Industrial chain cooperation-based intelligent packaging engineering cloud service analysis system and method

Abstract

This application relates to the fields of industrial internet and intelligent packaging manufacturing technology, and discloses an intelligent packaging engineering cloud service analysis system and method based on supply chain collaboration. The system includes: a data perception and fusion module that extracts environmental and material parameters, calculates compensation coefficients, and generates an engineering parameter set; a cloud-based map management module that constructs a proxy model and distance matrix based on grid coordinates and response states; a reverse modeling and verification module that generates a two-dimensional tool set based on the compensation coefficients and extracts the minimum bounding rectangle; a collaborative optimization decision-making module that calculates fitness by integrating the aforementioned features and outputs the optimal parameter set and factory; and a fault-tolerant scheduling execution module that converts real-time verification results into instructions for issuing or triggering closed-loop state rollback. This invention employs a technical solution that integrates environmental and material parameters to calculate dynamic folding compensation coefficients, corrects the dimensionality reduction mapping matrix based on these coefficients, and performs polygonal surface region self-intersection detection. This achieves the technical effect of improving the dimensional accuracy of packaging unfolding drawings and intercepting interference from drawings in advance.

Description

Technical Field

[0001] This invention relates to the fields of industrial internet and intelligent packaging manufacturing technology, specifically to an intelligent packaging engineering cloud service analysis system and method based on supply chain collaboration. Background Technology

[0002] The production model for packaging engineering has fully shifted to cross-regional supply chain collaboration. The product flow, from cloud-based concepts to physical edge factories, has been significantly extended. Three-dimensional digital models need to be precisely converted into executable two-dimensional cutting drawings. Massive manufacturing tasks need to be distributed across geographical boundaries to distributed contract manufacturers. Modern manufacturing systems must establish efficient information exchange channels between cloud platforms and physical production lines to maintain the normal operation of the entire supply chain network.

[0003] Existing intelligent packaging cloud platforms typically employ fixed parameters based on historical data to perform 2D unfolding calculations of 3D entities. This static conversion method consumes minimal system computing power, ensuring a high-frequency generation rate of packaging drawings. Regarding production node scheduling, conventional systems often assign tasks based on the total remaining capacity reported by the contract manufacturer or absolute geographical distance. This mechanism can quickly fill capacity gaps in large factories, significantly reducing the computational complexity of task allocation algorithms. The instruction issuance stage typically adopts a unidirectional network distribution architecture, where the system directly pushes the production queue to the control terminal of the underlying equipment, shortening the initial response time of order processing.

[0004] The physical properties of paper-based packaging materials are highly susceptible to fluctuations in ambient temperature, humidity, and batch thickness. Fixed static unfolding parameters cannot adapt to the actual production environment. After drawings are put into production, spatial interference often occurs in the folding area, resulting in high trial-and-error costs due to repeated machine shutdowns and adjustments at the lower levels. Macro-level capacity-oriented scheduling mechanisms ignore the lower-level boundaries of machine processing limits. Single-distance assignment completely excludes logistics load and carbon emission accounting. Allocation nodes often find that the equipment size cannot accommodate the target cutting line after receiving the task, leading to a break in the supply chain collaboration. The physical operational status is constantly changing. Single-direction dispatching instructions lacking two-way verification cannot detect sudden failures of edge devices. When encountering partial production line shutdowns or overlapping production schedules, the rigid distribution mechanism can cause deadlocks in the cloud system, severely disrupting on-time delivery of business orders. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent packaging engineering cloud service analysis system and method based on supply chain collaboration. This system aims to solve the problems in existing technologies where cross-regional packaging outsourcing often results in drawing interference with underlying physical machines due to static fixed parameters, and where a single macro-distance assignment and unidirectional dispatch mechanism easily leads to edge production line matching failures and cloud-based collaborative scheduling deadlocks.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A cloud-based intelligent packaging engineering analysis system based on supply chain collaboration includes: The data perception and fusion module is used to acquire environmental and material parameters of the collaborative foundry nodes to calculate the dynamic folding compensation coefficient and generate a unified set of engineering parameters. The cloud-based graph management module is used to extract the geographic grid coordinates and Boolean response status of collaborative contract manufacturing nodes, construct an in-memory proxy model, and generate a gridded logistics distance benchmark matrix. The reverse modeling verification module is used to correct the preset dimension reduction mapping matrix using the dynamic folding compensation coefficient to generate a two-dimensional unfolded tool line coordinate set, perform self-intersection detection on the two-dimensional unfolded tool line coordinate set and extract the minimum bounding rectangle feature vector. The collaborative optimization decision-making module is used to calculate the comprehensive fitness in the memory proxy model based on the unified engineering parameter set, the gridded logistics distance benchmark matrix and the minimum bounding rectangle feature vector, update the target weight coefficient and output the optimal parameter set and target factory node. The fault-tolerant scheduling execution module is used to send a network handshake request to the target factory node and, based on the real-time verification results fed back by the target factory node, convert the optimal parameter set into an instruction to be issued or to trigger a closed-loop state rollback procedure.

[0007] Preferably, the data perception and fusion module is specifically used for: The workshop ambient humidity, equipment reference pressure, and initial material thickness are obtained via industrial bus. The independent characteristic components and coupled cross-term coefficients corresponding to the workshop environmental humidity, the equipment reference pressure, and the initial thickness of the material are extracted using a pre-set multivariable polynomial fitting model. The dimensionless scalar values ​​are then calculated as the dynamic folding compensation coefficients. The three-dimensional design baseline parameters and the dynamic folding compensation coefficient are subjected to range standardization to generate a normalized engineering feature vector; The normalized engineering feature vector is spliced ​​using a weighted fusion equation to output the unified engineering parameter set.

[0008] Preferably, the cloud-based map management module is specifically used for: Retrieve the geographic grid coordinates and Boolean response status of the collaborating contract manufacturer nodes; The geographic grid coordinates and the Boolean response state are converted into key-value pairs and loaded into a high-concurrency in-memory database cluster to construct the in-memory proxy model; A live timestamp attribute is set for the key-value pair structure. When the live timestamp attribute times out and no heartbeat message is received, the Boolean response status is forcibly overwritten to false. Based on the geographic grid coordinates of the collaborative contract manufacturer nodes, the Manhattan distance algorithm is used to calculate the relative distance between each pair of collaborative contract manufacturer nodes, generate the gridded logistics distance reference matrix, and fill the main diagonal elements of the gridded logistics distance reference matrix with a preset minimum constant.

[0009] Preferably, the reverse modeling verification module is specifically used for: Extract the angle variable between the direction vector of the two-dimensional blade line segment contained in the dimensionality reduction mapping matrix and the direction of the original texture of the material; The dynamic folding compensation coefficient is used as an offset operator, and the corrected two-dimensional line segment mapping weight is calculated by combining the included angle variable and the material anisotropy constant. The actual cutting length is calculated by multiplying the corrected two-dimensional line segment mapping weight as a scale scaling factor by the reference length of the two-dimensional knife line segment. Using the topological connection node of the two-dimensional knife line segment as the reference origin, perform a geometric affine transformation along the direction vector to derive the absolute coordinates of the extended endpoints. For the adjacent line segment breakpoints that have undergone relative offset, perform line equation solving and vector intersection processing to regenerate the coordinates of the closed corner points and output the coordinate set of the two-dimensional unfolded knife line.

[0010] Preferably, the collaborative optimization decision-making module is specifically used for: Extract the maximum length and width dimensions of the two-dimensional boundary contained in the feature vector of the minimum bounding rectangle, and select the cooperative foundry nodes with a Boolean response state of true and a device limit processing area size greater than or equal to the maximum length and width dimensions of the two-dimensional boundary in the memory proxy model as feasible foundry nodes. The baseline physical distance corresponding to the feasible contract manufacturing node is obtained by calling the gridded logistics distance benchmark matrix. The absolute values ​​of carbon emission estimates and logistics time are calculated by combining the single-vehicle load parameters and the benchmark physical distance. These are then mapped and transformed into carbon emission estimates and logistics suitability indicators through a range normalization function. Material cost and structural strength indicators are extracted from the unified engineering parameter set, and a joint objective function is constructed by combining the carbon emission estimate and the logistics adaptability indicator to calculate the comprehensive adaptability.

[0011] Preferably, the fault-tolerant scheduling execution module is specifically used for: Extract the available production queue idle time window of the target factory node and perform time sequence overlap verification with the expected delivery timestamp; extract the actual available area of ​​the die-cutting machine in the idle state and perform spatial inclusion comparison with the feature vector of the minimum bounding rectangle; and combine the equipment health status words to determine the real-time verification result. When the real-time verification result is true, the optimal parameter set is converted into a manufacturing execution command and issued for production. When the real-time verification result is false, the closed-loop state rollback procedure is initiated to overwrite the Boolean response state of the target factory node in the memory proxy model as false and reset the live timestamp attribute. Extract the second-best solution node from the optimal parameter set, ranked first in descending order of comprehensive fitness, and resend the network handshake request.

[0012] Preferably, when the reverse modeling verification module performs self-intersection detection on the two-dimensional unfolded toolline coordinate set and extracts the minimum bounding rectangle feature vector, it is specifically used for: The two-dimensional unfolded toolline coordinate set is closed and divided into discrete polygonal regions, and a polygonal Boolean intersection operation is performed. When it is detected that any two non-adjacent closed surface regions have an intersecting set and the calculated area of ​​the intersecting set is greater than the preset interference tolerance area threshold, it is determined that the surface region self-intersection phenomenon has occurred and the candidate solution verification failure is marked. When no self-intersection of the surface region occurs, the two-dimensional spatial length and width components of the bounding box of the two-dimensional unfolded knife line coordinate set are recorded to extract the feature vector of the minimum bounding rectangle.

[0013] Preferably, when updating the target weight coefficients, the collaborative optimization decision-making module is specifically used for: During the optimization iteration process, the difference between the average carbon emission estimate of all individuals in the current population and the average carbon emission estimate of the previous generation population is calculated. When the average carbon emission estimate exceeds the tolerance limit, a cross-dimensional linkage mechanism is triggered, which amplifies the environmental weight coefficient by a preset step size within the single weight safety limit threshold range, and compresses the remaining weight coefficients proportionally and equally. When the feasible foundry node is empty after completing the full graph traversal in the memory proxy model, the carbon emission estimate and the logistics fit index are set as the worst penalty benchmark, and a penalty factor is introduced to perform a downgrade process on the comprehensive fitness to construct a penalty fitness function.

[0014] Preferably, when the fault-tolerant scheduling execution module extracts the second-best solution node ranked first in descending order of comprehensive fitness from the optimal parameter set and resends the network handshake request, it is specifically used for: Record the cumulative number of supplementary retries and introduce a maximum fault tolerance depth coefficient; When the cumulative number of supplementary retries is less than or equal to the maximum fault tolerance depth coefficient, a new network handshake request is generated for the suboptimal solution node to perform a fault tolerance loop; When the cumulative number of replacement retries exceeds the maximum fault tolerance depth coefficient, a circuit breaker mechanism is triggered to terminate the replacement operation and a manual intervention alert is sent.

[0015] This invention also provides a cloud service analysis method for intelligent packaging engineering based on supply chain collaboration, including the following steps: Obtain environmental and material parameters of the collaborative foundry nodes to calculate the dynamic folding compensation coefficient and generate a unified set of engineering parameters; Extract the geographic grid coordinates and Boolean response states of the collaborative contract manufacturer nodes, construct an in-memory proxy model, and generate a gridded logistics distance benchmark matrix; The preset dimension reduction mapping matrix is ​​corrected using the dynamic folding compensation coefficient to generate a two-dimensional unfolded tool line coordinate set. Self-intersection detection is performed on the two-dimensional unfolded tool line coordinate set and the minimum bounding rectangle feature vector is extracted. Based on the unified engineering parameter set, the gridded logistics distance benchmark matrix, and the minimum bounding rectangle feature vector, the comprehensive fitness is calculated in the memory proxy model, the target weight coefficient is updated, and the optimal parameter set and target factory node are output. A network handshake request is sent to the target factory node, and the optimal parameter set is converted into an instruction to be issued or a closed-loop state rollback procedure is triggered based on the real-time verification result fed back by the target factory node.

[0016] This invention provides a cloud service analysis system and method for intelligent packaging engineering based on supply chain collaboration. It has the following beneficial effects: 1. This invention employs a technical solution that integrates environmental and material parameters to calculate a dynamic folding compensation coefficient. Based on this coefficient, the dimensionality reduction mapping matrix is ​​corrected, and polygonal surface region self-intersection detection is performed. This achieves the technical effect of improving the dimensional accuracy of packaging unfolding drawings and intercepting interference drawings in advance. Compared with the existing technology that relies solely on fixed empirical values ​​for two-dimensional static unfolding, this invention solves the problem of local physical interference easily occurring when facing different workshop temperatures and humidity levels or material batches, thus leading to high machine trial-and-error costs.

[0017] 2. This invention employs a memory proxy model that incorporates geographic grid coordinates and response states, combined with a logistics distance matrix and the feature of the circumscribed rectangle of the cutting line to calculate the comprehensive fitness of multiple objectives. This achieves precise matching of factory nodes and low-carbon production under multi-dimensional constraints. Compared to existing technologies that rely solely on a single capacity indicator or manually assign orders based on proximity, this invention addresses the shortcomings of calculating carbon emissions without considering the equipment's maximum processing capacity and actual carbon emissions, which leads to low efficiency in cross-regional supply chain collaboration and uncontrolled logistics time.

[0018] 3. This invention employs a technical solution that triggers closed-loop state rollback based on real-time verification results from the target factory node, and extracts the suboptimal solution node based on comprehensive fitness to re-execute the supplementary network handshake. This achieves the technical effect of ensuring the successful flow of scheduling instructions and enhancing the system's fault tolerance and response capabilities. Compared to existing technologies that issue dispatch instructions unidirectionally and lack dynamic automatic error correction mechanisms, this invention addresses the shortcomings of these technologies, which are prone to deadlocks in the scheduling system and delays in order delivery when encountering sudden equipment malfunctions at the manufacturing node or scheduling queue timing conflicts. Attached Figure Description

[0019] Figure 1 This is a system architecture diagram of an embodiment of the present invention; Figure 2 This is a flowchart of a method according to an embodiment of the present invention; Figure 3 This is a comparison diagram of dynamic compensation for two-dimensional unfolded blade lines in an embodiment of the present invention; Figure 4 This is a fitness convergence curve of the multi-objective optimization algorithm in an embodiment of the present invention.

[0020] Among them, 10 is the data perception and fusion module; 20 is the cloud-based graph management module; 30 is the reverse modeling and verification module; 40 is the collaborative optimization decision-making module; and 50 is the fault-tolerant scheduling and execution module. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Reference Figure 1 The present invention provides an intelligent packaging engineering cloud service analysis system based on industrial chain collaboration, which may include: a data perception and fusion module 10, a cloud-based graph management module 20, a reverse modeling and verification module 30, a collaborative optimization decision-making module 40, and a fault-tolerant scheduling and execution module 50.

[0023] The data sensing and fusion module 10 is deployed on a cloud server and in the edge gateways of multiple collaborative foundry nodes. The data sensing and fusion module 10 is used to acquire local environmental humidity, equipment reference pressure, and initial material thickness at the collaborative foundry nodes via an industrial bus. The data sensing and fusion module 10 incorporates a polynomial fitting model to calculate dynamic folding compensation coefficients based on environmental and material parameters. The data sensing and fusion module 10 receives 3D design reference parameters and transforms the collected feature data into a unified set of engineering parameters.

[0024] The cloud-based geographic mapping management module 20 is communicatively connected to the data perception and fusion module 10. The cloud-based geographic mapping management module 20 extracts the geographic grid coordinates of each node in the collaborative foundry node set from the database. It also obtains the Boolean response status of each node with respect to the equipment limit size constraint matrix. Finally, the cloud-based geographic mapping management module 20 loads the geographic grid coordinates and Boolean response status into a high-concurrency in-memory database to construct an in-memory proxy model and generate a gridded logistics distance benchmark matrix.

[0025] The reverse modeling verification module 30 is communicatively connected to the data perception fusion module 10 and the cloud-based map management module 20. The reverse modeling verification module 30 is used to establish a directed bipartite graph mapping matrix between the candidate 3D entity edge sequence set and the 2D toolline segment sequence set. The reverse modeling verification module 30 corrects the node weights of the directed bipartite graph mapping matrix according to the dynamic folding compensation coefficient to generate a 2D unfolded toolline coordinate set. The reverse modeling verification module 30 performs polygon Boolean intersection operations on the 2D unfolded toolline coordinate set to detect surface self-intersections and extracts the minimum bounding rectangle feature vector of the 2D unfolded toolline coordinate set.

[0026] The collaborative optimization decision-making module 40 is communicatively connected to the reverse modeling verification module 30 and the cloud-based graph management module 20. The collaborative optimization decision-making module 40 performs multi-objective iterative calculations. It calculates the fitness of candidate solutions based on the joint objective function. Based on the interference judgment results and limit violation verification results output by the reverse modeling verification module 30, the collaborative optimization decision-making module 40 traverses candidate production nodes in the memory proxy model and dynamically updates the target weight coefficients to control the convergence direction of the optimization algorithm.

[0027] The fault-tolerant scheduling execution module 50 is communicatively connected to the collaborative optimization decision-making module 40. The fault-tolerant scheduling execution module 50 obtains the optimal parameter set output by the collaborative optimization decision-making module 40 and the associated target factory node. The fault-tolerant scheduling execution module 50 sends a network handshake request to the target factory node. Based on the real-time verification results returned by the target factory node, the fault-tolerant scheduling execution module 50 executes production scheduling instructions or triggers a state rollback procedure.

[0028] Reference Figure 2 This invention provides a cloud service analysis method for intelligent packaging engineering based on supply chain collaboration, comprising the following steps: S100 collects workshop environment and equipment parameters through the edge gateway to calculate dynamic folding compensation coefficient, and integrates three-dimensional design benchmark parameters with various feature data into a unified engineering parameter set; S200: Extract the geographic grid coordinates of the collaborative contract manufacturer nodes and the Boolean response states of the equipment limit constraints, load them into the in-memory database to build an in-memory proxy model and generate a gridded logistics distance benchmark matrix; S300, based on the dynamic folding compensation coefficient, corrects the mapping matrix of the candidate three-dimensional solution to generate a two-dimensional unfolded tool line coordinate set, performs self-intersection detection on the two-dimensional unfolded tool line coordinate set and extracts the minimum bounding rectangle feature vector for boundary limit verification; S400 performs candidate node reconstruction optimization in the memory proxy model based on the verification results, dynamically updates the target weight coefficient and calculates the fitness, until the multi-objective optimization algorithm converges and outputs the optimal solution set and associated target factory nodes. The S500 initiates a real-time network handshake confirmation to the target factory node, and executes instructions or triggers a fault-tolerant scheduling program that includes state rollback and suboptimal solution replacement based on the verification result.

[0029] In this embodiment, the data-aware fusion module 10 mainly operates in the underlying data link of the system architecture. Specifically, it is deployed in the edge gateway and cloud server of the collaborative foundry node, responsible for the collection, de-identification calculation, and preprocessing of multi-source data at the underlying level. The edge computing-based process parameter de-identification and dynamic compensation coefficient extraction mechanism can effectively prevent the direct leakage of underlying production process data, thereby protecting the data privacy of the collaborative foundry. Specifically, the operation of the data-aware fusion module 10 can be divided into the following sub-steps: S101, the edge gateway acquires environmental and equipment parameters from sensors at the bottom of the workshop via an industrial bus. In one specific implementation, the edge gateway establishes a communication connection with the programmable logic controllers and environmental sensors in the production workshop. This communication connection can be implemented using the Modbus TCP protocol or the OPC UA protocol. Furthermore, the edge gateway reads the workshop ambient humidity from the environmental sensors. Read the device reference pressure from the programmable logic controller. And obtain the initial thickness of the packaging material. Because the fiber structure of paper-based packaging materials is highly sensitive to environmental moisture, and the penetration depth of the die-cutting tool is directly limited by the mechanical force applied by the equipment and the original thickness of the material, this invention selects the above three parameters as the core input quantities for evaluating the material deformation characteristics. To ensure the physical consistency of multi-source data, the edge gateway adds a unified local timestamp when collecting data to strictly align the working condition data of the same die-cutting cycle. Throughout the acquisition phase, the edge gateway performs data reading operations in a local offline state and does not exchange data with the external wide area network in real time. For the specific communication message format and underlying handshake process for reading programmable logic controller register data based on the OPC UA protocol, those skilled in the art can refer to relevant industrial communication protocol standards for implementation; these are well-known technologies in the field and will not be elaborated upon here.

[0030] S102, the edge gateway runs a polynomial fitting model locally, converting the collected environmental and material parameters into dynamic folding compensation coefficients. In actual industrial settings, workshop humidity... relative to equipment reference pressure These are core internal process parameters of the contract manufacturer, and directly uploading them to the cloud poses a high risk of data leakage. Before specific calculations, the fitting process is based on the principle of multivariate nonlinear regression, using historical experimental data to calibrate the mapping weights of each physical quantity to the cardboard folding elongation rate. As a preferred implementation, the edge gateway uses a pre-built multivariate polynomial fitting model to extract the independent and coupled features of the input parameters, and then calculates the anonymized scalar value, i.e., the dynamic folding compensation coefficient. The core computational logic of this polynomial fitting model is configured as follows: ; in, For the fitting constant term, , , These are the coefficients of the first-order term corresponding to a single physical parameter. This refers to the coupling cross-term coefficients of environmental humidity and material thickness. In this embodiment, the value range of these coefficients is determined by each foundry based on local historical die-cutting test data through offline training using the least squares method. Each coefficient itself has dimensions to offset the physical units of the input parameters, thereby ensuring the output dynamic folding compensation coefficients. This is a dimensionless, purely numerical scalar quantity used to accurately characterize the geometric expansion properties of materials under current physical constraints. After completing the above calculations, the edge gateway will only carry the dynamic folding compensation coefficient. Data packets are uploaded to the cloud server through an encrypted channel, without exposing the original underlying measurement values ​​to the outside world, thereby achieving physical isolation and substantial desensitization of industrial data.

[0031] S103, the cloud server receives the 3D design baseline parameters input from the design end and the feature data uploaded by each edge gateway, performs normalization processing, and finally generates a unified engineering parameter set. Specifically, the three-dimensional design baseline parameters include the length, width, and height dimensions of the packaging structure, material designation, and the original texture direction parameters of the material. Here, the original texture direction parameters characterize the fiber orientation of the paper base material during compression molding and serve as the reference coordinate system for subsequent anisotropic compensation calculations. Considering the three-dimensional design baseline parameters and the dynamic folding compensation coefficient... With drastically different dimensions and data distribution ranges, to eliminate the interference of outlier data on global optimization, the cloud server uses a range normalization method to map the heterogeneous data to the same numerical range, forming a normalized engineering feature vector. When performing range standardization calculations, if the absolute value of the difference between the historical maximum and minimum values ​​of a certain feature is less than a preset minimum constant, then... (i.e., the denominator approaches 0), the system determines that the feature has no physical fluctuations in the current period and directly assigns it a normalized baseline value of 0.5, thereby avoiding the underlying program from triggering a division-by-zero exception. Based on the above preprocessing results, the data perception fusion module 10 uses a weighted fusion equation to perform feature concatenation on the normalized data set and outputs a unified engineering parameter set. To fully preserve the independent geometric and physical properties used for subsequent 3D inverse modeling, the system constructs the feature set as a multi-dimensional feature vector, rather than a single scalar. The weighted fusion equation is defined as: ; in, The total number of feature dimensions of the input system. For the first Adaptive weight coefficients for class features For the first Each normalized independent feature component (including 3D size component and dynamic folding compensation component). This represents the transpose of the vector. This fusion process constructs multi-dimensional feature vectors from discrete scalar features, effectively avoiding the one-sided influence of a single feature extremum on the global model representation, while fully preserving the independent dimensions of heterogeneous data. Finally, the data-aware fusion module 10 generates a unified set of engineering parameters. The data is transmitted to the collaborative optimization decision-making module 40 and the reverse modeling verification module 30, and used as the initial state basis for subsequent multi-objective iterative calculations and geometric expansion.

[0032] In one specific embodiment of the present invention, the cloud-based graph management module 20 is mainly used to construct the underlying data structure supporting the high-frequency, latency-free computation of the system. In conventional multi-objective optimization algorithms, frequent data interactions across physical networks can easily cause communication network latency, leading to iteration timeouts. Based on this, the present invention proposes a memory proxy model architecture for high-frequency iterative optimization to achieve offline mapping of manufacturing constraint resources. Specifically, the working process of the cloud-based graph management module 20 can be divided into the following sub-steps: S201, extract the geographic coordinates and equipment status data of the entire network of collaborative contract manufacturer nodes from the underlying persistent database. As a preferred implementation, when extracting multi-source data, to ensure that the underlying physical state upon which cloud optimization is based and the actual operating conditions are within the same effective time window, the cloud-based graph management module 20 retrieves the entire network of collaborative contract manufacturer nodes from the persistent relational database in the cloud based on a preset timed polling mechanism or a status change event triggering mechanism. Each node The system uses geographic grid coordinates instead of raw latitude and longitude in this step. This is to map the nonlinear geographic space into a discrete two-dimensional orthogonal feature space, facilitating the rapid calculation and construction of the subsequent distance matrix. Simultaneously, it addresses the device limit size constraint matrix within each node. This matrix essentially contains physical boundary parameters such as the machine's maximum cutting width, maximum paper feed thickness, and allowable stress. To avoid directly exposing these sensitive underlying process details to the cloud, the cloud-based graph management module 20 does not directly obtain the specific physical machine boundary dimensions. Instead, it obtains the Boolean response status calculated and fed back by the equipment based on the current available capacity and the machine's physical parameters. By reducing the multi-dimensional physical constraints of the underlying devices to Boolean response states containing only true and false states, this data retrieval rule can not only strictly shield the detailed machine data of the underlying foundry to prevent the leakage of trade secrets, but also significantly reduce the computational complexity of the subsequent cloud optimization algorithm in the feasibility verification stage.

[0033] S202, the geographic grid coordinates and Boolean response states are mapped and loaded into a high-concurrency in-memory database to construct an in-memory proxy model. In this embodiment, considering that the multi-objective optimization process usually involves tens of thousands of population generations, the traditional disk-based database access method will produce a serious input / output bottleneck. To overcome this hardware physical limitation, the cloud-based map management module 20 converts the extracted static coordinates and dynamic constraint states into a key-value pair structure and loads it into a high-concurrency in-memory database cluster in the cloud. In this key-value pair data format, the key field is used to uniquely identify a specific collaborative foundry node number, and the value field compactly stores the latitude and longitude grid coordinate tuples corresponding to the node and the current Boolean response state through serialization. Furthermore, to prevent dirty reads of node state data due to network fluctuations, the memory proxy model sets a timestamp attribute for each key-value pair that matches the communication cycle of the real device. Once this timestamp expires and no new heartbeat message is uploaded at the underlying level, the system forcibly changes the corresponding node's boolean response status. The data is overwritten as false to ensure the physical integrity of the fault-tolerant logic. Based on this fully in-memory resident data structure, the upper-level optimization algorithm can complete an offline query for the manufacturing feasibility of any foundry within microseconds, without initiating actual network interaction requests to physical edge nodes. For the specific distributed deployment architecture of the high-concurrency in-memory database and the memory fragmentation reclamation management algorithm for the key-value pair structure, those skilled in the art can refer to the general design specifications of existing high-performance caching systems; these are well-known technologies in the field and will not be elaborated upon here.

[0034] S203, Generate a gridded logistics distance benchmark matrix A high-frequency addressing and reading mechanism was established. Based on the memory proxy model, the cloud-based graph management module 20 calculates the logistics transportation distance between all nodes in the network based on the geographical grid coordinates of each foundry node. Before the specific calculation steps, the system, based on the weighted adjacency matrix principle in graph theory, treats each foundry node as a vertex of the graph and the physical distance as the edge weight. Considering that actual logistics networks often follow the orthogonal distribution characteristics of road networks, the system directly uses the Manhattan distance algorithm for calculation and generates a gridded logistics distance benchmark matrix composed of the relative distances of each node. The specific calculation logic for the elements of this matrix is ​​the sum of the absolute values ​​of the differences between the horizontal and vertical grid coordinates of the corresponding two nodes. This is then multiplied by a preset grid spatial resolution coefficient (e.g., a scale constant representing 10 kilometers per grid), thereby converting the dimensionless grid offset into the actual physical transportation distance. Specifically, since the physical distance between nodes within the same factory is zero, the system defaults to filling the main diagonal elements of the matrix with a preset minimal constant. (For example, a constant with a value in the range of 0.01 to 0.1, used to equivalently characterize the minimum distance baseline for microscopic material flow within the plant area), rather than an absolute zero value. The technical purpose of using this minimal constant to replace the zero value is to prevent the subsequent system from triggering an abnormal termination error where the denominator approaches zero when calculating macroscopic indicators such as logistics carbon emissions or efficiency conversion ratio per unit distance; at the same time, it ensures the integrity of this gridded logistics distance baseline matrix. To ensure strict full-rank nonsingularity, the algorithm avoids dimensional degradation during spatial inverse mapping calculations. During high-frequency algorithm iteration in the collaborative optimization decision module 40, when faced with production node changes or optimization matching, the cloud-based graph management module 20 provides data support through a high-frequency addressing mechanism. This mechanism directly uses the factory node's index number as the matrix's row and column biases, reading distance values ​​using direct memory addressing with constant time complexity. Through this design, the mechanism effectively eliminates redundant computational costs associated with repeated coordinate distance calculations during optimization iterations, thus ensuring the algorithm's global convergence efficiency.

[0035] In a further embodiment of the present invention, the reverse modeling verification module 30 is mainly responsible for the dimensionality reduction interference detection task from 3D design to 2D manufacturing. In the conventional packaging engineering field, the design end often ignores the physical compression effect of material thickness during the folding process, resulting in frequent dimensional deviations or structural overlaps in the generated 3D model during actual 2D die-cutting. To solve the above-mentioned technical defects, the present invention proposes a multi-order physical constraint-driven 2D tool line reverse dynamic compensation and topological interference detection algorithm. Specifically, the complete working process of the reverse modeling verification module 30 can be divided into the following sub-steps: S301, establish a directed bipartite graph mapping matrix between the candidate 3D entity edge sequence set and the 2D tool segment sequence set. The dimensionality reduction transformation from 3D design to 2D manufacturing is essentially a planar deconstruction and reconstruction of spatial topological relationships. Before performing geometric dimensionality reduction calculations, the system, based on the bipartite graph matching principle in graph theory, deconstructs the solid topological structure of the 3D packaging box into discrete edge features. As a specific implementation method, the system defines the candidate 3D entity edge sequence set. and the corresponding set of two-dimensional knife line segment sequences Since a 3D solid edge is highly likely to split into multiple 2D line segments during physical unfolding due to process requirements, the reverse modeling and verification module 30 constructs a directed bipartite graph mapping matrix. This is used to record the topological relationship between the two. The matrix is... A numerical matrix of dimension, whose internal matrix elements Used to characterize three-dimensional edges For two-dimensional line segments The mapping weights and connection directions are determined. Considering the numerical stability of subsequent matrix operations, if the projection of a 3D entity edge onto the 2D unfolded plane degenerates into an isolated point (i.e., the line segment length approaches 0), the system does not directly set its weight to zero, but instead assigns it a preset minimal constant. This ensures that the matrix does not generate singular nodes. The core technical purpose of preserving this minimal constant is to maintain the topological connectivity of the bipartite graph mapping network, thereby effectively preventing abnormal interruptions in the graph traversal program due to graph structure breaks (i.e., the generation of isolated dangling nodes) during subsequent execution of 2D line segment intersection and closed region traversal algorithms. This is achieved by establishing this directed bipartite graph mapping matrix. The system realizes mathematical dimensionality reduction mapping from three-dimensional spatial entities to two-dimensional manufacturing drawings.

[0036] S302 introduces a dynamic folding compensation coefficient as an offset operator to perform anisotropic correction on the node weights of the directed bipartite graph mapping matrix, thereby inversely generating a two-dimensional unfolded toolline coordinate set. In actual manufacturing scenarios, corrugated cardboard and other paper-based packaging materials are made of multiple layers of fiber-oriented pressing, and their folding resistance and elongation differ significantly in the direction parallel to and perpendicular to the fiber texture. Based on this general material mechanical property, the system must perform anisotropic dimensional compensation. In this embodiment, the reverse modeling verification module 30 obtains the dynamic folding compensation coefficient uploaded by the underlying data perception fusion module 10. And it is used as an offset operator on the directed bipartite graph mapping matrix. On the non-zero elements. Furthermore, for any line segment in the set of two-dimensional knife-line segment sequences... The system extracts the angle between its direction vector and the direction of the material's original texture. Based on the above parameters, the system dynamically allocates compensation weights, and the corresponding weight correction function is defined as follows: ; in, The corrected two-dimensional line segment mapping weights; The first element in the original mapping matrix Three-dimensional edge pair The initial topological weights of the two-dimensional line segments are set to a range of [0,1]. It is the anisotropy constant of the material, usually taken in the range of [0.05, 0.3] based on laboratory tensile test data, and is used to quantitatively characterize the degree of deformation difference of the material in different stress directions; This refers to the direction angle variable. The technical purpose of this correction formula is to: when the two-dimensional toolline tends to be parallel to the material texture (i.e., When the absolute value of approaches 1, the system applies maximum process dimension compensation to offset the risk of cardboard breakage during processing and folding; when the two are perpendicular, it reverts to the baseline compensation amount. This is because the range of the cosine function is bounded and controlled by the material anisotropy constant. This calculation ensures the numerical stability and continuity of the compensation weights during the nonlinear scaling process. Subsequently, the reverse modeling and verification module 30 recalculates the endpoint coordinates of each line segment based on the corrected weight matrix. The specific calculation logic is as follows: the system maps the corrected two-dimensional line segments to weights. As a scaling factor, the reference length of the original two-dimensional knife line segment is extracted. Calculate the actual cutting length after compensation. Next, using the topological connection nodes of the line segments as the origin, a geometric affine transformation is performed along the original direction vector to derive the absolute coordinates of the extended endpoints. To prevent the connection between adjacent line segments from breaking due to anisotropic compensation, the system, based on the topological closure constraints of the original 2D toolpath, performs linear equation solving and vector intersection processing on the break points of adjacent line segments that have experienced relative offsets to regenerate the coordinates of the closed corner points. Based on the above coordinate update and topological reconstruction traversal, the final output is a 2D unfolded toolpath coordinate set with process compensation. Through this correction process, the system accurately maps the material thickness and equipment pressure constraints in the physical space onto a two-dimensional geometric drawing.

[0037] S303, performs polygon Boolean intersection operation on the 2D unfolded toolline coordinate set to detect self-intersections of the surface regions and extracts the minimum bounding rectangle feature vector. After generating the compensated 2D unfolded diagram, the system needs to perform rigorous geometric verification of its manufacturing feasibility. This step is mainly based on the polygon topological intersection principle in computational geometry. As a preferred implementation, the reverse modeling verification module 30 performs polygon topological intersection operation on the 2D unfolded toolline coordinate set. The closed loop is divided into multiple discrete polygonal regions. For each of these regions, the system performs a polygon Boolean intersection operation. In practical engineering calculations, due to the truncation error of the computer's underlying floating-point system, directly determining the intersection to be empty may lead to misjudgment of tolerance. To avoid one-sided judgments relying on a single extreme value, the system introduces an area tolerance dimension for comprehensive evaluation. Specifically, if any two non-adjacent closed regions are detected to have a non-empty intersection set, and the calculated area of ​​this intersection set is greater than a preset interference tolerance area threshold, then the system will perform an intersection operation. (For example, taking 0.1 times the square of the original material thickness), the system determines that the current solution has a substantial surface region self-intersection phenomenon. Physically, this phenomenon means that the packaging box has structural interference after die-cutting and unfolding, and the system will directly mark the candidate solution as failing verification. Under the ideal premise that no self-intersection occurs, the system further extracts the coordinate set of the two-dimensional unfolding cutter line. The minimum bounding rectangle eigenvector This feature vector, by directly recording the two-dimensional spatial length and width components of the bounding box, precisely establishes the rectangular boundary size required to accommodate the entire unfolded drawing, which characterizes the minimum raw material area required to process this design. For the specific low-level code implementation of polygon Boolean intersection operations (such as the scan line algorithm in the lower-order features) and minimum bounding rectangle boundary extraction (such as the rotating caliper algorithm in the lower-order features), those skilled in the art can refer to standard geometric algorithm libraries related to computer graphics; these are well-known technologies in the field and will not be elaborated upon here. Finally, the system uses this minimum bounding rectangle feature vector... The output is sent to the collaborative optimization decision module 40, which serves as the core basis for subsequent limit verification and device matching.

[0038] In a specific embodiment of the present invention, the collaborative optimization decision module 40 mainly serves as the core of the cloud computing engine, responsible for constructing the global logical closed loop and scheduling model. Addressing the technical bottleneck of the severe disconnect between structural design and underlying manufacturing resources in traditional optimization algorithms, this invention proposes a cross-dimensional linkage and flexible constraint reconstruction mechanism involving microscopic two-dimensional topological interference and macroscopic target weights. Specifically, the complete execution process of the collaborative optimization decision module 40 can be divided into the following sub-steps: S401, construct the joint objective function for multi-objective iterative computation and calculate the fitness. Before executing specific calculation steps, the system, based on the multi-objective decision-making principle in heuristic optimization algorithms, specifically employs weighted scalarization technology to aggregate multiple heterogeneous physical and economic evaluation indicators of the packaging design into a single scalar fitness value to evaluate the overall merits of the current design scheme. As a preferred implementation, the collaborative optimization decision-making module 40 calculates four core normalized indicators for each joint candidate scheme bound by the three-dimensional design structure and the target foundry node: a material cost indicator characterizing the design itself. Structural strength indicators And carbon emission estimates based on the physical characteristics of associated nodes. Logistics compatibility index The technical reason for selecting the above four indicators as core input parameters is that they comprehensively cover the four key dimensions of modern green manufacturing: economy, mechanical reliability, environmental friendliness, and supply chain response speed. This effectively avoids quality degradation or high carbon emissions caused by solely pursuing low costs. Furthermore, the material cost indicator is established based on the product of the unfolded two-dimensional area and the unit price of consumables; the structural strength indicator is calculated based on the three-dimensional stress characteristics and material thickness; the carbon emission estimate is positively correlated with the logistics transportation distance and the carbon emission rate per unit load; and the logistics adaptability indicator is measured based on the negative correlation function between the actual delivery cycle and the customer's expected time (e.g., selecting the reciprocal of the deviation), ensuring that the smaller the deviation, the larger the numerical value of the adaptability indicator. To eliminate the inherent differences in the dimensions and orders of magnitude of the above indicators, the system pre-processes the range normalization of each indicator, distributing them within the [0,1] interval. Based on the above processing results, the system constructs a fitness calculation model with multi-objective iterative computation, whose joint objective function is defined as: ; in, This represents the dimensionless comprehensive fitness value of the current candidate design scheme; a larger value indicates better overall performance. Due to material costs... Compared with carbon emission estimates In engineering terms, this is a negative indicator where smaller is better; the formula specifically uses... and A forward flip is performed to ensure that the optimization direction of all added options remains completely consistent, thereby avoiding logical contradictions in multi-objective optimization. Meanwhile, , , , These are the macro-level basic weight coefficients for each corresponding indicator. In this embodiment, the initial values ​​of these weight coefficients are determined by the analytic hierarchy process (AHP) combined with historical production experience. Their values ​​are strictly set within the range of (0,1), and the sum of all weight coefficients is required to be constant at 1 to satisfy the normalization constraint. By using this joint objective function, the system not only mathematically integrates discrete objectives such as cost reduction, physical quality improvement, and low-carbon emission reduction into a single dimension, but also establishes a clear evaluation benchmark for subsequent algorithm population evolution.

[0039] S402, the system rapidly traverses candidate production nodes in the memory proxy model using the minimum bounding rectangle feature vector and calculates carbon emissions and logistics changes by calling the distance benchmark matrix. After obtaining the microstructural characteristics of an individual, the system needs to establish its physical matching relationship with macroscopic manufacturing resources. In this embodiment, the collaborative optimization decision module 40 receives the minimum bounding rectangle feature vector extracted by the reverse modeling verification module 30. The system then parses the maximum length and width dimensions of its contained two-dimensional boundaries. Subsequently, the system reads the high-concurrency in-memory database built by the cloud-based graph management module 20 in a high-frequency concurrent manner, directly traversing all collaborative foundry nodes across the network. During the traversal and comparison, the system only includes factories that simultaneously satisfy two joint constraints into the feasible solution set: one is the Boolean response state currently fed back by the target node. The first condition is true; the second condition is that the equipment's maximum processing area at the target node must be strictly greater than or equal to the eigenvector in both the horizontal and vertical directions. The system specifies the length and width dimensions. For the selected feasible contract manufacturer nodes, the system directly calls the gridded logistics distance benchmark matrix by passing in the node index number as the memory offset. This direct memory addressing method rapidly obtains the baseline physical distance from the current node to the target delivery location with constant time complexity. Combining preset single-vehicle load parameters and a basic transportation timeliness model, the system then calculates the absolute value of the carbon emission estimate and the absolute value of the logistics time allocated to that node, and dynamically maps them to the 0-1 range using the aforementioned range normalization function, converting them into normalized carbon emission estimates. Logistics compatibility index The above calculations yielded... and This serves as the core input variable for the joint objective function in S401, used to evaluate the fitness of the current node's matching scheme. It's worth noting that the physical distance to itself was set to a preset minimum constant during the initial construction of the distance baseline matrix. This operation not only ensures the non-singularity of the matrix state, but also completely avoids the floating-point division-to-zero operation crash caused by the denominator approaching 0 when performing index conversion and derivative calculation based on unit distance at this point, thus effectively guaranteeing the continuous stability of high-frequency optimization iteration.

[0040] S403, based on a cross-dimensional linkage mechanism, dynamically updates the target weights and applies a penalty factor to unmatched nodes in generational evolution control. To overcome the limitations of statically fixed weights in traditional optimization algorithms, this invention creatively establishes a flexible constraint reconstruction mechanism between microscopic geometric constraints and macroscopic scheduling objectives. After each generation of population evolution, the collaborative optimization decision module 40 globally analyzes the evolutionary trends of evaluation indicators between generations. The system calculates the average carbon emission estimate and logistics fit index of all individuals in the current population and compares them with the average values ​​of the previous generation to obtain the carbon emission change. With changes in logistics As a preferred control strategy, if the average carbon emissions of the current population exceed the tolerance limit set by the average of the previous generation, the system will automatically trigger cross-dimensional linkage. That is, within the preset single-item weight safety upper limit threshold (e.g., 0.8, to prevent the algorithm from experiencing dimensional degradation in multi-objective solutions under extreme conditions), the environmental weight coefficient will be dynamically amplified by a preset step size (e.g., 0.05). The system compresses the other three weight coefficients proportionally to maintain a system-level constraint where the total weight sum is 1, thus forcing subsequent crossover and mutation actions in the population to tilt towards low-carbon emissions. Furthermore, as a necessary fallback mechanism for generational evolution control rules, if the system completes a full graph traversal in the memory proxy model and finds that no cooperating foundry's equipment size can accommodate the minimum bounding rectangle of the candidate solution (i.e., the candidate node set is empty), the system does not directly force the removal of that individual. To maintain the continuous layering of genetic diversity in the algorithm's optimization population within the multidimensional solution space, the system estimates the carbon emissions of the candidate solution. Logistics compatibility index The default setting is the worst-case penalty benchmark (e.g., set to 1 and 0 respectively), and the base fitness is calculated based on the S401 joint objective function. Then, the system introduces a penalty factor that takes on a very large value. (For example, the default value is 10) 6 ), and by constructing a penalty fitness function The original fitness level is downgraded. This is due to the penalty factor. Since the denominator is a positive real number much greater than 1, this degradation formula does not have the computational risk of the denominator approaching 0. Through this high-intensity computational penalty operation, the overall fitness of abnormal individuals will decay exponentially. Under the action of the survival-of-the-fittest selection operator, the system forcibly guides the optimization model to spontaneously converge to the legal feasible domain that satisfies the underlying machine's physical constraints in subsequent iterations. Ultimately, it perfectly realizes the powerful intervention and logical closed-loop control of the microscopic two-dimensional topology manufacturing conditions on the macroscopic algorithm evolution.

[0041] In a specific embodiment of the present invention, the fault-tolerant scheduling execution module 50 is mainly used to establish an instruction issuance guarantee mechanism in a dynamic industrial environment. In the asynchronous collaborative architecture of cloud optimization and edge execution, due to network communication latency and high-frequency fluctuations in physical production line status, concurrent timing state failures are often difficult to completely avoid. Based on the above technical challenges, the present invention creatively proposes a closed-loop state rollback and suboptimal solution supplementation fault-tolerant mechanism. Specifically, the complete working process of the fault-tolerant scheduling execution module 50 can be divided into the following sub-steps: S501, a real-time network handshake request containing the minimum bounding rectangle feature vector is sent to the target factory node. Before executing the specific handshake steps, the system adds a pre-verification step before formally issuing production instructions, based on the two-phase commit principle in distributed transactions. Specifically, after the collaborative optimization decision module 40 outputs the globally optimal design and scheduling joint scheme, the fault-tolerant scheduling execution module 50 does not immediately issue the execution instruction, but enters the confirmation waiting stage. As a preferred implementation, the system constructs the real-time network handshake request based on the Transmission Control Protocol. The technical reason for selecting the minimum bounding rectangle feature vector and the expected delivery timestamp as the core parameters of the request is that they respectively constitute the rigid boundary conditions of the manufacturing task in physical space and delivery sequence, which can cover the core processing constraints to the greatest extent. Furthermore, the data packet structure of this request not only includes the conventional frame header, checksum, and node routing addressing information, but its payload segment is also specially encapsulated with the minimum bounding rectangle feature vector of the candidate scheme. and expected delivery timestamp In this embodiment, the triggering condition for the aforementioned handshake request is strictly set to the natural convergence of the cloud-based optimization iteration, and the overall fitness of the optimal solution remains stable for multiple generations. By issuing this real-time network handshake request, the system can force a physical state alignment across the cloud and edge devices before officially locking the physical production schedule, thereby effectively avoiding the risk of physical mismatch caused by state lag in the cloud memory proxy model.

[0042] In the S502, the edge gateway returns real-time verification results based on the latest local physical production line queue. To ensure consistency in timestamp verification among multi-source network nodes, the edge gateway pre-synchronizes its clock with the cloud server via a network time protocol to establish strict timing alignment logic. Upon receiving the network handshake request, the target collaborating foundry immediately triggers the underlying verification logic through its locally deployed edge gateway. As a specific implementation, the edge gateway does not rely on any local historical cached data but directly reads the real-time programmable logic controller register status of the underlying machine via the industrial communication bus. Specifically, the edge gateway extracts the current available production queue idle time window. and compare it with the expected delivery timestamp in the request packet. Perform timing overlap verification; simultaneously, extract the actual usable area of ​​the die-cutting machine currently in an idle state. and compare it with the feature vector of the minimum bounding rectangle. The system performs spatial inclusion comparison based on the two-dimensional length and width dimensions. To avoid relying on a single dimension for biased judgment, the system constructs a multi-dimensional joint verification function: ; in, The output is the real-time verification result, and its value is a boolean. For timing verification functions, when Completely fall into The internal time output is true; For spatial verification functions, when Both length and width dimensions are greater than or equal to When the length and width dimensions are true, the output is true; This is the device health status word; true indicates normal operation, and false indicates a fault alarm if the register reports a fault. This is a logical AND operator. The technical purpose of this multi-dimensional joint verification function is to comprehensively determine the spatiotemporal availability and physical health of the equipment. For the above verification, the edge gateway only sends a real-time verification result of a Boolean true state to the cloud when the time window meets the production schedule margin, the physical space meets the requirements, and there are no fault shutdown signals on the equipment; otherwise, it sends a Boolean false state. For the specific data frame parsing of the underlying industrial bus protocol and the addressing and reading / writing operations of the programmable logic controller registers, those skilled in the art can refer to the communication specifications of standard industrial control systems, which are well-known technologies in the field and will not be elaborated upon here.

[0043] S503, trigger closed-loop state rollback and extract suboptimal solution to re-initiate handshake loop. In this embodiment, after the fault-tolerant scheduling execution module 50 receives the real-time verification result from the edge node, the system immediately executes the branch control logic. If the verification result... If true, the system directly converts the joint solution into standard manufacturing execution instructions and issues them for production. If the verification result... A false result indicates a typical concurrent timing state failure, at which point the system immediately initiates a closed-loop state rollback procedure. During the rollback operation, the fault-tolerant scheduling execution module 50 directly sends a system-level interrupt command to the cloud-based graph management module 20, forcing the target node's Boolean response state in the memory proxy model to be changed. The value is overwritten as false, and its timestamp attribute is reset simultaneously. The technical and physical significance of this overwriting operation is to immediately prevent subsequent optimization groups from including the failed node in the feasible solution set again, thus preventing system-level scheduling deadlock. Subsequently, the system extracts the second-best solution currently ranked first from the globally optimal solution set cached by the collaborative optimization decision module 40, arranged in descending order of comprehensive fitness, as the new target solution. Furthermore, to prevent endless handshake retries during widespread network congestion, the system introduces a maximum fault tolerance depth coefficient. This coefficient is typically set to an integer value within the range [3, 5], based on the complexity of the supply chain network topology and the requirements for fault tolerance response time. Based on this, the system records the current cumulative number of supplementary retries. And perform logical judgment: if The fault-tolerant scheduling execution module 50 regenerates and sends out network handshake requests for the newly extracted suboptimal solution nodes, and continues to execute the fault-tolerant loop; if This indicates that the entire network's manufacturing resources have reached their maximum load. The system then triggers a circuit breaker mechanism, terminating the replacement operation and sending a manual intervention alert to the business terminals. Through this suboptimal replacement fault-tolerance mechanism, the system can still smoothly transition scheduling tasks with relatively low computing power costs when facing extreme conditions of sudden fluctuations in the underlying physical network, thereby greatly improving the overall robustness of the industrial internet network.

[0044] To further clarify the collaborative working process of the technical solution described in this invention, the overall operating logic of the system will be explained in detail below with reference to a specific packaging engineering production scenario.

[0045] First, the data perception and fusion module 10 performs low-level feature acquisition and preprocessing. In this embodiment, the business terminal issues a 3D design task for a corrugated paper-based packaging structure. Multiple collaborative contract manufacturer nodes exist within the target supply area. For a candidate contract manufacturer node, its deployed edge gateway reads the current workshop's ambient humidity (65%) and equipment reference pressure (2.5 MPa) via the industrial bus, and obtains the initial material thickness of the selected cardboard as 3.0 mm. The edge gateway, in a local offline state, calls a polynomial fitting model, substitutes the aforementioned physical parameters, and calculates the dimensionless dynamic folding compensation coefficient. The value is 1.042. The edge gateway uploads this coefficient value to the cloud server via an encrypted channel. The cloud server receives the 3D packaging dimension parameters input from the design end and performs range standardization processing on them along with the uploaded dynamic folding compensation coefficient. After eliminating dimensional differences, the system uses a weighted fusion equation to concatenate the data to generate a multi-dimensional unified engineering parameter set. .

[0046] Subsequently, the cloud-based graph management module 20 constructs a memory proxy model. Based on a timed polling mechanism, the system retrieves the geographic grid coordinates of the aforementioned collaborative foundry node set from the underlying persistent database, while simultaneously receiving Boolean response statuses from each node's edge gateway based on the underlying machine's physical constraints. The system uses the factory node index number as the key field and the grid coordinates and Boolean response status as value fields, loading them into a high-concurrency in-memory database. Based on this, the system uses the Manhattan distance algorithm to calculate the relative spatial offset between each node and generates a gridded logistics distance benchmark matrix. To ensure the nonsingularity of the matrix, the system uniformly assigns a very small constant value to the elements on the main diagonal. This establishes the address space for subsequent high-frequency offline queries.

[0047] Next, the reverse modeling verification module 30 performs dimensionality reduction interference detection and feature vector extraction. Based on a directed bipartite graph mapping matrix, the system reduces the 3D topological structure of the cardboard box entity into a sequence of 2D knife line segments. During coordinate transformation, the system incorporates the dynamic folding compensation coefficients passed from the underlying layer. As the core offset operator, and combined with the preset material anisotropy constant Anisotropic correction is applied to each two-dimensional line segment. Combined with... Figure 3 The graphical mapping results shown in the figure indicate that the thin solid line represents the theoretical baseline cutter line profile without compensation, while the thick dashed line represents the actual cutting cutter line profile after physical property correction. As can be seen from the figure, the dimensional scaling offset of the 2D cutter line in the direction parallel to the original fiber texture of the cardboard is significantly greater than the offset perpendicular to the force direction. After coordinate reconstruction, the system performs a polygon Boolean intersection operation on the generated 2D unfolded cutter line coordinate set. Through traversal calculations, the intersection area between each closed surface region is less than the preset interference tolerance area threshold, and the system determines that the surface regions do not self-intersect. The system then extracts the minimum bounding rectangle feature vector of the bounding box of this 2D coordinate set. The length and width of its physical boundary were recorded as 850mm and 620mm, respectively.

[0048] Based on this, the collaborative optimization decision module 40 executes cross-dimensional linkage optimization control. The system will use the aforementioned feature vectors... Input a high-concurrency in-memory database and directly iterate through and compare the maximum processing area of ​​all factory nodes in the network. This is based on the condition that the size tolerance meets the requirements and the current Boolean response state. For genuine, legitimate production nodes, the system invokes the gridded logistics distance benchmark matrix using direct memory addressing. The carbon emission estimate of the current matching scheme is calculated by combining the single-vehicle load weight parameters. Logistics compatibility index During the population's generational evolution, the system uses a joint objective function to calculate the overall fitness of each individual. Combining Figure 4 The iterative trend is shown, with the horizontal axis representing the number of iterations in the optimization algorithm and the vertical axis representing the dimensionless overall fitness value. When the iteration reached approximately the 40th generation, the system detected that the average carbon emissions of the current population exceeded the tolerance limit set by the previous generation's average. The system then automatically triggered a cross-dimensional linkage mechanism, amplifying the environmental weight coefficient by a preset step size within the safety upper limit threshold. The value of is determined, and other weights are proportionally compressed. Guided by this dynamic weight update, the curve experiences a brief fitness reassessment oscillation after the 40th generation. Subsequently, the entire optimization population is forced to converge towards the physically feasible region with low carbon emissions, reaching a convergent and stable state of the global optimum around the 120th generation. The system ultimately locks in the top-ranked joint scheme and extracts the associated target factory node.

[0049] Finally, after optimization and convergence, the fault-tolerant scheduling execution module 50 performs closed-loop verification and fault-tolerant deployment. The system sends a real-time network handshake request to the selected target factory node, and the data packet payload encapsulates the minimum bounding rectangle feature vector. (i.e., the rigid boundary requirement of 850mm × 620mm) and the expected delivery timestamp. After receiving the request, the edge gateway of the target node directly reads the register status of the underlying programmable logic controller. In this working scenario, assuming that a main die-cutting unit in the target foundry suddenly experiences a servo motor overload power failure, the alarm bit of the equipment health status word is set to logic 1. The edge gateway detects this abnormal word, determines that the multi-dimensional joint verification function is invalid, and then sends the real-time verification result of the Boolean false status to the cloud. After receiving this false status, the system determines that a concurrent timing state failure has occurred, immediately triggers the closed-loop state rollback procedure, and forcibly changes the state of the target node in the memory proxy model. The value is overwritten as false, and the system physically isolates it to prevent it from being selected again by the optimization algorithm. Subsequently, the system extracts the second-best solution node in terms of overall fitness from the solution set cached by the collaborative optimization decision module (module 40), and regenerates the instruction to initiate a network handshake request. The edge gateway of the second-best solution node is verified to have its available production scheduling time window and machine physical space both meeting the inclusiveness requirements, indicating normal equipment operation. It then reports a Boolean true state to the cloud. Upon receiving the true confirmation, the system directly converts the manufacturing plan into standard execution instructions and issues them for production. The entire state rollback and second-best addressing process completes closed-loop scheduling in a very short time, without causing deadlock or stagnation in the upper-layer business logic.

[0050] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent packaging engineering cloud service analysis system based on industry chain collaboration, characterized in that, include: The data perception and fusion module is used to acquire environmental and material parameters of the collaborative foundry nodes to calculate the dynamic folding compensation coefficient and generate a unified set of engineering parameters. The cloud-based graph management module is used to extract the geographic grid coordinates and Boolean response status of collaborative contract manufacturing nodes, construct an in-memory proxy model, and generate a gridded logistics distance benchmark matrix. The reverse modeling verification module is used to correct the preset dimension reduction mapping matrix using the dynamic folding compensation coefficient to generate a two-dimensional unfolded tool line coordinate set, perform self-intersection detection on the two-dimensional unfolded tool line coordinate set and extract the minimum bounding rectangle feature vector. The collaborative optimization decision-making module is used to calculate the comprehensive fitness in the memory proxy model based on the unified engineering parameter set, the gridded logistics distance benchmark matrix and the minimum bounding rectangle feature vector, update the target weight coefficient and output the optimal parameter set and target factory node. The fault-tolerant scheduling execution module is used to send a network handshake request to the target factory node and, based on the real-time verification results fed back by the target factory node, convert the optimal parameter set into an instruction to be issued or to trigger a closed-loop state rollback procedure. 2.The industrial chain coordination based intelligent packaging engineering cloud service analysis system according to claim 1, characterized in that, The data perception and fusion module is specifically used for: The workshop ambient humidity, equipment reference pressure, and initial material thickness are obtained via industrial bus. The independent characteristic components and coupled cross-term coefficients corresponding to the workshop environmental humidity, the equipment reference pressure, and the initial thickness of the material are extracted using a pre-set multivariable polynomial fitting model. The dimensionless scalar values ​​are then calculated as the dynamic folding compensation coefficients. The three-dimensional design baseline parameters and the dynamic folding compensation coefficient are subjected to range standardization to generate a normalized engineering feature vector; The normalized engineering feature vector is spliced ​​using a weighted fusion equation to output the unified engineering parameter set. 3.The industrial chain coordination based intelligent packaging engineering cloud service analysis system according to claim 1, characterized in that, The cloud-based map management module is specifically used for: Retrieve the geographic grid coordinates and Boolean response status of the collaborating contract manufacturer nodes; The geographic grid coordinates and the Boolean response state are converted into key-value pairs and loaded into a high-concurrency in-memory database cluster to construct the in-memory proxy model; A live timestamp attribute is set for the key-value pair structure. When the live timestamp attribute times out and no heartbeat message is received, the Boolean response status is forcibly overwritten to false. Based on the geographic grid coordinates of the collaborative contract manufacturer nodes, the Manhattan distance algorithm is used to calculate the relative distance between each pair of collaborative contract manufacturer nodes, generate the gridded logistics distance reference matrix, and fill the main diagonal elements of the gridded logistics distance reference matrix with a preset minimum constant. 4.The industrial chain coordination based intelligent packaging engineering cloud service analysis system according to claim 1, characterized in that, The reverse modeling verification module is specifically used for: Extract the angle variable between the direction vector of the two-dimensional blade line segment contained in the dimensionality reduction mapping matrix and the direction of the original texture of the material; The dynamic folding compensation coefficient is used as an offset operator, and the corrected two-dimensional line segment mapping weight is calculated by combining the included angle variable and the material anisotropy constant. The actual cutting length is calculated by multiplying the corrected two-dimensional line segment mapping weight as a scale scaling factor by the reference length of the two-dimensional knife line segment. Using the topological connection node of the two-dimensional knife line segment as the reference origin, perform a geometric affine transformation along the direction vector to derive the absolute coordinates of the extended endpoints. For the adjacent line segment breakpoints that have undergone relative offset, perform line equation solving and vector intersection processing to regenerate the coordinates of the closed corner points and output the coordinate set of the two-dimensional unfolded knife line. 5.The industrial chain coordination based intelligent packaging engineering cloud service analysis system according to claim 1, characterized in that, The collaborative optimization decision-making module is specifically used for: Extract the maximum length and width dimensions of the two-dimensional boundary contained in the feature vector of the minimum bounding rectangle, and select the cooperative foundry nodes with a Boolean response state of true and a device limit processing area size greater than or equal to the maximum length and width dimensions of the two-dimensional boundary in the memory proxy model as feasible foundry nodes. The baseline physical distance corresponding to the feasible contract manufacturing node is obtained by calling the gridded logistics distance benchmark matrix. The absolute values ​​of carbon emission estimates and logistics time are calculated by combining the single-vehicle load parameters and the benchmark physical distance. These are then mapped and transformed into carbon emission estimates and logistics suitability indicators through a range normalization function. Material cost and structural strength indicators are extracted from the unified engineering parameter set, and a joint objective function is constructed by combining the carbon emission estimate and the logistics adaptability indicator to calculate the comprehensive adaptability. 6.The industrial chain coordination based intelligent packaging engineering cloud service analysis system according to claim 1, characterized in that, The fault-tolerant scheduling execution module is specifically used for: Extract the available production queue idle time window of the target factory node and perform time sequence overlap verification with the expected delivery timestamp; extract the actual available area of ​​the die-cutting machine in the idle state and perform spatial inclusion comparison with the feature vector of the minimum bounding rectangle; and combine the equipment health status words to determine the real-time verification result. When the real-time verification result is true, the optimal parameter set is converted into a manufacturing execution command and issued for production. When the real-time verification result is false, the closed-loop state rollback procedure is initiated to overwrite the Boolean response state of the target factory node in the memory proxy model as false and reset the live timestamp attribute. Extract the second-best solution node from the optimal parameter set, ranked first in descending order of comprehensive fitness, and resend the network handshake request. 7.The industrial chain collaboration based intelligent packaging engineering cloud service analysis system according to claim 1, characterized in that, The reverse modeling verification module, when performing self-intersection detection on the two-dimensional unfolded toolline coordinate set and extracting the minimum bounding rectangle feature vector, is specifically used for: The two-dimensional unfolded toolline coordinate set is closed and divided into discrete polygonal regions, and a polygonal Boolean intersection operation is performed. When it is detected that any two non-adjacent closed surface regions have an intersecting set and the calculated area of ​​the intersecting set is greater than the preset interference tolerance area threshold, it is determined that the surface region self-intersection phenomenon has occurred and the candidate solution verification failure is marked. When no self-intersection of the surface region occurs, the two-dimensional spatial length and width components of the bounding box of the two-dimensional unfolded knife line coordinate set are recorded to extract the feature vector of the minimum bounding rectangle.

8. The intelligent packaging engineering cloud service analysis system based on supply chain collaboration according to claim 5, characterized in that, When updating the target weight coefficients, the collaborative optimization decision-making module is specifically used for: During the optimization iteration process, the difference between the average carbon emission estimate of all individuals in the current population and the average carbon emission estimate of the previous generation population is calculated. When the average carbon emission estimate exceeds the tolerance limit, a cross-dimensional linkage mechanism is triggered, which amplifies the environmental weight coefficient by a preset step size within the single weight safety limit threshold range, and compresses the remaining weight coefficients proportionally and equally. When the feasible foundry node is empty after completing the full graph traversal in the memory proxy model, the carbon emission estimate and the logistics fit index are set as the worst penalty benchmark, and a penalty factor is introduced to perform a downgrade process on the comprehensive fitness to construct a penalty fitness function.

9. The intelligent packaging engineering cloud service analysis system based on supply chain collaboration according to claim 6, characterized in that, When the fault-tolerant scheduling execution module extracts the second-best solution node ranked first in descending order of comprehensive fitness from the optimal parameter set and resends the network handshake request, it is specifically used for: Record the cumulative number of supplementary retries and introduce a maximum fault tolerance depth coefficient; When the cumulative number of supplementary retries is less than or equal to the maximum fault tolerance depth coefficient, a new network handshake request is generated for the suboptimal solution node to perform a fault tolerance loop; When the cumulative number of replacement retries exceeds the maximum fault tolerance depth coefficient, a circuit breaker mechanism is triggered to terminate the replacement operation and a manual intervention alert is sent.

10. A cloud service analysis method for intelligent packaging engineering based on supply chain collaboration, applied to the cloud service analysis system for intelligent packaging engineering based on supply chain collaboration as described in any one of claims 1-9, characterized in that, Includes the following steps: Obtain environmental and material parameters of the collaborative foundry nodes to calculate the dynamic folding compensation coefficient and generate a unified set of engineering parameters; Extract the geographic grid coordinates and Boolean response states of the collaborative contract manufacturer nodes, construct an in-memory proxy model, and generate a gridded logistics distance benchmark matrix; The preset dimension reduction mapping matrix is ​​corrected using the dynamic folding compensation coefficient to generate a two-dimensional unfolded tool line coordinate set. Self-intersection detection is performed on the two-dimensional unfolded tool line coordinate set and the minimum bounding rectangle feature vector is extracted. Based on the unified engineering parameter set, the gridded logistics distance benchmark matrix, and the minimum bounding rectangle feature vector, the comprehensive fitness is calculated in the memory proxy model, the target weight coefficient is updated, and the optimal parameter set and target factory node are output. A network handshake request is sent to the target factory node, and the optimal parameter set is converted into an instruction to be issued or a closed-loop state rollback procedure is triggered based on the real-time verification result fed back by the target factory node.

Images