi-source real-time data from the production line, construct a multi-component state evolution model for multi-component baked goods and synchronize it with a digital twin to generate a real-time digital twin dataset.

[0027] Multi-source real-time data of the production line can be a collection of data reflecting the operating parameters of the equipment and the physical state of the materials, collected in real time from various sensors and equipment (such as temperature sensors, vision inspection cameras, etc.) in the baking production line. The data originates from the physical equipment of the production line.

[0028] Multi-component baked goods can be baked products that are ultimately assembled from multiple components (such as dough, filling, puff pastry, and decorations) with different physical properties and requiring different processing (such as fermentation, baking, cooling, shaping, and decoration) independently or in combination. Examples include sandwich cakes, fruit tarts, and multi-layered pastries.

[0029] A real-time digital twin dataset can be a collection of data created in virtual space that is synchronized with and dynamically updated in real time with the physical production line. It contains information such as the real-time status of each baking component, the future state evolution trajectory based on model predictions, and the occupancy status of production resources (such as ovens and mixers).

[0030] Specifically, traditional baking production line control relies primarily on experience-based scheduling with fixed timelines or single threads. This fails to accurately perceive and predict the independent, dynamically changing physical states (such as temperature, humidity, and texture) of multiple components like dough and fillings within a complex process chain. This leads to difficulties in precisely matching the readiness times of each component, often resulting in "waiting for materials" or "rushing to complete tasks," severely impacting efficiency and quality. The necessity of this step lies in introducing a "multi-component state evolution model" and "digital twin synchronization." This transforms the complex, asynchronous state changes of the physical world into calculable and predictable models and data in a virtual space. Acquiring multi-source real-time data from the production line provides the model with realistic input and calibration benchmarks. Model construction allows for the quantitative prediction of the evolution of each component from its current state to the target state, while digital twin synchronization ensures continuous consistency between virtual predictions and physical reality. The generated "real-time digital twin dataset" becomes the sole and accurate data foundation for all subsequent optimization decisions, upgrading production line control from experience-based "blind scheduling" to "transparent and computable scheduling" based on real-time perception and prediction of all elements. This is an absolute prerequisite for achieving subsequent precise synchronization and flexible scheduling.

[0031] S202. Obtain baking production order data. Based on the real-time digital twin dataset and baking production order data, calculate the golden window period for multi-component state synchronization, and perform multi-component state reverse synchronization planning with this window period as a constraint to generate a synchronization process planning instruction set.

[0032] Baking production order data can be customer order data containing information such as target product type, quantity, delivery time, and possible special process requirements. The data comes from the enterprise order management system.

[0033] The golden window period can be calculated as the optimal time interval during which all components are in a physical state suitable for final assembly, provided that the final quality requirements of all components are met.

[0034] Synchronous process planning instruction set can be a set of instructions for each component, which is a precise process path, parameter settings and start-up time point derived in reverse with the "golden window period" as the unified endpoint.

[0035] Specifically, after obtaining a real-time state overview (digital twin dataset), the core challenge is how to coordinate multiple asynchronous production lines to reach the "correct state" at the "correct time" and converge. Relying solely on forward sequential planning can easily lead to overall assembly delays or rushed execution due to differences in process duration and uncertainties. The necessity of this step lies in its proposal of a "starting with the end in mind" collaborative planning paradigm. First, the final assembly quality target (target state window) is determined by combining baking production order data. Then, based on the predictive model in the real-time digital twin dataset, the "golden window period" is innovatively calculated—a dynamic, optimized synchronization time constraint, rather than a fixed time point. This calculation process actively seeks the optimal coupling point of multiple threads in the time dimension. Subsequently, using this window period as a rigid constraint, "multi-component state reverse synchronization planning" is performed, that is, deducing the process time that each component must start and the process trajectory it should follow from the common assembly completion time. This ensures that even if the process paths and durations of each component are different, they can be guided to a definite synchronization endpoint. The generated "synchronous process planning instruction set" decomposes the macroscopic synchronization goal into a series of microscopic, executable, and time-coupling instructions, fundamentally solving the "time alignment" problem in the asynchronous advancement of multiple processes, and is a key decision-making link for achieving efficient collaboration.

[0036] S203. Based on the real-time digital twin dataset and the synchronous process planning instruction set, perform urgency-weighted scheduling and resource conflict pre-arbitration to generate an asynchronous collaborative scheduling instruction set.

[0037] Urgency-weighted scheduling is a scheduling strategy that considers not only process logic and time constraints in scheduling decisions, but also introduces "order urgency" (such as the urgency of delivery date and customer priority) as a weighting factor to differentiate the production tasks of different orders.

[0038] Resource conflict pre-arbitration can be a process of analyzing the potential competition for the same key equipment (such as a specific oven or decorating table) among multiple process paths in the "synchronous process planning instruction set" before production execution, and making a ruling in advance based on rules (such as urgency weight) to allocate the right to use resources.

[0039] An asynchronous collaborative scheduling instruction set can be a final production instruction set that incorporates resource conflict resolution results and urgency weight adjustments based on synchronous planning, and can be issued to specific equipment for execution.

[0040] Specifically, even with perfect synchronized time planning, conflicts inevitably arise in actual production lines when multiple orders and process instructions for multiple components compete for limited critical resources such as ovens and workstations. Traditional first-come, first-served or fixed-priority scheduling cannot adapt to dynamically changing production demands and urgent orders, easily leading to delays in high-value urgent orders or sacrificing quality control to rush production. The necessity of this step lies in its addition of a "resource conflict pre-arbitration" layer between planning and execution, and the innovative introduction of a "urgency-weighted" mechanism. Based on real-time digital twin datasets, future resource occupancy can be predicted, and conflicts can be detected in advance. Based on the synchronized process planning instruction set, the process requirements and time constraints of each conflicting party are clearly defined. Introducing order urgency as an arbitration weight allows the scheduling system to intelligently balance "ensuring synchronization" and "meeting delivery deadlines." For urgent orders, a "resource preemption" scheme can be initiated to accelerate the process; for regular orders, a "robust avoidance" scheme is adopted to prioritize ensuring their quality stability. This differentiated and forward-looking conflict resolution mechanism generates an "asynchronous collaborative scheduling instruction set," enabling the production line to operate flexibly according to a globally optimal (rather than locally optimal) strategy even under resource-constrained conditions, greatly improving the responsiveness and order fulfillment capabilities of the production system.

[0041] S204. Based on the real-time digital twin dataset and asynchronous collaborative scheduling instruction set, perform joint judgment and synchronous triggering of the status readiness of multiple components to generate an intelligent control report for the baking production line.

[0042] The intelligent control report for the baking production line can be a summary report automatically generated by the system after a production cycle or order is completed. It includes synchronous performance indicators, resource conflict handling results, and actual and planned deviation analysis, which are used for review and continuous optimization.

[0043] Specifically, due to fluctuations in the production environment (such as batch differences in raw materials and minor performance drift of equipment), strictly triggering assembly according to a specific time point may pose risks: a component may not reach its optimal state due to a slight delay, and forced assembly would affect the quality of the finished product. The necessity of this step lies in its upgrade of the final control decision from time-driven to a "state-time joint-driven" approach. Based on a real-time digital twin dataset, the system can obtain the most accurate real-time state of each component, rather than simply relying on predictive models. Based on an asynchronous collaborative scheduling instruction set, the system knows the planned synchronous trigger time. Jointly judging the readiness of multiple components means that, near the planned time point, the system, like an experienced chef, personally "inspects" the "quality" of each component. Only when all components simultaneously report "ready and stable" will the final synchronous trigger instruction be issued. This final decision based on the actual state constitutes the last intelligent checkpoint in quality control. The subsequently generated "Intelligent Control Report for Baking Production Line" not only records the results but also, through quantitative analysis (such as synchronization deviation), distills the experience gained during the production process, forming a closed-loop optimization. This ensures that the method is not only automated, but also an intelligent system with "perception-judgment-learning" capabilities.

[0044] The method provided in this embodiment first collects multi-source data from the production line in real time through a sensor network, constructing a digital twin model reflecting the physical state evolution of multiple components such as pastry and fillings. Based on order requirements and the twin data, the system forward-deduces the time required for each component to meet standards, reverse-calculates the optimal "golden window" for synchronous assembly, and generates a precise reverse process instruction set with this window as the endpoint. Facing competition for resources from multiple orders, an urgency weight is introduced for conflict pre-arbitration, forming a differentiated scheduling scheme. Finally, a joint judgment is made on whether the real-time status of all components simultaneously and stably meets the standards, triggering synchronous assembly at the optimal time, generating a quantitative performance report to drive iterative optimization, and generating and outputting a report to baking staff. This solution achieves precise synchronization and flexible scheduling of asynchronous processes for multi-component baking products, significantly improving assembly efficiency and resource utilization. Through state-driven intelligent control, it ensures a high degree of consistency and stability in the final product quality. Simultaneously, the system enhances the production line's agile response capability to order changes and emergency order insertions, comprehensively promoting the advancement of baking production towards intelligence and adaptability.

[0045] In some embodiments, based on multi-source real-time data from the production line, core physical state data that determines the final availability of each component is collected in parallel for each independent component constituting a multi-component baked goods. By analyzing the evolution of the core physical state of each independent component over time under different process treatments, a state evolution prediction model is constructed for each independent component category. By dynamically comparing and calibrating the multi-source real-time data from the production line with the state evolution prediction model data, the synchronization between the physical production line and the virtual model is achieved, generating a real-time digital twin dataset containing the real-time state of components, state prediction trajectories, and process resource states.

[0046] Core physical state data can be a set of key quantitative physical parameters that determine the final usability of each individual component in a multi-component baked goods.

[0047] A state evolution prediction model can be a mathematical model trained by machine learning algorithms (such as LSTM long short-term memory network) on historical process data for a specific category of independent components (such as "puff pastry" or "cream filling"). It can simulate the dynamic evolution of the core physical state (such as temperature and texture) of the component over time under specific processes (such as heating and cooling).

[0048] Dynamic comparison and calibration can be an adaptive process that continuously compares the component state prediction sequence calculated in real time by the state evolution prediction model with the actual component state sequence collected by the sensor, and automatically and in reverse adjusts the internal parameters of the model (such as the reaction rate coefficient) according to the deviation between the two, so that the model prediction trajectory continuously approaches the actual physical production line situation.

[0049] Specifically, traditional baking production lines lack comprehensive, refined, and predictable monitoring of asynchronous processes involving multiple components. Production line control is like "the blind men and the elephant," relying solely on single-point sensors (such as oven temperature) and fixed timing for coarse scheduling. It fails to quantify the actual evolution of individual components like pastry and fillings within the complex process chain, leading to a severe disconnect between planned assembly times and the actual readiness of each component. This is the root cause of inefficiency and quality fluctuations. To address this, the core of this step is to construct a digital twin that is synchronized in real-time with the physical production line and can proactively predict state evolution. The implementation process is as follows: First, the system uses physical equipment deployed at key nodes of the production line (such as vision cameras, thermocouples, texture probes, and viscometers) to collect core physical state data that determines the final usability of each individual component (such as sponge cake base, mousse filling, and chocolate decoration pieces) that constitutes a baked product (such as a "fruit mousse cake"). For example, the surface temperature distribution of the cake base (e.g., 82°C at the center and 78°C at the edge) is collected non-contactly using an infrared thermal imager array, while the texture state (e.g., elastic modulus of 1200 Pa) is obtained using an integrated texture probe. Subsequently, based on massive amounts of historical production data (such as recording the temperature and texture change curves of cake bases with different initial humidity levels baked at 170°C per minute), an LSTM neural network is used to train a dedicated state evolution prediction model for each type of component (such as a "6-inch chiffon cake base"). This model takes the current state and process parameters (such as set temperature and time) as input and outputs the predicted state value at any future time. During the production process, the system inputs real-time sensor data streams (such as the current cake base temperature of 75°C and texture modulus of 800Pa) into the corresponding model to obtain the predicted trajectory (such as predicting that the temperature will reach 82°C in 5 minutes). At the same time, the system uses a Kalman filter algorithm to dynamically compare the predicted trajectory with the actual data collected at the next moment (such as the actual measured temperature of 81.5°C in 5 minutes), calculates the prediction error, and uses this error to fine-tune the model parameters to complete online calibration. Finally, it generates a real-time digital twin dataset that integrates the real-time status of all components, high-precision predicted trajectory, and the occupancy status of process equipment.

[0050] The method provided in this embodiment constructs a high-fidelity, predictable digital twin, enabling global transparent monitoring and forward-looking simulation of asynchronous processes. This provides a unique and reliable dynamic data foundation for subsequent accurate time window calculation and collaborative scheduling.

[0051] In some embodiments, for the puff pastry or bread dough layer components constituting the main body of the product, their temperature and texture state characterizing crispness or softness are collected; for the filling or sauce layer components, their temperature and viscosity state characterizing flow or shaping ability are collected; for decorative components, their temperature and hardness or shaping state characterizing structural stability are collected.

[0052] Texture can be a physical parameter used to quantify the mechanical properties of puff pastry or bread dough, such as crispness, softness, or elasticity.

[0053] Viscosity state can be a rheological physical parameter used to quantify the flow resistance and shaping ability of fillings or sauce layers.

[0054] Hardness or shape set can be a physical parameter used to quantitatively assess the resistance to deformation and structural stability of decorative components (such as chocolate inserts and icing strips).

[0055] Specifically, traditional baking production relies heavily on single-point temperature measurements and manual judgment to monitor component conditions, which has fundamental flaws. For example, even if the puff pastry reaches the target temperature (e.g., 180°C), it may still be too soft (crispness index only 0.6) due to uneven moisture distribution, failing to support the filling. Similarly, chocolate decorations may meet the temperature requirement (e.g., 29°C) but lack sufficient hardness (Shore hardness only 50HD), causing them to collapse during assembly. This crude understanding that "meeting the temperature requirement equals quality" fails to capture the complex physical essence of multimodal states such as texture, viscosity, and hardness that determine the final product's appearance, and is the core reason for assembly failures and inconsistent tastes in multi-component products. To address this, this step deploys a multimodal sensor fusion network to synchronously collect the core physical state data that determines the final usability of each independent component. The implementation process is as follows: First, the system automatically identifies the component category on the production line (e.g., identifying it as "apple pie puff pastry") through image recognition (e.g., a vision station based on the YOLO algorithm) and RFID tags. Subsequently, for the puff pastry / bread dough, while non-contactly collecting its surface and core temperatures (e.g., surface 185°C, core 82°C) using an infrared thermometer, a texture analyzer probe presses down at a constant rate to measure its force-deformation curve, and the crispness index is calculated in real time (e.g., 0.92, target range 0.85-1.0). For the filling / sauce, in the insulated conveying pipeline, the rotor of an online rotating viscometer shears the material at a fixed speed, measuring its torque and converting it into a dynamic viscosity value (e.g., 4.8 Pa·s, target window 4.5-5.5 Pa·s), while simultaneously measuring its temperature (e.g., 75°C) using a thermocouple. For the decorative components, at the end of the cooling tunnel, a micro-hardness tester performs a micro-indentation test on its surface to obtain a surface hardness value (e.g., 72HD), and an infrared thermometer array scans its temperature distribution (e.g., 28±1°C). All data is uploaded in real time, forming the precise input for the subsequent digital twin model.

[0056] The method provided in this embodiment enables precise, online quantitative perception of the multi-dimensional and decisive core physical states of various components of multi-component baked goods, transforming the previously vague "sensory quality" into precise and controllable physical parameters, providing an irreplaceable data foundation for building a high-fidelity state evolution model and achieving precise synchronous control.

[0057] In some embodiments, by analyzing the historical state data of each independent component under specific heating, cooling, stirring, or settling processes, the dynamic change rules of the core physical state of the independent component evolving from the initial value to the process target value are summarized, and an initial state evolution prediction model is established for each category of independent components. In the current production process, the core physical state data collected in real time is input into the initial state evolution prediction model to obtain the predicted state sequence. By continuously comparing the deviation between the predicted state sequence and the actual state sequence, the response parameters of the initial state evolution prediction model are dynamically adjusted in reverse, so that the predicted trajectory output by the model can adaptively fit the real evolution process under the current raw material characteristics, equipment operating conditions, and environmental conditions, thus completing the online calibration of the model.

[0058] The predicted state sequence can be a sequence of continuous predicted values ​​of the component's state over a future period of time, calculated by the state evolution prediction model, using real-time physical state data of the core components collected during the current production process (such as the current temperature of the puff pastry being 65°C) as input. For example, it can predict the temperature value per minute over the next 10 minutes.

[0059] An actual state sequence can be a continuous data sequence consisting of the actual values ​​of the component's state measured and recorded by sensors within the same future time period.

[0060] Dynamic reverse adjustment can be a model online optimization mechanism that calculates the deviation between the predicted state sequence and the actual state sequence (such as root mean square error) and uses optimization algorithms (such as gradient descent) to automatically and reversely adjust the response parameters (such as thermal conductivity coefficient and water evaporation rate constant) inside the model to reduce subsequent prediction bias.

[0061] Specifically, traditional baking production relies on static empirical models based on standard recipes and fixed equipment parameters to predict process time. These models cannot adapt to batch variations in raw materials (e.g., flour water absorption fluctuations of ±5%), equipment performance degradation (e.g., decreased oven heating wire efficiency), and environmental disturbances (e.g., changes in workshop temperature and humidity). This leads to significant deviations between predicted trajectories and actual state evolution. For example, a standard model might predict that butter filling needs 20 minutes of cooling to the target viscosity, but due to higher ambient temperatures that day, it may not reach the target even after 25 minutes, causing subsequent process blockages. To address this, the core of this step is to construct a dynamic prediction model with "initial knowledge" that can "learn and adapt" in real time. The implementation process is as follows: First, the system retrieves historical data from the Manufacturing Execution System (MES) for specific components (e.g., "chocolate ganache filling") from past production. This data records the temperature and viscosity change curves per minute under specific cooling processes (e.g., natural cooling from 80°C). Based on this dataset, an LSTM neural network algorithm is used for training to summarize the inherent laws of its state evolution, constructing an initial state evolution prediction model. In current production, the model receives real-time data on the current state of the filling (e.g., temperature 70°C, viscosity 2.1 Pa·s) and outputs a predicted state sequence (e.g., predicted temperature 45°C, viscosity 5.1 Pa·s in 10 minutes). The system simultaneously acquires the actual state sequence 10 minutes later (e.g., measured temperature 43°C, viscosity 5.3 Pa·s), and continuously compares the predicted and actual sequences using a Kalman filter algorithm to calculate the deviation. Once the deviation exceeds a threshold (e.g., viscosity prediction deviation greater than 0.2 Pa·s), the system initiates dynamic reverse adjustment, automatically and finely adjusting key response parameters in the model using gradient descent (e.g., adjusting the heat transfer coefficient from 1.0 to 0.96), making the next prediction closer to the true value. This cycle repeats, completing the online calibration of the model and ensuring its output trajectory adaptively matches the actual evolution of the raw materials, equipment, and environment.

[0062] The method provided in this embodiment endows the prediction model with the ability to continuously learn and adapt, enabling it to dynamically compensate for uncertainties in the production process, ensure high fidelity in state prediction, and provide a reliable core basis for subsequent accurate synchronous planning.

[0063] In some embodiments, the target product type is extracted from the baking production order data, and the target state window requirements for each component at the final assembly time of the target product are retrieved from the baking knowledge base. The target state window defines the range of core physical state parameters suitable for assembly of each component. Based on the state evolution prediction model, the shortest and longest process time required for each component to reach its respective target state window from the current state is forward-deduced. With the constraint that all components can reach the target state, the maximum common overlap time interval of all forward-deduced time intervals is calculated, and this interval is solved as the golden window period for this production. With the starting point of the golden window period as the synchronization trigger time point, the state evolution prediction model is used to reverse-deduce the precise state trajectory and process start sequence that each component must follow to achieve synchronization from the synchronization trigger time point, and generate a synchronization process planning instruction set.

[0064] The target state window requirement can be a range of acceptable parameters for the core physical state parameters of each component of a specific product at the moment of final assembly, retrieved from a preset baking knowledge base.

[0065] Forward inference can be a process that starts with the current real-time state of a component and uses a state evolution prediction model to simulate the time point when its core physical state evolves to the first time it enters the range defined by the target state window requirement (earliest completion time) under a given process path, and the time point when the state evolves to the point when it is about to exceed that range (latest allowed time).

[0066] The maximum common overlap time interval can be the intersection of the time intervals (from the earliest completion time to the latest allowed time) of all components obtained through forward deduction, which are the time intervals in which each component reaches and remains within the target state window.

[0067] Reverse engineering can take the starting point of the golden window period (synchronous trigger time point) as the time anchor point, and use the state evolution prediction model to calculate in reverse (time reverse order) the complete state evolution trajectory that each component must go through in order to reach the starting value of its target state window at that precise moment, as well as the precise moment when each process step (such as starting baking or starting cooling) must be initiated.

[0068] The precise state trajectory and process start-up sequence can be a detailed path diagram containing time-state correspondences generated through reverse engineering. It specifies the state reference value that each component should reach at each moment (e.g., the temperature should drop to 52°C) and the start-up and shutdown schedule of all process equipment (e.g., the No. 3 cooling fan should be turned on at T+3 minutes).

[0069] Specifically, traditional multi-component baking production relies on fixed recipes and timelines or manual experience to arrange processes. Essentially, it's a linear, "each doing its own thing" process, unable to dynamically calculate a precise time window that ensures all components reach their optimal assembly state simultaneously based on the real-time status of each component. This often leads to resource waste and quality loss, such as "the pastry is baked and waiting half an hour, but the filling hasn't cooled down enough" or "the pastry softens while waiting for the filling." To address this, the core of this step is to start with the order target, forward optimize the "greatest common divisor" time window of all component states, and backward generate a collaborative path that ensures precise arrival at this window. The implementation process is as follows: The system extracts the target product from the order (e.g., "Black Forest Cake") and retrieves its target state window requirements for each component from the baking knowledge base. For example, the center temperature window for the chocolate cake base is 18-22°C, the humidity window is 30-35%, the liqueur absorption rate window for cherry liqueur-soaked cherries is 40-45%, and the whipped cream stiffness window is 650-750g. Subsequently, based on the calibrated state evolution prediction model, a forward deduction was performed for each component: Given the current state of the cake base (temperature 65°C, humidity 25%), the model predicted the earliest time it would cool to the target temperature and humidity window to be 25 minutes, and the latest allowable time (to prevent over-drying) to be 35 minutes, forming a time interval [25, 35]. Similarly, the time for the butter to reach the target hardness range was deduced to be [28, 40] minutes. The maximum common overlap between these two time intervals was calculated, yielding an intersection of [28, 35] minutes, which is the golden window period for this production. Next, using the starting point of this window (28 minutes) as the synchronous trigger point, a reverse deduction was performed: starting from the lower limit of temperature (18°C) and the lower limit of humidity (30%) that the cake base should be at in the 28th minute, the model calculated its cooling curve in reverse, deducing that it must be transferred from the oven to the cooling zone starting from the 10th minute; similarly, the butter whipping process was deduced in reverse, finding that high-speed whipping must be started at the 24th minute. Ultimately, the system integrates all the precise state trajectories derived from reverse engineering (such as the expected temperature and humidity of the cake base per minute) with the process start-up sequence (such as "T=10 minutes: transfer the cake base to the cooling station; T=24 minutes: start mixer No. 2 to whip the cream") to generate a set of synchronous process planning instructions to be issued to each piece of equipment.

[0070] The method provided in this embodiment enables the dynamic calculation and locking of a globally optimal synchronization time window in a multi-component asynchronous, variable-duration process, and the generation of a reverse collaborative path that ensures accurate matching of this window, fundamentally solving the spatiotemporal matching problem of multi-component production.

[0071] In some embodiments, when there is a direct overlap in the process processing time intervals of each component, the direct overlap is directly calculated as the golden window period; when there is no direct overlap in the process processing time intervals, several process parameter fine-tuning instructions are applied to the state evolution prediction model. The process parameter fine-tuning instructions are used to adjust the process processing intensity or path of several components within the allowable quality deviation range to change the time interval required for them to reach the target state window; the overlap is recalculated based on the fine-tuned process processing time intervals, and the first successfully obtained direct overlap, or the optimal overlap obtained after a finite number of fine-tunings, is calculated as the golden window period for this production.

[0072] Process parameter fine-tuning commands can be specific operation commands generated by the system to make limited adjustments to the process control parameters of a specific component within the allowable quality deviation range.

[0073] Adjusting the intensity or path of the process can be done by fine-tuning the process parameters according to the instructions. There are two types of adjustments: one is to adjust the intensity, that is, to change the parameter values ​​of the process (such as increasing / decreasing the temperature, or increasing / decreasing the stirring speed); the other is to adjust the path, that is, to change the order or combination of the processes (such as changing "natural cooling for 10 minutes first, then forced cooling" to "directly enter the forced cooling stage").

[0074] The time interval required to reach the target state window can be shifted by adjusting the process intensity or path, so that the earliest completion time and the latest allowable time for the component to reach its target state window are moved, thereby shifting or scaling its entire qualified time interval on the time axis.

[0075] Specifically, in complex multi-component production, the time intervals of each component derived from standard process parameters often do not directly overlap (for example, the cake base needs to be assembled within [20, 25] minutes, while the cream filling needs [28, 33] minutes). Traditional scheduling methods often fall into a stalemate in this situation, and can only choose to sacrifice the quality of a certain component (such as overbaking the cake base) or delay the entire order, resulting in wasted resources or decreased customer satisfaction. To address this, the core of this step is to introduce an intelligent conflict resolution mechanism that allows for fine-tuning of the process within acceptable quality tolerances, in order to "create" a synchronization window that did not originally exist. The implementation process is as follows: The system first checks the time intervals derived from the forward derivation of each component. If there is a direct overlap (for example, the overlap between interval A [18, 30] and interval B [22, 35] is [22, 30]), it is directly calculated as the golden window period. If there is no direct overlap (for example, interval C [15, 20] and interval D [25, 30] are completely separated), the system initiates an intelligent adjustment process. It first retrieves the allowable quality deviation range (e.g., acceptable fluctuation of ±2°C in center temperature) for the affected component (e.g., "light cheesecake base") from the baking knowledge base. Then, based on a preset optimization rule base (e.g., the "prioritize shortening time" rule), the system applies a process parameter fine-tuning instruction to the state evolution prediction model controlling the steaming and baking process of the "cake base," such as "increasing the steam injection pressure during the steaming and baking stage from 0.12MPa to 0.15MPa (allowable upper limit)." The model re-performs forward deduction based on this new parameter, obtaining the adjusted new time interval C'[12,18]. The system recalculates the overlap with interval D[25,30] based on the fine-tuned time interval. If there is still no overlap, it continues to apply the next fine-tuning instruction (e.g., "reducing the target soluble solids content at the cooking endpoint from 68% to 65% (allowable lower limit)") to another component (e.g., "reducing the target soluble solids content at the cooking endpoint from 68% to 65% (allowable lower limit)"), thus obtaining interval D'[22,28]. After one or more such finite number of fine-tunings, the calculated intervals C'[12,18] and D'[22,28] still do not overlap, but intervals C'[12,18] and D”[20,26] obtained after another round of fine-tuning produce an overlapping interval [20,18]? The calculation here is incorrect; [20,26] and [12,18] do not overlap. The example needs to be corrected. A more reasonable example is: after fine-tuning, C' becomes [14,19] and D” becomes [17,23], then the overlapping part is [17,19]. The system will use this first successfully obtained direct overlap of [17,19] minutes, or the optimal overlap obtained after multiple attempts (such as the overlap interval with the longest time span), as the final calculation for the golden window period of this production.

[0076] The method provided in this embodiment offers a flexible intelligent conflict resolution mechanism that, while ensuring the core quality of the final product, can actively and finitely adjust process parameters to dynamically shape the synchronization window of each component's status, thereby significantly improving the feasibility of production planning and the success rate of scheduling.

[0077] In some embodiments, when multiple process paths in a synchronous process planning instruction set compete for the same critical resource, order urgency is introduced as an arbitration weight based on baking production order data. For orders with urgency higher than a preset urgency threshold, the process scheme that can shorten the total time to reach the target state is selected first in its competing path, and a higher resource preemption priority is given, generating a resource preemption scheduling scheme. For orders with urgency lower than or equal to the preset urgency threshold, the process scheme with the most stable state evolution and the lowest quality fluctuation risk is selected first in its competing path, and concessions are made in resource allocation, generating a robust avoidance scheduling scheme. The resource preemption scheduling scheme and the robust avoidance scheduling scheme are combined to generate an asynchronous collaborative scheduling instruction set.

[0078] The preset urgency threshold can be a threshold value that the system presets based on historical production data and business rules. It is used to classify orders into two categories: high urgency and low (or equal) urgency, and trigger different scheduling strategies. This threshold can be dynamically adjusted according to the production load. For example, it can be set to 60 points during peak production periods and 40 points during off-peak periods.

[0079] A resource preemptive scheduling scheme can be a set of scheduling instructions that, when generating process paths for orders with urgency levels higher than a preset urgency threshold, prioritizes those that can compress the overall production cycle (even if it may slightly increase energy consumption or equipment wear). When encountering resource competition, it grants priority to use or interrupts the current occupant (if allowed). For example, it allows an urgent "holiday cake" order to interrupt an ongoing "regular bread" order's use of a specific shaped sprayer.

[0080] A robust, yield-avoidance scheduling scheme can be a set of scheduling instructions that, when generating process paths for orders with an urgency level lower than or equal to a preset urgency threshold, prioritizes the schemes that result in the smoothest evolution of component states, the highest quality consistency, and the least impact on equipment. When encountering resource competition, it proactively delays its own process or switches to backup resources (such as using another idle mixing tank) to avoid conflicts.

[0081] Specifically, traditional production line scheduling often employs static, rule-based scheduling (such as FIFO) when faced with multiple orders and intertwined process paths for multiple components. This cannot dynamically respond to urgent orders, VIP customer orders, or the priority demands of high-value products. For example, when both a regular order's cake base and a wedding cake order requiring urgent delivery need to use a single high-precision laser cutting machine, static scheduling may lead to delays in critical orders, while a crude "urgent priority" rule may cause production line chaos and low resource utilization due to frequent adjustments. To address this, the core of this step is to design a two-layer arbitration mechanism based on dynamic urgency weights to achieve intelligent trade-offs and coordination between "efficiency" and "stability," and between "urgent" and "regular." The implementation process is as follows: The system monitors all executing synchronous process planning instruction sets in real time. When it detects multiple process paths competing for the same critical resource (such as two different orders both planning to use the "chocolate tempering machine" between 2:15 PM and 2:30 PM), the conflict arbitration process is triggered. The system first extracts the urgency of relevant baking production orders from their data. This value is calculated by an algorithm based on weighted factors such as the difference between the promised delivery time and the current time, the total order amount, and the customer's historical importance level (for example, a VIP customer wedding cake order that must be delivered within 2 hours has an urgency of 95). The system compares this urgency value with a preset urgency threshold (e.g., 80). For wedding cake orders with an urgency higher than the threshold (95>80), the system prioritizes Option A from its available process options (e.g., Option A: rapid temperature adjustment using a thermostat, taking 5 minutes but with a slight risk of temperature fluctuation; Option B: natural temperature adjustment, taking 15 minutes but with the most stable quality) to shorten the total time to reach the target state. It also assigns a higher priority to the thermostat, forcing adjustments to the originally planned regular order time, thus generating a resource-preemptive scheduling scheme. For regular orders with an urgency level below a threshold (e.g., urgency of 60), the system prioritizes Option B, which has a more stable state evolution, from its available options. It also makes concessions in resource allocation, accepting that the use of the temperature control machine be postponed until after the wedding cake order, thus generating a robust, yield-avoidance scheduling scheme. Finally, the system integrates these two schemes generated for all identified conflict points to form a unified, coordinated, and resource-contention-free asynchronous collaborative scheduling instruction set, which is then distributed to each device controller.

[0082] The method provided in this embodiment establishes an arbitration mechanism based on dynamic urgency assessment, achieving an optimal balance between rapid response to highly urgent orders and stable operation of regular orders. This ensures the timeliness of critical orders while maximizing the stability of the overall production line and resource utilization.

[0083] In some embodiments, the actual production data after executing the asynchronous collaborative scheduling instruction set is integrated to quantitatively evaluate the dynamic deviation between the actual evolution trajectory of each order component's status and the corresponding predicted trajectory in the synchronous process planning instruction set; based on the dynamic deviation, the overall synchronization efficiency index and resource conflict resolution success rate of this production cycle are calculated, and the key process links that cause the largest deviation are identified; the overall synchronization efficiency index, resource conflict resolution success rate, and key process link identification information are jointly structured and encapsulated to generate a baking production line intelligent control report.

[0084] Dynamic deviation can be a quantitative indicator of the degree of difference between the actual evolution trajectory of the components obtained from the production line sensors (such as the actual measurement sequence of temperature and humidity) and the component state prediction trajectory corresponding to the synchronous process planning instruction set, compared point by point (such as every minute) in the time dimension after the execution of the asynchronous collaborative scheduling instruction set.

[0085] The overall synchronization efficiency index is a comprehensive evaluation score calculated using mathematical methods such as weighted averaging and normalization based on the dynamic deviation of all relevant components within the current production cycle. It is used to quantitatively characterize the degree to which each component actually achieves synchronized assembly in the entire order or production batch.

[0086] The resource conflict resolution success rate can be defined as the percentage of resource contention events successfully identified and resolved by the system through the generation of asynchronous collaborative scheduling instruction sets during the current production cycle, relative to the total number of resource contention events occurring during the same period.

[0087] The key process step can be identified by analyzing and comparing the dynamic deviation of each component at different process stages, and identifying the specific process step or equipment station that causes the greatest deviation (i.e. the most significant difference between prediction and reality).

[0088] Structured encapsulation can be the process of organizing and storing evaluation results from different dimensions, such as the overall synchronization efficiency index, the success rate of resource conflict resolution, and the identification information of key process links, according to a predefined, machine-readable report template (such as JSON, XML format, or a specific database table structure).

[0089] Specifically, traditional production processes often only record whether completion is achieved, lacking quantitative review and closed-loop feedback on "accuracy of collaborative synchronization" and "effectiveness of scheduling conflict resolution." This leads to production process optimization relying on fuzzy experience and failing to systematically pinpoint bottlenecks (e.g., time deviations are always caused by the temperature control of a particular oven model). To address this, the core of this step is to construct a multi-dimensional, quantifiable production efficiency assessment and traceability system, transforming process data into decision-making knowledge that guides continuous optimization. The implementation process is as follows: After the production of a "Black Forest Cake" order is completed, the system integrates all actual production data, including the actual temperature and hardness time-series data of the cake base collected by an infrared thermal imager and texture analyzer, and the actual viscosity change curve of the cherry liqueur filling monitored online by a viscometer. The system uses the root mean square error (RMSE) algorithm to quantitatively evaluate the dynamic deviation between the actual temperature curve and the predicted curve of the cake base (e.g., a calculated deviation of 1.8°C), and performs similar calculations for all components. Subsequently, based on these specific deviation values, the system calculates the overall synchronization efficiency index (e.g., a comprehensive score of 88 / 100) for this production using a preset weighted average model (e.g., the weight of major components with a significant impact on final assembly is set to 0.6, and the weight of decorative components is set to 0.4). Simultaneously, the system reviews the logs and counts 5 scheduling arbitration events triggered by competition for "chocolate spraying machine" and "cold storage space" during this production. Of these, 4 conflicts were successfully resolved by adjusting the timing or path, resulting in a resource conflict resolution success rate of 80%. Next, by comparing the deviations of each stage, the system identifies the key process causing the largest deviation (e.g., the deviation in the stiffness of the whipped cream filling is as high as 4.5, far exceeding the average) as "insufficient speed stability of the No. 2 high-speed mixer during the whipping stage." Finally, the system structurally encapsulates the above index, success rate, and key process identifiers (e.g., "Process: P023 - Cream Whipping; Equipment: Stirrer-002") into a JSON format report conforming to industry standards, generating a complete intelligent control report for the baking production line.

[0090] The method provided in this embodiment realizes the transformation from production data to management insights, provides a multi-dimensional indicator system for quantitatively evaluating the accuracy of production synchronization and scheduling efficiency, and accurately locates process bottlenecks, providing a solid data-driven decision-making basis for continuous optimization of production lines, preventive maintenance, and scheduling strategy iteration.

[0091] In some embodiments, in response to a real-time customized adjustment instruction for a baking production order received after the golden window period calculation, the real-time customized adjustment instruction includes a modification of the target state window requirement for at least one component; based on the modified target state window requirement, the state target parameters of the affected component in the state evolution prediction model are locally updated; the forward inference and the calculation of the maximum common overlap time interval are re-executed with the updated state evolution prediction model, and the golden window period and synchronous process planning instruction set for this production are dynamically calculated and refreshed; the refreshed synchronous process planning instruction set and asynchronous collaborative scheduling instruction set are conflict detected and re-coordinated, and updated production control instructions are generated and issued, so that the production line can adapt to customized changes without interrupting the overall process.

[0092] Real-time customized adjustment instructions can refer to the intelligent baking production line control system, which senses and accepts modification instructions initiated by the operator or upstream order system after the calculation during the golden window period (i.e., when the production plan has been generated but not fully executed).

[0093] The target state window requirement can be a precise description of the range of core physical state parameters that each individual component must achieve at the time of final assembly, retrieved from the baking knowledge base and specific to the target product type.

[0094] Partial updates can occur when a system receives a customization instruction that involves only some components. Instead of completely overturning the existing state evolution prediction model, it selectively modifies the state target parameters of specific components in the model that are directly related to the customization requirements (such as "target value of filling viscosity" and "target value of decoration hardness") with minimal impact, while keeping the model parameters and prediction trajectories of all other unmentioned components unchanged.

[0095] Specifically, traditional rigid production lines struggle to respond to customer customization changes during the planning and execution process once the plan is initiated (such as temporarily changing the cake filling jam from strawberry to blueberry). Forced adjustments often lead to production interruptions, material waste, or component mismatches. For example, changing the filling recipe after the puff pastry has entered the baking stage will disrupt the original assembly synchronization window due to changes in its cooling and setting time. If the system cannot dynamically replan, it will either reject the change, harming the customer experience, or accept the change, leading to subsequent assembly failures. To address this, the core of this step is to endow the production line with "online flexible reconfiguration" capabilities. Through a closed-loop "perception-replanning-rescheduling" mechanism, customized disturbances are smoothly absorbed into the ongoing production rhythm. The implementation process is as follows: 30 minutes after production begins (when the "golden window" has been calculated), the operator receives a real-time customized adjustment instruction for "Order A - Deluxe Mille Crepe Cake" via the central control console HMI, specifying that "the third layer of cream filling should be changed from plain to include 5% coffee espresso." The system immediately parses the instruction and, based on the modified target state window requirements (the target viscosity of coffee cream is adjusted from 3500 cP for the original flavor to 3800 cP), locally updates the viscosity target parameter for the "coffee-flavored cream" component category in the state evolution prediction model in memory. Next, the system re-executes the forward extrapolation for all components (including partially processed puff pastry layers, unprocessed cream layers, etc.) starting from the current moment using the updated state evolution prediction model, and recalculates the maximum common overlap time interval for each component's new time interval. Assuming the original window period is [minutes 45, 55], the new calculation finds that because coffee cream requires a longer whipping time to stabilize, its new time interval does not overlap with the original puff pastry layer interval. Therefore, the system dynamically calculates a new golden window period [minutes 48, 58] after process fine-tuning (such as slightly increasing the initial whipping temperature of the cream) and refreshes the corresponding synchronous process planning instruction set. Subsequently, the system automatically performed conflict detection between the updated synchronous process planning instruction set and the currently executing asynchronous collaborative scheduling instruction set. It was found that the time required for whipping coffee cream in the new plan, which would occupy the "No. 2 high-speed mixer," conflicted with the time required for whipping meringue in another order. Therefore, re-coordination was conducted, and after arbitration based on urgency, the start time of the meringue order was slightly delayed. Updated production control instructions were then generated and issued. Throughout the entire process, the baking of the puff pastry and the preparation of other creams were not interrupted, demonstrating the system's adaptation to this customized change.

[0096] The method provided in this embodiment enables efficient and lossless response to real-time customized needs during the production process. Through online replanning and dynamic rescheduling, it ensures that the status of multiple components remains accurately synchronized after changes, greatly improving the flexibility of the production line and customer satisfaction, while avoiding waste of materials and time caused by changes.

[0097] Figure 3A schematic diagram of a smart baking production line control system based on multi-process collaboration is provided in one embodiment of this application, as shown below. Figure 3 As shown, the intelligent baking production line control system 300 based on multi-process collaboration in this embodiment includes: a digital twin module 301, a synchronous instruction module 302, an asynchronous instruction module 303, a control report module 304, and a customized production module 305.

[0098] The digital twin module 301 is used to acquire multi-source real-time data from the production line, and based on the multi-source real-time data from the production line, to construct a multi-component state evolution model for multi-component structured baked goods and synchronize it with the digital twin to generate a real-time digital twin dataset. The synchronization instruction module 302 is used to acquire baking production order data, calculate the golden window period for multi-component state synchronization based on the real-time digital twin dataset and the baking production order data, and perform multi-component state reverse synchronization planning with this window period as a constraint to generate a synchronization process planning instruction set. The asynchronous instruction module 303 is used to perform urgency-weighted scheduling and resource conflict pre-arbitration based on the real-time digital twin dataset and the synchronous process planning instruction set, and generate an asynchronous collaborative scheduling instruction set. The control report module 304 is used to perform joint judgment and synchronous triggering of the status of multiple components based on the real-time digital twin dataset and the asynchronous collaborative scheduling instruction set, and generate an intelligent control report for the baking production line.

[0099] The system in this embodiment can be used to execute the methods of any of the above embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.

Claims

1. A control method for an intelligent baking production line based on multi-process collaboration, characterized in that, include: Acquire multi-source real-time data from the production line; based on the multi-source real-time data from the production line, construct a multi-component state evolution model for multi-component structured baked goods and synchronize it with a digital twin to generate a real-time digital twin dataset. Acquire baking production order data, and based on the real-time digital twin dataset and the baking production order data, calculate the golden window period for multi-component state synchronization, and perform multi-component state reverse synchronization planning with this window period as a constraint to generate a synchronization process planning instruction set. Based on the real-time digital twin dataset and the synchronous process planning instruction set, urgency-weighted scheduling and resource conflict pre-arbitration are performed to generate an asynchronous collaborative scheduling instruction set; Based on the real-time digital twin dataset and the asynchronous collaborative scheduling instruction set, a joint judgment and synchronous triggering of the readiness status of multiple components are performed to generate an intelligent control report for the baking production line.

2. The method according to claim 1, characterized in that, The generation process of the real-time digital twin dataset includes: Based on the multi-source real-time data of the production line, core physical state data that determines the final availability of each component is collected in parallel for each independent component that constitutes the multi-component structure of the baked goods. By analyzing the evolution of the core physical state of each independent component over time under different process treatments, a state evolution prediction model is constructed for each category of independent components. By dynamically comparing and calibrating the multi-source real-time data of the production line with the data of the state evolution prediction model, the synchronization between the physical production line and the virtual model is achieved, generating the real-time digital twin dataset containing the real-time status of components, the state prediction trajectory, and the status of process resources.

3. The method according to claim 2, characterized in that, The core physical state data includes: For the puff pastry or bread dough components that make up the main body of the product, their temperature and texture state, which characterizes the crispness or softness, are collected. For filling or sauce layer components, their temperature and viscosity, which characterize their flow or plasticity, are collected. For decorative components, their temperature and the hardness or shape stability that characterizes structural stability are collected.

4. The method according to claim 2, characterized in that, The process of constructing and calibrating the state evolution prediction model includes: By analyzing the historical state data of each independent component under specific heating, cooling, stirring or settling processes, the dynamic change rules of the core physical state of the independent component from the initial value to the process target value are summarized, and an initial state evolution prediction model is established for each category of independent components. In the current production process, the core physical state data collected in real time is input into the initial state evolution prediction model to obtain the predicted state sequence; By continuously comparing the deviation between the predicted state sequence and the actual state sequence, the response parameters of the initial state evolution prediction model are dynamically adjusted in reverse, so that the predicted trajectory output by the model can adaptively fit the real evolution process under the current raw material characteristics, equipment operating conditions and environmental conditions, thus completing the online calibration of the model.

5. The method according to claim 2, characterized in that, The generation process of the synchronous process planning instruction set includes: Extract the target product type from the baking production order data, and retrieve the target state window requirements of each component at the final assembly time of the target product from the baking knowledge base. The target state window defines the range of core physical state parameters suitable for assembly of each component. Based on the state evolution prediction model, the shortest and longest process times required for each component to reach its respective target state window from the current state are forward- deduced. With the constraint that all components can reach the target state, calculate the maximum common overlap time interval of all forward extrapolation time intervals, and solve this interval as the golden window period for this production; Using the starting point of the golden window period as the synchronization trigger time point, the state evolution prediction model is used to reverse-engineer the precise state trajectory and process start-up sequence that each component must follow to achieve synchronization from the synchronization trigger time point, and generate the synchronization process planning instruction set.

6. The method according to claim 5, characterized in that, The calculation of the maximum common overlap time interval for all forward extrapolation time intervals includes: When there is a direct overlap in the process time intervals of each component, the direct overlap is directly calculated as the golden window period. When there is no direct overlap in the process processing time interval, several process parameter fine-tuning instructions are applied to the state evolution prediction model. The process parameter fine-tuning instructions are used to adjust the process processing intensity or path of several components within the allowable quality deviation range, so as to change the time interval required for them to reach the target state window. The overlapping portion is recalculated based on the fine-tuned process time interval, and the first successfully obtained direct overlapping portion, or the optimal overlapping portion obtained after a finite number of fine-tunings, is calculated as the golden window period for this production.

7. The method according to claim 6, characterized in that, The generation process of the asynchronous cooperative scheduling instruction set includes: When multiple process paths in the synchronous process planning instruction set compete for the same key resource, the order urgency is introduced as an arbitration weight based on the baking production order data. For orders with an urgency level higher than the preset urgency threshold, the process solution that can shorten the total time to reach the target state is selected first in the competitive path, and a higher resource preemption priority is given to generate a resource preemption scheduling scheme. For orders with an urgency level lower than or equal to the preset urgency threshold, the process scheme with the most stable state evolution and the lowest quality fluctuation risk is selected first in its competitive path, and concessions are made in resource allocation to generate a robust avoidance scheduling scheme. By combining the resource preemptive scheduling scheme and the robust avoidance scheduling scheme, the asynchronous collaborative scheduling instruction set is generated.

8. The method according to claim 7, characterized in that, The process of generating the intelligent control report for the baking production line includes: Integrate the actual production data after executing the asynchronous collaborative scheduling instruction set, and quantitatively evaluate the dynamic deviation between the actual evolution trajectory of each order component status and the corresponding predicted trajectory in the synchronous process planning instruction set; Based on the dynamic deviation, the overall synchronization efficiency index and resource conflict resolution success rate of this production cycle are calculated, and the key process links that cause the largest deviation are identified. The overall synchronization efficiency index, the success rate of resource conflict resolution, and the identification information of key process links are jointly structured and encapsulated to generate the intelligent control report of the baking production line.

9. The method according to claim 8, characterized in that, The method further includes: In response to a real-time customized adjustment instruction received after the golden window period is calculated, the real-time customized adjustment instruction includes a modification to the target state window requirement for at least one component; Based on the modified target state window requirements, the state target parameters of the affected components in the state evolution prediction model are locally updated; The forward deduction and the calculation of the maximum common overlap time interval are re-executed using the updated state evolution prediction model, and the golden window period and the synchronous process planning instruction set for this production are dynamically solved and refreshed. The updated synchronous process planning instruction set and the asynchronous collaborative scheduling instruction set are subjected to conflict detection and re-coordination to generate and issue updated production control instructions, enabling the production line to adapt to customized changes without interrupting the overall process.

10. A smart baking production line control system based on multi-process collaboration, characterized in that, The method applied to any one of claims 1-9 includes: The digital twin module is used to acquire multi-source real-time data from the production line. Based on the multi-source real-time data from the production line, a multi-component state evolution model of multi-component baked goods is constructed and synchronized with the digital twin to generate a real-time digital twin dataset. The synchronization instruction module is used to acquire baking production order data, calculate the golden window period for multi-component state synchronization based on the real-time digital twin dataset and the baking production order data, and perform multi-component state reverse synchronization planning with this window period as a constraint to generate a synchronization process planning instruction set. The asynchronous instruction module is used to perform urgency-weighted scheduling and resource conflict pre-arbitration based on the real-time digital twin dataset and the synchronous process planning instruction set, and generate an asynchronous collaborative scheduling instruction set. The control report module is used to perform joint judgment and synchronous triggering of the status readiness of multiple components based on the real-time digital twin dataset and the asynchronous collaborative scheduling instruction set, and generate an intelligent control report for the baking production line.

Images