Environment-friendly plastic product extrusion molding intelligent parameter control system and control method

Abstract

The application relates to the technical field of plastic processing, and discloses an intelligent parameter control system and control method for extrusion molding of environment-friendly plastic products. The system comprises a data acquisition and fusion module, a cause-and-effect correlation construction module, a digital twin and virtual-real interaction verification module, an intelligent control module, a self-adaptive reconstruction module, a man-machine interaction and visualization module. The method comprises the following steps: collecting data in real time and generating features through deep fusion; dynamically constructing a knowledge graph based on the fusion features and synchronously updating a cause-and-effect topology; constructing a digital twin model, virtually running and predicting; generating instructions to control microstructures; identifying and classifying disturbances, reconstructing the cause-and-effect topology to guide strategy adjustment; and displaying the system state in real time to support operation intervention. The application realizes real-time acquisition and deep fusion of online microstructure data through the data acquisition and fusion module, and combines the microstructure to guide a reverse deduction unit, constructs a deep correlation model, and effectively solves the problems of large batch differences and unstable physical properties of environment-friendly plastic materials.

Description

Technical Field

[0001] This invention relates to the field of plastic processing technology, specifically to an intelligent parameter control system and control method for extrusion molding of environmentally friendly plastic products. Background Technology

[0002] The increasingly severe global environmental problems have driven the rapid development and widespread application of environmentally friendly plastics (such as biodegradable plastics, recycled plastics, and bio-based plastics). These materials, with their unique environmentally friendly properties, are gradually replacing traditional plastics, demonstrating enormous market potential in packaging, agricultural films, automobiles, and electronics. However, compared to traditional general-purpose plastics, environmentally friendly plastics have more complex material properties and exhibit significant batch-to-batch variations.

[0003] Existing environmentally friendly plastic extrusion molding typically follows and improves upon the processing technology of general-purpose plastics. Its core lies in transforming solid plastic particles into continuous products through screw rotation, heating, and the shaping action of a die. In actual production, engineers primarily rely on preset process parameters (such as temperature, pressure, screw speed, and traction speed), combined with operational experience for manual adjustments and offline quality inspection.

[0004] However, when existing technologies are applied to environmentally friendly plastics with variable properties, they neglect the decisive influence of microstructure on the final mechanical, optical, and barrier properties of these plastics. This makes it difficult to control product performance at the microscopic level. When material batches change, equipment wears out, or external environmental disturbances occur, traditional systems struggle to autonomously and quickly identify and adapt to these batch changes, resulting in often delayed system adjustments and reliance on repeated trial and error based on human experience. Therefore, this invention provides an intelligent parameter control system and method for extrusion molding of environmentally friendly plastic products to address the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent parameter control system and method for extrusion molding of environmentally friendly plastic products. This system solves the problems of complex and variable material properties, strong coupling of multiple parameters, difficulty in online control of microstructure, and susceptibility to unexpected disturbances in the extrusion molding process of existing environmentally friendly plastics.

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

[0007] The first aspect of this invention provides an intelligent parameter control system for extrusion molding of environmentally friendly plastic products, the structure of which is as follows:

[0008] This system includes a data acquisition and fusion module, a causal relationship construction module, a digital twin and virtual-real interaction verification module, an intelligent control module, an adaptive reconstruction module, and a human-computer interaction and visualization module.

[0009] The data acquisition and fusion module is used to acquire macroscopic process parameter data, product macroscopic quality data, melt property data, and molecular microstructure data of the melt or semi-solid state in real time during the extrusion process through configured macroscopic process parameter sensors, product macroscopic quality sensors, online property sensors, and online microstructure characterization units. The macroscopic process parameter sensors can acquire the temperature of each heating zone of the extruder. Melt pressure Die head pressure Screw speed Feed rate, motor power Motor current Torque and product output The macroscopic quality sensor for the product can obtain the product width. ,thickness, Surface defect index and surface smoothness The online property sensor can acquire the apparent viscosity of the melt. Elastic modulus Component characteristics Degradation index Purity of recycled materials and the content of specific additives The online microstructure characterization unit can employ an online small-angle X-ray scattering module, an online wide-angle X-ray diffraction module, or an online dielectric spectroscopy module to obtain molecular orientation. Crystallinity Crystallographic characteristics Degree of phase separation Molecular chain relaxation time and polarity index The data acquisition and fusion module further includes a data fusion unit, used to perform timestamp synchronization, data cleaning, missing value processing, noise filtering, and normalization on the real-time acquired macroscopic process parameter data, product macroscopic quality data, melt physical property data, and molecular microstructure data. The normalization process can be performed using the following formula:

[0010] ;

[0011] In the formula, This is the original data; The minimum value of the data; This represents the maximum value of the data.

[0012] The data fusion unit then performs deep fusion of the processed data through tensor decomposition or multimodal deep neural networks to generate fused features. .

[0013] The causal association construction module, connected to the data acquisition and fusion module, is used to dynamically construct and update in real time the extrusion molding knowledge graph and the dynamic causal association topology between multimodal data, process parameters, and product performance based on the fusion features. The causal association construction module includes a knowledge graph construction unit and a dynamic causal association learning unit. The knowledge graph construction unit integrates domain expert experience, historical production batch data, and a materials database to construct an initial extrusion molding knowledge graph. In the formula For a collection of entities, This represents a set of relations. The knowledge graph construction unit dynamically updates the knowledge graph using knowledge graph embedding techniques (e.g., the TransE model) and graph neural networks (e.g., GCN or GraphSAGE). The message passing mechanism of the graph neural network can be represented as:

[0014] ;

[0015] In the formula, For nodes In the The feature vector of the layer; For nodes The set of neighbors; This is a normalization constant; The weight matrix is ​​a learnable weight matrix; This is the activation function.

[0016] The dynamic causal association learning unit is used to autonomously learn and dynamically construct a nonlinear causal relationship network between multimodal data, process parameters, and product performance based on the fused features and through causal discovery algorithms (e.g., PC algorithm or FCI algorithm). In the formula For a set of variables, A set of directed edges represents a causal relationship. The nonlinear causal relationship network is the dynamic causal association topology. The dynamic causal association topology is updated in real time based on new data.

[0017] The digital twin and virtual-real interaction verification module, connected to the causal correlation construction module, is used to construct a high-fidelity digital twin model of the environmentally friendly plastic extrusion molding process. The digital twin and virtual-real interaction verification module includes a multiphysics coupling modeling unit, a material constitutive model integration unit, and a virtual-real interaction prediction unit. The multiphysics coupling modeling unit is used to construct a digital twin model including the extruder geometry, thermodynamics, fluid dynamics, and mass transfer processes. The physical model may include continuity equations, momentum equations, and energy equations. For example, the momentum equation can be expressed as:

[0018] ;

[0019] In the formula, Density; It is the velocity vector; For pressure; For stress tensor; For gravity. The material constitutive model integration unit is used to integrate non-Newtonian rheological constitutive models (e.g., power-law models) of environmentally friendly plastics. In the formula, Apparent viscosity; This is the consistency coefficient; Shear rate; (e.g., the Avrami equation) (for example, the law exponent), thermodynamic parameter models, and crystallization kinetic models. In the formula Crystallinity; This is the crystallization rate constant; The Avrami index and degradation kinetics model are integrated into the digital twin model. The digital twin model can be corrected in real time based on the online physical property data. The virtual-real interaction prediction unit is used to map the fused features to the digital twin model in real time, synchronously drive the digital twin model to run virtually, and predict the potential impact of the control strategy generated by the intelligent control module on melt flow, temperature distribution, microstructure evolution, and macroscopic performance of the product.

[0020] The intelligent control module, connected to the digital twin and virtual-real interaction verification module, is used to generate process parameter adjustment instructions based on the prediction results of the digital twin model, the knowledge graph, and the dynamic causal relationship topology, to achieve macroscopic performance control guided by the microstructure of the product. The intelligent control module includes a microstructure-guided inverse deduction unit, a reinforcement learning prediction control unit, and a meta-learning generalization unit. The microstructure-guided inverse deduction unit is used to construct a deep correlation model between microscopic features and macroscopic performance using a deep generative model (e.g., variational autoencoder or generative adversarial network). Combining the knowledge graph and the dynamic causal relationship topology, the microstructure-guided inverse deduction unit calculates the instantaneous physical field distribution required for the target microstructure (e.g., target molecular orientation, crystallinity, or crystal form characteristics) or macroscopic performance using an inverse deduction optimization algorithm (e.g., gradient descent-based inverse optimization or Monte Carlo tree search). (e.g., a specific shear rate) Temperature gradient Pressure curve Cooling rate This calculation can be represented as an optimization problem:

[0021] ;

[0022] In the formula, For the target microstructure; For target macroscopic performance; Predict( ) is a physical field The predicted microstructure and macroscopic properties under the influence of the action.

[0023] The microstructure guides the reverse inference unit, which then reverse maps the instantaneous physical field distribution to generate specific process parameter adjustment commands (e.g., screw speed adjustment). Temperature adjustment amount for each heating zone Die head clearance adjustment amount traction speed adjustment amount ).

[0024] The reinforcement learning prediction control unit is used to construct a dynamic environment model of the extrusion molding process. This model can predict actions under different process parameters. State transition under and rewards :

[0025] ;

[0026] In the formula, The current state. This is the current action. The reinforcement learning prediction control unit interacts with the environment model or the digital twin model through a reinforcement learning agent (e.g., based on a deep Q-network or actor-critic architecture) to learn process parameter adjustment strategies. .

[0027] The reinforcement learning agent selects the optimal action by maximizing long-term cumulative reward, the reward function. Taking into account the overall quality of the product Production efficiency Energy consumption Deviation of process parameters in The current state. This is the current action. The reinforcement learning prediction control unit interacts with the environment model or the digital twin model through a reinforcement learning agent (e.g., based on a deep Q-network or actor-critic architecture) to learn process parameter adjustment strategies. The reinforcement learning agent selects the optimal action by maximizing long-term cumulative reward, and the reward function... Taking into account the overall quality of the product Production efficiency Energy consumption Deviation of process parameters and the deviation between the microstructure and the target value .

[0028] The meta-learning generalization unit is used to enable the reinforcement learning agent to quickly adapt to new environments and tasks through meta-learning methods (e.g., MAML or Reptile), and to quickly adjust the optimal control strategy after the dynamic causal association topology reconstruction or when encountering new material batches.

[0029] The adaptive reconstruction module, connected to the data acquisition and fusion module and the intelligent control module, is used to identify and classify unexpected disturbances and drive the causal association construction module to adaptively reconstruct the dynamic causal association topology, thereby guiding the intelligent control module to adjust its control strategy. The adaptive reconstruction module includes an anomaly pattern recognition and classification unit, a dynamic association topology reconstruction unit, and an intrinsic safety and trust unit. The anomaly pattern recognition and classification unit uses deep metric learning or unsupervised anomaly detection algorithms (e.g., Isolation Forest or One-Class SVM) to monitor the data flow in the fused features in real time, identifying and classifying anomalous data flows that significantly deviate from known patterns. These anomalous data flows indicate sudden changes in material properties, sensor failures, equipment wear, or changes in the external environment. When the deviation between the actual production effect and the prediction of the digital twin model exceeds a preset threshold, the intrinsic safety and trust unit can switch the control strategy and trigger a rapid learning process.

[0030] The human-computer interaction and visualization module, connected to the aforementioned modules, is used to display the operating status, decision-making process, prediction results, and abnormal early warning information of the control system in real time, and supports operators in setting targets and intervening in parameters. The human-computer interaction and visualization module includes a data visualization interface, a knowledge and decision visualization interface, a digital twin simulation result display interface, an early warning and diagnosis interface, and a parameter setting and intervention interface.

[0031] A second aspect of the present invention provides a method for intelligent parameter control in the extrusion molding of environmentally friendly plastic products, applied to the intelligent parameter control system for the extrusion molding of environmentally friendly plastic products described in the first aspect of the present invention. The method includes the following steps:

[0032] S1. Macroscopic process parameters, product quality data, melt property data, and molecular microstructure data of melt or semi-solid state are acquired in real time through multi-source sensors during the extrusion process, and the acquired data are deeply fused to generate fusion features.

[0033] S2. Based on the generated fusion features, dynamically construct and update the extrusion molding knowledge graph in real time, and synchronously update the dynamic causal relationship topology between multimodal data, process parameters and product performance.

[0034] S3. Based on the knowledge graph, dynamic causal relationship topology and fusion features, construct a digital twin model of the environmentally friendly plastic extrusion molding process, and enable the digital twin model to perform virtual operation and effect prediction by real-time mapping of fusion features;

[0035] S4. Based on the prediction results of the digital twin model, the knowledge graph, and the dynamic causal relationship topology generation process parameter adjustment instructions, the microstructure of the product is controlled, for example, the orientation degree, crystallinity, or crystal form characteristics of the target molecules determined by the predictive analysis are controlled.

[0036] S5. Identify and classify unexpected disturbances, and guide the adjustment of control strategies by reconstructing the dynamic causal relationship topology.

[0037] S6 displays the real-time operating status, decision-making process, prediction results, and abnormal early warning information of the control system, and supports operators in setting targets and intervening in parameters.

[0038] This invention provides an intelligent parameter control system and method for extruding environmentally friendly plastic products. It offers the following advantages:

[0039] 1. This invention acquires and deeply integrates online microstructure data in real time through a data acquisition and fusion module, and combines it with a microstructure-guided reverse deduction unit in the intelligent control module to construct a deep correlation model between microscopic features and macroscopic performance. This effectively addresses the problems of large batch-to-batch differences and unstable physical properties of environmentally friendly plastic materials, ensuring that the final product's mechanical properties, optical properties, or barrier properties, and other macroscopic indicators, remain within the target range.

[0040] 2. This invention introduces a causal relationship construction module, which can autonomously learn and dynamically update the dynamic causal relationship topology between the extrusion molding knowledge graph and multimodal data, process parameters and product performance. The adaptive reconstruction module can identify and classify abnormal data streams and drive the dynamic causal relationship topology to be reconstructed quickly. It can respond in real time to changes in causal relationships caused by factors such as material batch changes, equipment wear or environmental disturbances. It does not require preset fixed models or a lot of manual experience to adjust, which improves the automation level of the system and reduces the difficulty of operation and the requirements for the professional skills of operators.

[0041] 3. The high-fidelity digital twin model constructed by this invention can perform virtual operation and effect prediction based on real-time data, providing a safe strategy verification environment for the intelligent control module, avoiding the risk of direct trial and error in actual production, enhancing the reliability and security of system decision-making, and effectively avoiding production accidents or waste caused by improper parameter adjustment. Attached Figure Description

[0042] Figure 1This is a schematic diagram of the intelligent parameter control system for extrusion molding of environmentally friendly plastic products according to the present invention;

[0043] Figure 2 This is a schematic diagram of the data acquisition and fusion module structure of the present invention;

[0044] Figure 3 This is a schematic diagram of the causal association construction module structure of the present invention;

[0045] Figure 4 This is a schematic diagram of the structure of the digital twin and virtual-real interaction verification module of the present invention;

[0046] Figure 5 This is a schematic diagram of the intelligent control module structure of the present invention;

[0047] Figure 6 This is a schematic diagram of the adaptive reconfiguration module structure of the present invention;

[0048] Figure 7 This is a schematic diagram of the human-computer interaction and visualization module structure of the present invention.

[0049] Among them, 100 is the data acquisition and fusion module; 200 is the causal relationship construction module; 300 is the digital twin and virtual-real interaction verification module; 400 is the intelligent control module; 500 is the adaptive reconstruction module; and 600 is the human-computer interaction and visualization module. Detailed Implementation

[0050] 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.

[0051] See attached document Figure 1 , Figure 1 This is a schematic diagram of an intelligent parameter control system for extrusion molding of environmentally friendly plastic products according to an embodiment of the present invention. The present invention provides an intelligent parameter control system for extrusion molding of environmentally friendly plastic products, including: a data acquisition and fusion module 100, a causal correlation construction module 200, a digital twin and virtual-real interaction verification module 300, an intelligent control module 400, an adaptive reconstruction module 500, and a human-computer interaction and visualization module 600.

[0052] The data acquisition and fusion module 100 is used to acquire multi-source data in real time during the extrusion process, specifically including macroscopic process parameters, macroscopic quality data of the product, melt physical property data, and molecular microstructure data of the melt or semi-solid state. This module performs timestamp synchronization, data cleaning, missing value processing, noise filtering, and normalization on the acquired heterogeneous data, and further generates fusion features through deep fusion technology.

[0053] The causal association construction module 200, connected to the data acquisition and fusion module 100, is used to receive fused features. This module dynamically constructs and updates the extrusion molding knowledge graph in real time based on the fused features, and synchronously updates the dynamic causal association topology between multimodal data, process parameters, and product performance.

[0054] The digital twin and virtual-real interaction verification module 300, connected to the causal association construction module 200, is used to construct a high-fidelity digital twin model of the environmentally friendly plastic extrusion molding process. This module, based on knowledge graphs, dynamic causal association topologies, and fusion features, maps fusion features in real time to drive the digital twin model to perform virtual operation and effect prediction.

[0055] The intelligent control module 400 is connected to the digital twin and virtual-real interaction verification module 300. Based on the prediction results of the digital twin model, the knowledge graph, and the dynamic causal relationship topology, this module generates process parameter adjustment instructions to regulate the microstructure of the product.

[0056] The adaptive reconfiguration module 500 is connected to the data acquisition and fusion module 100 and the intelligent control module 400. This module is used to identify and classify unexpected disturbances appearing in the data stream. When an unexpected disturbance is identified, this module drives the causal association construction module 200 to adaptively reconstruct the dynamic causal association topology. The reconstructed dynamic causal association topology then guides the intelligent control module 400 to adjust its control strategy.

[0057] The human-computer interaction and visualization module 600 is connected to the modules mentioned above. This module is used to display the operating status, decision-making process, prediction results, and abnormal early warning information of the control system in real time, and supports operators in setting targets and intervening in parameters.

[0058] The system's workflow is as follows: The data acquisition and fusion module 100 continuously acquires multi-dimensional real-time data from the extrusion production line and processes and fuses it into unified fusion features. These fusion features are transmitted to the causal association construction module 200 for continuous updating and improvement of the extrusion molding knowledge graph and dynamic causal association topology. Simultaneously, the digital twin and virtual-real interaction verification module 300 utilizes this knowledge, topology, and fusion features to construct and drive a high-fidelity digital twin model in real time, performing virtual simulation and effect prediction of the actual extrusion process. The intelligent control module 400 receives the prediction results from the digital twin model, as well as the knowledge graph and dynamic causal association topology, and generates optimized process parameter adjustment instructions based on this information to control the extruder, thereby achieving precise control of the product's microstructure. The adaptive reconstruction module 500 monitors the data flow; once an unexpected disturbance is detected, it triggers the causal association construction module 200 to reconstruct the causal association topology and guides the intelligent control module 400 to adjust its control strategy to cope with the changes. The Human-Computer Interaction and Visualization Module 600 provides comprehensive information display and operation interfaces, ensuring the transparency and controllability of system operation.

[0059] See attached document Figure 2 , Figure 2 This is a schematic diagram of the structure of a data acquisition and fusion module 100 according to an embodiment of the present invention.

[0060] The data acquisition and fusion module 100 is used to acquire multi-source, multi-dimensional data during the extrusion molding process. This module includes macroscopic process parameter sensors, product macroscopic quality sensors, online physical property sensors, an online microstructure characterization unit, and a data fusion unit.

[0061] Macroscopic process parameter sensors are deployed at corresponding locations within the extruder to collect real-time temperatures in each heating zone of the extruder. Melt pressure Die head pressure Screw speed Feed rate Motor current Torque and product output These sensors output data at a preset frequency (e.g., 10 times per second).

[0062] A macroscopic quality sensor for the extruded product is deployed in the detection area after the product exits the extrusion line to obtain the product width in real time. ,thickness Surface defect index (e.g., identification and quantification via machine vision systems) and surface finish Data such as...

[0063] Online property sensors are integrated into the extruder die or melt delivery pipeline to obtain the apparent viscosity of the melt in real time. Elastic modulus Component characteristics Degradation index Purity of recycled materials and the content of specific additives Data such as online rheometers, near-infrared spectrometers, or differential scanning calorimeters (DSC) can be used to obtain online physical property sensors.

[0064] Online microstructure characterization units are deployed near the extrusion die exit or in the cooling section to acquire molecular microstructure data of the melt or semi-solid state, specifically including molecular orientation. Crystallinity Crystallographic characteristics Degree of phase separation Molecular chain relaxation time and polarity index The online microstructure characterization unit can be any one or a combination of an online small-angle X-ray scattering module, an online wide-angle X-ray diffraction module, or an online dielectric spectroscopy module.

[0065] The data fusion unit connects to the aforementioned sensors and characterization units. Its function is to process the raw data acquired from macroscopic process parameter sensors, product macroscopic quality sensors, online physical property sensors, and online microstructure characterization units. First, the data fusion unit performs timestamp synchronization to ensure that data from different sources are aligned on the timeline. Second, it performs data cleaning and missing value handling, for example, by using interpolation or machine learning-based methods to fill in missing data. Third, it performs noise filtering, for example, by using Kalman filtering or wavelet denoising. Finally, it normalizes the processed data.

[0066] After preprocessing, the data fusion unit performs deep fusion of the processed multi-dimensional and multi-modal data through tensor decomposition or multimodal deep neural networks to generate unified fusion features. This fused feature, as a high-dimensional vector, can comprehensively characterize the macroscopic state, microstructure, and physical properties of the current extrusion process.

[0067] See attached document Figure 3 , Figure 3 This is a schematic diagram of the causal association construction module 200 according to an embodiment of the present invention.

[0068] The causal association construction module 200 is connected to the multi-scale multimodal data acquisition and fusion module 100 to receive fused features. This module's function is to dynamically construct and update the extrusion molding knowledge graph and the dynamic causal relationship topology between multimodal data, process parameters, and product performance in real time. The causal relationship construction module 200 includes a knowledge graph construction unit and a dynamic causal relationship learning unit.

[0069] The knowledge graph construction unit is used to integrate domain expert experience, historical production batch data, and materials databases to build an initial extrusion molding knowledge graph. .in, Represents a collection of entities, such as: environmentally friendly plastic grades, extruder models, process parameters (e.g., screw speed, temperature range), microstructural characteristics (e.g., crystallinity, molecular orientation) and product performance (e.g., tensile strength, impact toughness); Represents a set of relationships, such as: influence, consist of, is a parameter of.

[0070] Knowledge graph construction units dynamically update the knowledge graph using knowledge graph embedding techniques and graph neural networks. For example, the TransE model can be used to learn low-dimensional vector representations of entities and relations, with the objective function aiming to make the sum of the head entity vector and relation vector close to the tail entity vector.

[0071] ;

[0072] In the formula, It is a triple in a knowledge graph. These are the embedding vectors for the head entity, relation, and tail entity, respectively. For negative sample triples For interval parameters, This indicates that a positive value is taken.

[0073] Furthermore, graph neural networks (e.g., graph convolutional networks GCN or GraphSAGE) can be used to aggregate and update node features in knowledge graphs to capture more complex structural information. The message passing mechanism of graph neural networks can be represented as:

[0074] ;

[0075] In the formula, It is a node In the The feature vector of the layer, It is a node The set of neighboring nodes, It is an aggregation function (e.g., summation, averaging, or max pooling) used to aggregate information from neighboring nodes. By continuously receiving new fused features, the knowledge graph building unit can identify new entities and relationships, or update the attributes of existing entities and relationships, thereby keeping the knowledge graph real-time and accurate.

[0076] Dynamic causal association learning units are used to learn based on fused features It can autonomously learn and dynamically construct a nonlinear causal relationship network between multimodal data, process parameters, and product performance. In the formula, It is a set of variables, including various sensor data, process parameters, intermediate state variables, and product performance indicators; Let be a set of directed edges, representing causal relationships between variables.

[0077] This unit can employ causal discovery algorithms, such as PC algorithms, FCI algorithms, or deep learning-based causal discovery methods. These algorithms infer causal graphs by analyzing conditional independence in the data or using structural equation modeling. The nonlinear causal network is a dynamic causal association topology. This dynamic causal association topology is updated in real time based on newly input fused feature data. For example, when new data patterns emerge, the dynamic causal association learning unit can detect new causal paths or changes in the strength of existing causal paths and incrementally update the causal topology to ensure it accurately reflects the true causal mechanism in the current extrusion process.

[0078] See attached document Figure 4 , Figure 4 This is a schematic diagram of the structure of a digital twin and virtual-real interaction verification module according to an embodiment of the present invention.

[0079] The digital twin and virtual-real interaction verification module 300 is connected to the causal association construction module 200 to build a high-fidelity digital twin model of the environmentally friendly plastic extrusion molding process. This module is based on a knowledge graph. Dynamic causal topology and fusion features The system uses real-time mapping and fusion of features to drive the digital twin model in virtual operation and effect prediction. The digital twin and virtual-real interaction verification module 300 includes a multi-physics coupling modeling unit, a material constitutive model integration unit, and a virtual-real interaction prediction unit.

[0080] The multiphysics coupled modeling unit is used to construct a digital twin model that includes the extruder geometry, thermodynamics, fluid dynamics, and mass transfer processes. This digital twin model is a multi-scale, multiphysics coupled simulation model capable of accurately simulating the complex behavior of the melt inside the extruder and the product forming process. The digital twin model couples computational fluid dynamics (CFD) methods to simulate the flow, heat transfer, mixing, and shear behavior of the melt within the screw, barrel, and die. For example, the mass conservation equation (continuity equation) for melt flow can be expressed as:

[0081] ;

[0082] In the formula, For fluid density, For the fluid velocity vector, For time.

[0083] Meanwhile, the digital twin model, combined with the finite element analysis (FEA) method, simulates the deformation of the die head and the forming process of the product, including cooling shrinkage, stress distribution, and the formation of the final shape.

[0084] The material constitutive model integration unit is used to integrate non-Newtonian rheological constitutive models, thermodynamic parameter models, crystallization kinetic models, and degradation kinetic models of environmentally friendly plastics into a digital twin model. For example, the power-law model commonly used for non-Newtonian fluids can be expressed as:

[0085] ;

[0086] In the formula, For shear stress, This is the consistency coefficient. Shear rate, This is the power-law exponent.

[0087] These model parameters can be dynamically corrected based on real-time data acquired from online property sensors, ensuring that the digital twin model accurately reflects the actual physicochemical properties of the current batch of environmentally friendly plastics. The digital twin model can be corrected in real time based on online property data, thereby improving its simulation accuracy and prediction precision.

[0088] The virtual-real interaction prediction unit is used to integrate features. The data is mapped to the digital twin model in real time, synchronously driving the digital twin model to run virtually. This real-time mapping mechanism ensures that the state of the digital twin model is highly consistent with the actual extrusion process. The virtual-real interaction prediction unit then uses the digital twin model to virtually verify and predict the effect of the strategy to be generated by the intelligent control module 400, including the strategy's impact on melt flow state, temperature distribution, and microstructure evolution (e.g., molecular orientation). and crystallinity The system can predict the potential impact of changes in the product's macroscopic properties (e.g., final tensile strength, impact strength) and the expected effects of different control strategies, thereby selecting the optimal control scheme.

[0089] See attached document Figure 5 , Figure 5 This is a schematic diagram of the structure of an intelligent control module according to an embodiment of the present invention.

[0090] The intelligent control module 400 is connected to the digital twin and virtual-real interaction verification module 300, and is used for prediction results and knowledge graphs based on the digital twin model. and dynamic causal topology The system generates process parameter adjustment instructions to control the microstructure of the product. The intelligent control module 400 includes a microstructure-guided reverse inference unit, a reinforcement learning prediction control unit, and a meta-learning generalization unit.

[0091] The microstructure-guided inverse reasoning unit is used to construct a deep correlation model between microscopic features and macroscopic performance using deep generative models (e.g., variational autoencoders (VAEs) or generative adversarial networks (GANs)). This deep correlation model can learn microstructures (such as molecular orientation) from fused features. Crystallinity Crystallographic characteristics The complex mapping relationship between physical properties and macroscopic properties (such as tensile strength, impact toughness, and surface finish). Units are integrated with knowledge graphs. and dynamic causal relationship topology The instantaneous physical field distribution required to achieve the target microstructure or macroscopic performance is calculated using inverse deduction optimization algorithms (e.g., gradient descent-based inverse optimization or Monte Carlo Tree Search (MCTS)). This calculation can be formulated as an optimization problem, aiming to find a set of physical field parameters such that the microstructure and macroscopic performance simulated in the digital twin model are as close as possible to the target values:

[0092] ;

[0093] In the formula, Indicates in physical field Microstructure and macroscopic performance predicted by digital twin models under the influence of [the system / mechanism]. For the target microstructure (e.g., a specific degree of molecular orientation or crystallinity), For target macroscopic performance. The physical field distribution may include, but is not limited to, a specific shear rate. Temperature gradient Pressure curve Cooling rate The microstructure guides the reverse engineering unit, which then maps the instantaneous physical field distribution in reverse to generate specific process parameter adjustment commands, such as screw speed adjustment. Temperature adjustment amount for each heating zone Die head clearance adjustment amount Or traction speed adjustment amount .

[0094] Reinforcement learning predictive control unit is used to build a dynamic environment model of the extrusion molding process. This environmental model is able to predict the environment given the current state. and process parameter adjustment actions Under these circumstances, the state at the next moment and the rewards received The state transition process can be represented as:

[0095] ;

[0096] The reinforcement learning prediction control unit learns optimal process parameter tuning strategies by interacting with an environment model or digital twin model through a reinforcement learning agent (e.g., based on a deep Q-network (DQN) or an actor-critic architecture). This reinforcement learning agent selects actions by maximizing long-term cumulative rewards.

[0097] The meta-learning generalization unit is used to enable reinforcement learning agents to quickly adapt to new extrusion environments and new production tasks through meta-learning methods (e.g., Model-Independent Meta-Learning (MAML) or the Reptile algorithm). This unit allows the agent to quickly adjust its learned policy with limited new data or experience with new tasks, rather than learning from scratch. Therefore, when dynamic causal topology reconstruction occurs or new material batches are encountered, the meta-learning generalization unit can guide the reinforcement learning agent to quickly converge to a new optimal control policy.

[0098] See attached document Figure 6 , Figure 6 This is a schematic diagram of the adaptive reconfiguration module structure according to an embodiment of the present invention.

[0099] The adaptive reconstruction module 500 is connected to the multi-scale multimodal data acquisition and fusion module 100 and the intelligent control module 400. It is used to identify and classify unexpected disturbances and drive the causal correlation construction module 200 to reconstruct the dynamic causal correlation topology. The reconstructed dynamic causal correlation topology then guides the intelligent control module 400 to adjust its control strategy. The adaptive reconstruction module 500 includes an anomaly pattern recognition and classification unit, a dynamic correlation topology reconstruction unit, and an intrinsic security and trust unit.

[0100] The anomaly pattern recognition and classification unit is used to monitor fused features in real time using deep metric learning or unsupervised anomaly detection algorithms (e.g., Isolation Forest or One-Class SVM). The unit learns the distribution of normal data patterns to identify anomalous data streams that deviate significantly from known patterns. Anomalous data streams indicate potential unexpected disturbances, such as abrupt changes in material properties (e.g., inconsistent batches of recycled materials), sensor malfunctions, equipment wear, or changes in the external environment (e.g., fluctuations in workshop temperature). Anomaly detection typically involves calculating an "anomaly score" between data points and the normal data distribution. ,when If the threshold is exceeded, it is considered abnormal.

[0101] The dynamic correlation topology reconstruction unit is used to drive the dynamic causal correlation learning unit in the causal correlation construction module 200 to quickly rerun or reinforce the learning process after identifying unexpected perturbations. This reconstruction process aims to update or correct the current dynamic causal correlation topology. This ensures that the dynamic correlation topology reconstruction unit accurately reflects changes in the internal causal relationships of the system caused by disturbances. For example, changes in material composition may alter the strength of the causal relationship between certain process parameters and the microstructure of the product, or lead to new causal paths. The dynamic correlation topology reconstruction unit ensures that the causal topology can quickly adapt to these changes.

[0102] The intrinsic safety and trust unit is used to introduce safety constraint reinforcement learning to ensure that process parameter adjustment commands are always within the preset safe operating range. Internal execution. This unit achieves this goal by adding a penalty term to the reward function of reinforcement learning or through constraint optimization methods. For example, action selection can be constrained in the following ways:

[0103] ;

[0104] In the formula, To reinforce the loss function of learning, This is the safe operating range for process parameters. Furthermore, when the actual production results deviate from the predictions of the digital twin model 300 beyond a preset threshold (e.g., the actual product thickness deviates from the predicted value by more than a set percentage), the intrinsic safety and trust unit can trigger an emergency response mechanism, switch to a preset safety control strategy, and simultaneously trigger the intelligent control module 400 to quickly learn and adapt to the new operating conditions and restore optimal control as soon as possible.

[0105] See attached document Figure 7 , Figure 7 This is a schematic diagram of the human-computer interaction and visualization module structure according to an embodiment of the present invention.

[0106] The human-computer interaction and visualization module 600 connects to the aforementioned modules and is used to display the operating status, decision-making process, prediction results, and abnormal early warning information of the control system in real time, and supports operators in setting targets and intervening in parameters. The human-computer interaction and visualization module 600 includes a data visualization interface, a knowledge and decision visualization interface, a digital twin simulation result display interface, an early warning and diagnosis interface, and a parameter setting and intervention interface.

[0107] The data visualization interface is used to graphically display in real time various types of data collected by the multi-scale, multi-modal data acquisition and fusion module 100, including macroscopic process parameters (such as temperature curves and pressure fluctuations), macroscopic quality data of products (such as width and thickness trend charts), melt physical property data (such as viscosity changes), and molecular microstructure data (such as changes in crystallinity over time). The data can be presented in the form of line charts, bar charts, scatter plots, or real-time curves to facilitate operators' monitoring of various indicators in the production process.

[0108] The knowledge and decision visualization interface is used to display the extrusion molding knowledge graph and dynamic causal relationship topology constructed by the causal relationship construction module 200. The knowledge graph can be displayed in the form of a node-edge graph, with entities and relationships clearly visible. The dynamic causal relationship topology is presented in the form of a causal graph, showing the strength and direction of the causal relationships between variables. In addition, this interface can also display the reasoning path and basis for the decisions generated by the intelligent control module 400, for example, explaining that the adjustment of a certain process parameter is based on the reverse inference result of the target microstructure.

[0109] The digital twin simulation results display interface is used to showcase the simulation results of the digital twin and the virtual-real interaction verification module 300. This interface can display in real-time dynamic 3D visualizations of the temperature, pressure, and velocity fields of the melt simulated by the digital twin model within the screw, barrel, and die, as well as predictions of stress, strain, and final shape during the product forming process. For example, the shear rate distribution inside the die can be visualized; the shear rate at a certain location can be expressed as:

[0110] ;

[0111] In the formula, It is the deformation rate tensor; denoted as shear rate.

[0112] The early warning and diagnostic interface receives and displays abnormal warning information issued by the adaptive reconfiguration module 500. This interface clearly indicates the type of abnormality (such as sensor failure, batch mutation, or equipment wear) and provides a diagnostic report indicating possible causes and the scope of impact. Simultaneously, this interface provides real-time alerts to the operator when the intrinsic safety and trust unit triggers a safety mode or rapid learning process.

[0113] The parameter setting and intervention interface supports operators in setting targets and intervening in system parameters. Operators can input desired macroscopic performance or microscopic structural targets through this interface, such as setting target tensile strength or target crystallinity. Simultaneously, operators can also manually intervene or correct process parameter adjustment commands generated by the intelligent control module 400 under specific circumstances, and observe the system response and digital twin model prediction results after intervention.

[0114] 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 parameter control system for extrusion molding of environmentally friendly plastic products, characterized in that, include: The data acquisition and fusion module is used to acquire macroscopic process parameters, product quality data, melt property data, and molecular microstructure data of melt or semi-solid state in real time through multi-source sensors during the extrusion process, and to perform deep fusion of the real-time acquired data to generate fusion features. The causal association construction module is used to dynamically construct and update the extrusion molding knowledge graph in real time based on the generated fusion features, and synchronously update the dynamic causal association topology between multimodal data, process parameters and product performance. The digital twin and virtual-real interaction verification module is used to construct a digital twin model of the environmentally friendly plastic extrusion molding process based on knowledge graphs, dynamic causal relationship topology and fusion features, and enable the digital twin model to run virtually and predict effects through real-time mapping of fusion features. The intelligent control module is used to generate process parameter adjustment instructions based on the prediction results of the digital twin model, knowledge graph and dynamic causal relationship topology, so as to regulate the microstructure of the product. The adaptive reconfiguration module is used to identify and classify unexpected disturbances, and reconstruct the dynamic causal relationship topology through the causal relationship construction module. The reconstructed dynamic causal relationship topology guides the intelligent control module to adjust the control strategy. The human-computer interaction and visualization module is used to display the operating status, decision-making process, prediction results and abnormal warning information of the control system in real time, and supports operators to set targets and intervene in parameters. The digital twin and virtual-real interaction verification module includes: The multiphysics coupling modeling unit is used to construct a digital twin model that includes the geometry of the extruder, thermodynamics, fluid dynamics, and mass transfer process. The digital twin model is coupled with computational fluid dynamics to simulate melt flow and heat transfer, and combined with finite element analysis to simulate die deformation and product forming. The material constitutive model integration unit is used to integrate the non-Newtonian rheological constitutive model, thermodynamic parameter model, crystallization kinetic model and degradation kinetic model of environmentally friendly plastics into the digital twin model. The digital twin model can be corrected in real time based on online physical property data. The virtual-real interaction prediction unit is used to map the fused features to the digital twin model in real time, synchronously drive the digital twin model to run virtually, and predict the impact of the control strategy generated by the intelligent control module on melt flow, temperature distribution, microstructure evolution and product macroscopic performance.

2. The intelligent parameter control system for extrusion molding of environmentally friendly plastic products according to claim 1, characterized in that, The data acquisition and fusion module includes: Macroscopic process parameter sensors are used to acquire the temperature of each heating zone of the extruder, melt pressure, die pressure, screw speed, feed rate, motor current, torque, and product output; Macroscopic quality sensor for finished products, used to acquire product width, thickness, surface defect index and surface finish; Online physical property sensors are used to acquire melt apparent viscosity, elastic modulus, component characteristics, degradation index, recycled material purity, and the content of specific additives; Online microstructure characterization unit is used to obtain molecular orientation degree, crystallinity, crystal form characteristics, phase separation degree, molecular chain relaxation time and polarity index; The data fusion unit is used to perform timestamp synchronization, data cleaning, missing value handling, noise filtering and normalization on the acquired data, and to deeply fuse the processed data through tensor decomposition or multimodal deep neural networks to generate fused features.

3. The intelligent parameter control system for extrusion molding of environmentally friendly plastic products according to claim 1, characterized in that, The causal association construction module includes: The knowledge graph construction unit is used to integrate historical production data and material databases to build an extrusion molding knowledge graph. The knowledge graph is dynamically updated through knowledge graph embedding technology and graph neural networks. The dynamic causal association learning unit is used to autonomously learn and dynamically construct a nonlinear causal relationship network between multimodal data, process parameters and product performance based on fusion features. The nonlinear causal relationship network is a dynamic causal association topology, which is updated in real time based on new data.

4. The intelligent parameter control system for extrusion molding of environmentally friendly plastic products according to claim 1, characterized in that, The intelligent control module includes: The microstructure-guided reverse inference unit is used to construct a deep correlation model between micro-features and macro-performance using a deep generative model. Combined with knowledge graphs and dynamic causal relationship topologies, it calculates the instantaneous physical field distribution required for the target microstructure or macro-performance through reverse inference algorithms, and generates process parameter adjustment instructions through reverse mapping. The reinforcement learning predictive control unit is used to build a dynamic environment model of the extrusion molding process. It learns process parameter adjustment strategies by interacting with the environment model or digital twin model through reinforcement learning agents. Meta-learning generalization units are used to enable reinforcement learning agents to quickly adapt to new environments and tasks through meta-learning methods, and to quickly adjust the optimal control strategy after dynamic causal topology reconstruction or when encountering new material batches.

5. The intelligent parameter control system for extrusion molding of environmentally friendly plastic products according to claim 4, characterized in that, The intelligent control module guides the reverse inference unit through microstructure to regulate the orientation degree, crystallinity or crystal form characteristics of the target molecule determined by predictive analysis.

6. The intelligent parameter control system for extrusion molding of environmentally friendly plastic products according to claim 1, characterized in that, The adaptive reconstruction module includes: An anomaly pattern recognition and classification unit is used to identify and classify anomalous data streams in fused features using deep metric learning or unsupervised anomaly detection algorithms. The anomalous data streams indicate abrupt changes in material properties, sensor failures, equipment wear, or changes in the external environment. The dynamic correlation topology reconstruction unit is used to drive the dynamic causal correlation learning unit to quickly rerun or reinforce learning after identifying unexpected perturbations, and reconstruct the dynamic causal correlation topology. The intrinsic safety and trust unit is used to introduce safety constraint reinforcement learning to ensure that process parameters are within the safe operating range. When the deviation between the actual effect and the digital twin prediction exceeds the threshold, the control strategy is switched and rapid learning is triggered.

7. The intelligent parameter control system for extrusion molding of environmentally friendly plastic products according to claim 1, characterized in that, The human-computer interaction and visualization module includes: A data visualization interface is used to display data from various sensors, fusion features, and microstructure characterization results in real time. The knowledge and decision visualization interface is used to visualize the current structure, dynamic causal relationship topology, control strategies, and decision-making basis of the extrusion molding knowledge graph. The digital twin simulation results display interface is used to present the simulation results and predicted trends of the digital twin model in real time. The early warning and diagnosis interface is used to display the results of anomaly identification and tracing in a timely manner, and to provide the cause of the failure and suggested handling solutions; The parameter setting and intervention interface is used to support operators in setting product quality targets, microstructure targets, and energy consumption targets.

8. The intelligent parameter control system for extrusion molding of environmentally friendly plastic products according to claim 2, characterized in that, The online microstructure characterization unit employs an online small-angle X-ray scattering module, an online wide-angle X-ray diffraction module, or an online dielectric spectroscopy module.

9. A method for intelligent parameter control in extrusion molding of environmentally friendly plastic products, applied to the intelligent parameter control system for extrusion molding of environmentally friendly plastic products as described in any one of claims 1-8, characterized in that, Includes the following steps: S1. Macroscopic process parameters, product quality data, melt property data, and molecular microstructure data of melt or semi-solid state are acquired in real time through multi-source sensors during the extrusion process, and the acquired data are deeply fused to generate fusion features. S2. Based on the generated fusion features, dynamically construct and update the extrusion molding knowledge graph in real time, and synchronously update the dynamic causal relationship topology between multimodal data, process parameters and product performance. S3. Based on the knowledge graph, dynamic causal relationship topology and fusion features, construct a digital twin model of the environmentally friendly plastic extrusion molding process, and enable the digital twin model to perform virtual operation and effect prediction by real-time mapping of fusion features; S4. Based on the prediction results of the digital twin model, the knowledge graph, and the dynamic causal relationship topology, process parameter adjustment instructions are generated to regulate the microstructure of the product; S5. Identify and classify unexpected disturbances, and guide the adjustment of control strategies by reconstructing the dynamic causal relationship topology. S6 displays the real-time operating status, decision-making process, prediction results, and abnormal early warning information of the control system, and supports operators in setting targets and intervening in parameters.

Images