Intelligent PE pipe manufacturing whole process digital management system and method

Abstract

The invention discloses an intelligent PE pipe manufacturing whole process digital management system and method. The system comprises a data acquisition module, a digital twinning module, an analysis optimization module, a detection module and an integration module. The data acquisition module forms a first signal; the digital twinning module receives the first signal and constructs a digital twinning model, material characteristic parameters and environmental data are fused to simulate a pipe crystallization process, and a second signal is formed; the analysis optimization module receives simulation data of the prime number second signal and forms a third signal; the detection module receives a defect image and wall thickness data of the detection instrument, forms a fourth signal and transmits the fourth signal to the digital twinning module to update the digital twinning model; and the integration module integrates the optimization instruction of the third signal and the quality constraint of the fourth signal, generates an equipment control instruction and issues the equipment control instruction to a production line. The intelligent PE pipe manufacturing whole process digital management system can solve the problems of data islands, quality fluctuation, energy consumption waste and quality tracing difficulty in PE pipe manufacturing.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing of polymer materials, and particularly to a digital management system and method for the whole process of intelligent manufacturing of PE pipes. Background Art

[0002] As a key material in fields such as municipal engineering and gas transmission, the manufacturing process of polyethylene (PE) pipes involves the coupling of multiple links such as raw material ratio, extrusion molding, and cooling and shaping. The traditional manufacturing mode relies on manual experience to adjust equipment parameters, and there are significant limitations: First, the data of production equipment (such as extruders and cooling systems) are stored independently, and the SCADA system only realizes single-machine monitoring, lacking the ability of full-process data fusion and analysis, resulting in the lag of process parameter adjustment behind quality fluctuations; Second, PE materials are highly sensitive to temperature, and their melt flow characteristics are significantly affected by environmental humidity and cooling rate, but most existing control systems use fixed PID (Proportional-Integral-Differential Controller) parameters and cannot adapt to raw material batch differences or environmental disturbances; Third, quality inspection is usually carried out at the finished product stage, and tracing the root cause of defects requires cross-departmental manual investigation, which is time-consuming and prone to missing key process node anomalies.

[0003] In addition, some enterprises have tried to introduce MES (Manufacturing Execution System) systems to manage production data, but the following core problems have not been solved: (1) The data protocols of multi-source heterogeneous devices are incompatible, and the utilization rate of real-time data is low; (2) There is a lack of a process knowledge base based on material characteristics, and parameter adjustment relies on the trial-and-error method; (3) There is a lack of a dynamic balance strategy when the energy consumption optimization and quality control goals conflict; (4) Paper records and decentralized storage result in insufficient credibility of quality traceability.

[0004] With the development of industry, technologies such as digital twin and edge computing provide new ideas for the above problems, but existing applications have limitations: General digital twin platforms do not model the crystallization kinetics of PE materials, and the virtual simulation accuracy is insufficient; Most AI algorithms are trained with static data sets and do not realize online self-optimization of process parameters; Blockchain technology is mostly used for logistics traceability and is not deeply coupled with the production process. Therefore, there is an urgent need for a full-process digital solution that deeply integrates material science, the Internet of Things, and artificial intelligence to fundamentally improve the intelligent level of PE pipe manufacturing. Summary of the Invention

[0005] In view of the above disadvantages of the prior art, the purpose of the present invention is to provide a digital management system and method for the whole process of intelligent PE pipe manufacturing, which is used to solve the problems of data islands, quality fluctuations, energy consumption waste, and difficult quality traceability in PE pipe manufacturing. The present invention eliminates information islands through the penetration of data throughout the whole process, constructs a digital twin model to predict quality defects, dynamically optimizes process parameters based on the LSTM (Long Short-Term Memory) neural network, traces the root cause of defects by combining causal reasoning, solidifies production data using blockchain, and realizes the coordinated optimization of quality and energy efficiency through reinforcement learning to close-loop regulate the operation of equipment.

[0006] The present invention provides a digital management system for the whole process of intelligent PE pipe manufacturing, including: A data acquisition module, which acquires various data of raw materials through sensors deployed on detection equipment, processes them through an edge computing node to generate a real-time process data stream, and forms a first signal; A digital twin module, which receives the first signal and constructs a digital twin model, fuses material characteristic parameters and environmental data to simulate the pipe crystallization process, and forms a second signal; An analysis and optimization module, which receives the simulation data of the second signal, analyzes the correlation between process parameters and pipe performance through a neural network model, generates temperature compensation and draw speed correction instructions, and forms a third signal; A detection module, which receives the defect image and wall thickness data of the detection instrument, correlates the defect with process anomalies and traces back to the dryness of the raw materials, generates a quality traceability report, forms a fourth signal and transmits it to the digital twin module to update the digital twin model; An integration module, which integrates the optimization instructions of the third signal and the quality constraints of the fourth signal, dynamically adjusts the cooling water pump frequency and draw power through reinforcement learning, generates equipment control instructions and issues them to the production line.

[0007] In the digital management system for the whole process of intelligent PE pipe manufacturing of the present invention, the data acquisition module includes a vibration sensor and an infrared spectrometer. The vibration sensor is embedded in the screw bearing seat of the extruder to monitor the axial vibration spectrum, and the infrared spectrometer is installed at the entrance of the raw material conveyor belt for real-time analysis of the molecular chain branching degree of polyethylene particles. The edge computing node performs wavelet noise reduction processing on the vibration spectrum data, and matches the molecular chain branching degree with the preset raw material quality standard. When it detects that the branching degree deviates from the threshold range, it triggers a warning signal. The real-time process data stream includes vibration feature vectors, branching degree indices, and warning status identifiers. When the first signal is transmitted to the digital twin module through the industrial Ethernet protocol, it is appended with a data integrity check code.

[0008] In the digital management system for the whole process of intelligent PE pipe manufacturing of the present invention, in the digital twin module, the material property parameters obtain the historical batch data of the melt index corresponding to the polyethylene grade through the interface docking with the enterprise MES (Manufacturing Execution System). The environmental data includes the dew point temperature of the workshop air collected by the humidity sensor array and the real-time temperature rise gradient of the cooling water circulation pipeline. When simulating the pipe crystallization process, the unsteady-state thermodynamics equation is used to calculate the crystallization rate, and a virtual temperature compensation node is automatically inserted when it is detected that the cooling water temperature rise exceeds the preset rate. The second signal includes the crystallinity distribution cloud map, the predicted ring stiffness value, and the compensation node coordinate data. The digital twin model receives the device status feedback from the integration module through the OPC UA (Open Platform Communications Unified Architecture) protocol to synchronously update the virtual production line topology structure.

[0009] In the digital management system for the whole process of intelligent PE pipe manufacturing of the present invention, the neural network model of the analysis and optimization module adopts a bidirectional LSTM architecture (Long Short-Term Memory Network). The input layer receives the melt temperature time series data, the screw speed fluctuation curve, and the environmental humidity change gradient output by the digital twin module. The output layer generates the temperature compensation value for the third heating zone of the extruder and the dynamic correction coefficient of the traction speed. The training data set of the neural network model includes the mapping relationship between the pipe burst pressure test results and the corresponding process parameters in the historical production process. The temperature compensation instruction includes a gradient temperature rise strategy based on the prediction of the melt flow front. The traction speed correction coefficient is processed by sliding window mean filtering according to the real-time measured value of the pipe outer diameter. The third signal is synchronized with the device control cycle received by the integration module through the timestamp alignment mechanism.

[0010] In the digital management system for the whole process of intelligent PE pipe manufacturing of the present invention, the defect correlation process of the detection module includes: performing multi-scale convolutional feature extraction on the surface defect image to identify the morphological features of bubbles, converting the wall thickness laser scanning data into an axial distribution histogram and detecting the thickness mutation points. When the bubble diameter exceeds one-fifth of the pipe wall thickness or the thickness mutation gradient reaches the critical value of the material yield strength, the cross-linkage traceability process is started. The traceability process uses the process parameter historical curve recorded by matching the abnormal detection timestamp with the digital twin module. The hot air temperature fluctuation event in the raw material drying stage or the sudden drop in the die head pressure of the extruder is located through the process parameter historical curve. The quality traceability report includes the defect cause probability distribution matrix and the recommended process parameter calibration scheme. When the fourth signal is transmitted to the digital twin module, it triggers the model re-calibration service.

[0011] In the digital management system for the entire process of intelligent PE pipe manufacturing of the present invention, the reinforcement learning algorithm of the integration module adopts a deep deterministic policy gradient framework. The state space is defined as the temperature difference between the inlet and outlet of the cooling water, the amplitude of the pressure oscillation in the vacuum sizing box, and the harmonic distortion rate of the traction motor current. The action space is the adjustment step of the output frequency of the frequency converter and the proportional coefficient of the PID (Proportional-Integral-Derivative controller) control loop. The reward function comprehensively considers the reduction amount of the pipe ovality deviation, the reduction rate of unit energy consumption, and the penalty term for the number of equipment starts and stops. After the equipment control instruction is generated, it is verified through the virtual twin simulation. When the predicted execution result causes the pipe pressure resistance level to drop by more than the safety threshold, a multi-objective Pareto front search is started to generate an alternative control strategy, and an execution priority label and a timeout rollback mechanism are added when the instruction is issued.

[0012] In the digital management system for the entire process of intelligent PE pipe manufacturing of the present invention, the model correction mechanism of the digital twin module includes: when the relative error between the predicted value of the virtual pipe ring stiffness and the actual pressure test result feedback by the detection module continuously exceeds 8% for three cycles, the iterative optimization of the melt flow rate compensation coefficient is started. The particle swarm algorithm is used to search for the optimal compensation vector in the material characteristic parameter space, and the hidden layer weight matrix of the LSTM neural network in the analysis and optimization module is synchronously updated. The corrected digital twin model recalculates the key quality indicators of the production batches in the past two hours and generates a list of process parameter backtracking adjustment suggestions and pushes them to the MES system dashboard.

[0013] In the digital management system for the entire process of intelligent PE pipe manufacturing of the present invention, the analysis and optimization module further integrates a multi-objective optimization function. The function takes the maximization of the environmental stress cracking resistance index of the pipe and the minimization of the unit output energy consumption as parallel optimization objectives. The constraint conditions include the safety threshold of the extruder torque, the limit value of the heat exchange efficiency of the cooling tower, and the avoidance interval of the mechanical resonance frequency of the traction system. The NSGA-II (Non-dominated Sorting Genetic) algorithm is used to generate the Pareto optimal solution set, and the process parameter combination that takes into account both quality and energy efficiency is selected through fuzzy comprehensive evaluation. The temperature compensation instruction includes a feedforward control component based on the prediction of the extrusion rate change, and the traction speed correction coefficient calculates the dynamic compensation amount in real time according to the axial shrinkage rate of the pipe.

[0014] In the digital management system for the entire process of intelligent PE pipe manufacturing of the present invention, the blockchain evidence storage process of the detection module includes: structuring the raw material supplier code, process parameter version number, and grayscale image of the detection result in the quality traceability report through a Merkle tree (hash tree), generating a data fingerprint and writing it into the permissioned chain node. The electronic quality certificate includes a verifiable digital signature and a Beidou time stamp. When the pipe enters the logistics link, the full life cycle data on the blockchain is associated through the invisible two-dimensional code sprayed on the pipe wall. The two-dimensional code is printed with a ceramic substrate ink that resists high and low temperature deformation. The code scanning device can verify the integrity of the production data and the tampering traces layer by layer by parsing the hash pointer.

[0015] The present invention also provides a digital management method for the whole process of intelligent PE pipe manufacturing, including: S1: Collect various data of raw materials through sensors deployed on detection equipment, generate real-time process data streams through edge computing node processing, and form a first signal; S2: According to the first signal and construct a digital twin model, fuse material characteristic parameters and environmental data to simulate the pipe crystallization process, and form a second signal; S3: Based on the second signal and analyze the correlation between process parameters and pipe performance through a neural network model, generate temperature compensation and traction speed correction instructions, and form a third signal; S4: Receive defect images and wall thickness data of the detection instrument, associate defects with process anomalies and trace back to the dryness of raw materials, generate a quality traceability report, form a fourth signal and update the digital twin model; S5: Integrate the optimization instructions of the third signal and the quality constraints of the fourth signal, dynamically adjust the cooling water pump frequency and traction power through reinforcement learning, generate equipment control instructions and send them to the production line.

[0016] An intelligent PE pipe manufacturing whole-process digital management system and method provided by the present invention eliminate information islands through full-process data penetration, construct a digital twin model to predict quality defects, dynamically optimize process parameters based on the LSTM neural network, trace the root cause of defects through causal reasoning, solidify production data using blockchain, and achieve coordinated optimization of quality and energy efficiency through closed-loop regulation of equipment operation by reinforcement learning. Brief Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is the system architecture diagram of the intelligent PE pipe manufacturing whole-process digital management system; Figure 2 It is the method flow diagram of the intelligent PE pipe manufacturing whole-process digital management method. Detailed Embodiments

[0019] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand the other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0020] It should be noted that the diagrams provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner. Therefore, only the components related to the present invention are shown in the diagrams, rather than being drawn according to the number, shape, and size of the components in actual implementation. The type, quantity, and ratio of each component in actual implementation can be arbitrarily changed, and the component layout type may also be more complex.

[0021] In the following description, a large number of details are explored to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0022] Please refer to Figure 1-2 , which shows the digital management system and method for the whole process of intelligent PE pipe manufacturing of the present invention. The digital management system for the whole process of intelligent PE pipe manufacturing of the present invention includes a data acquisition module, a digital twin module, an analysis and optimization module, a detection module, and an integration module. The data acquisition module collects various data of raw materials through sensors deployed on detection equipment, processes them through edge computing nodes to generate real-time process data streams, and forms a first signal. The digital twin module receives the first signal and constructs a digital twin model, simulates the pipe crystallization process by integrating material characteristic parameters and environmental data, and forms a second signal. The analysis and optimization module receives the simulation data of the second signal, analyzes the correlation between process parameters and pipe performance through a neural network model, generates temperature compensation and draw speed correction instructions, and forms a third signal. The detection module receives the defect images and wall thickness data of the detection instrument, correlates the defects with process abnormalities and traces them back to the dryness of the raw materials, generates a quality traceability report, forms a fourth signal, and transmits it to the digital twin module to update the digital twin model. The integration module integrates the optimization instructions of the third signal and the quality constraints of the fourth signal, dynamically adjusts the cooling water pump frequency and draw power through reinforcement learning, generates equipment control instructions, and issues them to the production line.

[0023] As Figure 1As shown in the figure, the present invention relates to a digital management system for the entire process of intelligent PE pipe manufacturing. The constructed digital management system for the entire process of intelligent PE pipe manufacturing has a technical architecture based on a multi-module collaboration mechanism to achieve closed-loop optimization of the manufacturing process. The data acquisition module, as the core of the perception layer of the system, covers all production line nodes from raw material processing to finished product inspection through an embedded sensor network: a three-axis vibration sensor is installed on the screw bearing seat of the extruder to capture the axial vibration spectrum to indirectly reflect the melt flow uniformity; a near-infrared spectrometer is deployed at the entrance of the raw material conveyor belt to analyze the molecular chain branching degree of polyethylene particles through the absorption rate of characteristic wavelengths. This parameter directly affects the environmental stress cracking resistance of the pipe. The edge computing node adopts a multi-core heterogeneous processor architecture to perform wavelet packet transform noise reduction processing on the vibration signal, extract the low-frequency harmonic components related to the screw wear state, and at the same time dynamically match the spectral data with the standard branching degree curve in the raw material database. When it is detected that the branching degree deviates from the qualified range, an early warning signal containing the raw material batch number, deviation magnitude, and recommended blending ratio is generated. When the real-time process data stream is encapsulated as the first signal through the OPC UA protocol, a data integrity check code based on the SHA-256 algorithm is added to ensure that the process of transmitting to the digital twin module is protected from man-in-the-middle attacks. After receiving the first signal, the digital twin module starts a multi-physics field coupling simulation engine: first, it retrieves the historical melt index data corresponding to the polyethylene grade of the current production batch from the enterprise MES system, combines the dew point temperature of the workshop collected by the humidity sensor array and the temperature rise gradient of the cooling water circulation pipeline, constructs a non-steady-state thermodynamics model to calculate the pipe crystallization kinetics process, where a modified Avrami equation (a mathematical model describing the relationship between crystal growth rate and time in the crystallization kinetics process) is introduced to describe the change rate of crystallinity with time, and virtual temperature compensation nodes are dynamically inserted based on the cooling water temperature rise rate to offset the thermal lag effect. The second signal output by the simulation includes a three-dimensional crystallinity distribution cloud map, predicted ring stiffness value, and compensation node coordinate data. These information are simultaneously pushed to the input end of the AI model of the analysis and optimization module and the defect analysis unit of the detection module through the message queue service. The analysis and optimization module uses a bidirectional LSTM neural network architecture to process the time-series process data: the input layer receives the melt temperature fluctuation curve, screw speed deviation sequence, and environmental humidity change gradient output by the digital twin module, and captures the non-linear coupling relationship between parameters through the gating mechanism; the hidden layer contains 256 neurons, and Dropout (a technique to prevent neural network overfitting) regularization is used to prevent overfitting. The training data set integrates the mapping relationship between the pipe burst pressure test results and corresponding process parameters in historical production; the output layer generates the temperature compensation value for the third heating zone of the extruder and the dynamic correction coefficient of the traction speed. Among them, the temperature compensation instruction contains a gradient heating-up strategy based on the prediction of the melt flow front, and the traction speed correction coefficient is processed by sliding window mean filtering according to the real-time measured value of the pipe outer diameter feedback by the laser diameter gauge to ensure a smooth transition of the adjustment instruction.The third signal is synchronized with the device control cycle of the integration module through the timestamp alignment mechanism to avoid instruction execution conflicts. The defect correlation engine of the detection module adopts multi-modal data analysis: extracting multi-scale convolutional features from the surface defect image to identify the morphological features of bubbles (such as aspect ratio, edge sharpness), converting the wall thickness laser scanning data into an axial distribution histogram and detecting thickness mutation points. When the bubble diameter exceeds one-fifth of the wall thickness or the thickness mutation gradient reaches the critical value of the material yield strength, a cross-link traceability process is initiated. This process locates the hot air temperature fluctuation event in the raw material drying stage or the sudden drop in the die head pressure of the extruder by matching the abnormal detection timestamp with the historical process parameter curve recorded by the digital twin module. The generated quality traceability report includes a defect cause probability distribution matrix and a recommended process parameter calibration scheme. When the fourth signal is transmitted to the digital twin module, it triggers the model re-calibration service to synchronously update the virtual production line topology. The integration module adopts the Deep Deterministic Policy Gradient (DDPG) reinforcement learning framework, defines the state space as the temperature difference between the inlet and outlet of the cooling water, the pressure oscillation amplitude of the vacuum sizing tank, and the harmonic distortion rate of the traction motor current, and the action space as the adjustment step of the inverter output frequency and the proportional coefficient of the PID control loop. The reward function comprehensively considers the reduction in the pipe ovality deviation, the unit energy consumption reduction rate, and the penalty term for the number of equipment start-stop times. After the device control instruction is generated, it is simulated and verified through the virtual twin. When the predicted execution result causes the pipe pressure resistance level to drop by more than the safety threshold, a multi-objective Pareto front search is initiated to generate an alternative control strategy. An execution priority label and a timeout rollback mechanism are attached when the instruction is issued, forming a closed-loop control link from parameter optimization to execution verification.

[0024] Furthermore, the model construction and calibration mechanism of the digital twin module is refined. The material property parameters are dynamically obtained by docking with the standardized API interface of the enterprise MES system to obtain the historical batch data of the melt index of polyethylene raw materials, and a melt flow rate prediction model based on Gaussian process regression is established. The environmental data acquisition network consists of a humidity sensor array deployed distributively and temperature probes on the cooling water circulation pipeline. Among them, the humidity sensor measures the dew point temperature of the workshop air using the capacitive principle, with an accuracy of ±1.5%RH. The cooling water temperature rise gradient is obtained by differential calculation of the temperature change rate at the inlet and outlet. When simulating the pipe crystallization process, the digital twin module uses an improved unsteady thermodynamic equation to decompose the crystallization rate into a temperature-dependent term and a stress-induced term: the temperature-dependent term is calculated based on the Arrhenius formula (an empirical formula describing the relationship between the chemical reaction rate constant and temperature), considering the local heat transfer coefficient fluctuation caused by the change in cooling water flow; the stress-induced term introduces a shear rate correction factor generated by screw extrusion to reflect the influence of molecular chain orientation on the crystallinity distribution. When it is detected that the cooling water temperature rise rate exceeds the preset threshold, the system automatically inserts a temperature compensation node in the virtual model, and the coordinates of this node are dynamically adjusted according to the pipe surface temperature field collected by the thermal imager. The crystallinity distribution cloud map in the second signal uses a voxelized expression method, and each voxel contains the crystallinity percentage, the predicted grain size value, and the estimated residual stress value. The predicted ring stiffness value is obtained by calculating the deformation of the virtual pipe under the radial load through finite element analysis. The digital twin model receives the actual operating status of the equipment (such as the real-time value of the traction motor torque and the working frequency of the vacuum pump) fed back by the integration module through the OPC UA protocol, and dynamically updates the device node parameters in the virtual production line topology to ensure that the state synchronization accuracy between the simulation model and the physical device is controlled at the millisecond level. When the relative error between the predicted value of the virtual pipe ring stiffness and the actual pressure test result fed back by the detection module continuously exceeds 8% for three production cycles, the model self-calibration process is triggered: first, the particle swarm optimization algorithm is used to search for the optimal compensation vector in the material property parameter space to adjust the melt flow rate compensation coefficient; then, the calibrated parameters are synchronized to the LSTM neural network of the analysis and optimization module to update its hidden layer weight matrix to adapt to the new process conditions; finally, the key quality indicators (such as the average burst pressure and the standard deviation of roundness) of the production batches in the past two hours are recalculated, and a list of process parameter backtracking adjustment suggestions is generated and pushed to the MES system dashboard through the RESTful interface for the operator's decision-making reference.

[0025] In an embodiment of the present invention, the neural network model adopts a bidirectional LSTM architecture, and the input layer is designed as a multi-channel time series processing structure: the first channel receives the melt temperature time series data output by the digital twin module, and extracts the statistical features of temperature fluctuations (such as mean, variance, autocorrelation function) through a sliding window; the second channel processes the screw speed fluctuation curve and uses the fast Fourier transform to identify the harmonic components related to the natural frequency of the mechanical transmission system; the third channel analyzes the environmental humidity change gradient and calculates its dynamic influence factor on the cooling crystallization rate. The output layer uses a fully connected network to generate process parameter adjustment instructions: the temperature compensation value is calculated according to the melt flow front prediction model, and a gradient heating strategy is implemented in the third heating zone of the extruder to balance the melt viscosity distribution; the dynamic correction coefficient of the traction speed is combined with the real-time measured value of the pipe outer diameter and smoothed by the exponential weighted moving average method to avoid pipe stretching resonance caused by speed mutation. When constructing the training data set, virtual samples under process parameter perturbation scenarios are generated through data augmentation technology to improve the generalization ability of the model under abnormal working conditions. The multi-objective optimization function takes the maximization of the environmental stress cracking index of the pipe and the minimization of the energy consumption per unit output as parallel objectives, and the constraint conditions include the safety threshold of the extruder torque, the limit value of the heat exchange efficiency of the cooling tower, and the avoidance interval of the mechanical resonance frequency of the traction system. The improved NSGA-II algorithm is used for multi-objective optimization: first, the Pareto front solution set is screened through non-dominated sorting, then the crowding comparison operator is used to maintain the diversity of the solution set, and finally the fuzzy comprehensive evaluation method is applied to select the optimal process parameter combination that takes into account both quality and energy efficiency. The temperature compensation instruction contains a feedforward control component, which is dynamically adjusted based on the predicted value of the extrusion rate change, and the heating power is compensated in advance when the raw material feeding speed fluctuation is detected; the traction speed correction coefficient calculates the dynamic compensation amount in real time according to the axial shrinkage rate of the pipe, and reversely corrects the linear speed setting value of the traction roller through the deviation between the actual length and the theoretical value of the pipe fed back by the laser length measuring instrument. The third signal transmission protocol adopts the IEEE1588 standard with time synchronization function to ensure the phase alignment of the optimization instruction and the device control cycle, and avoid the asynchronous execution of the instruction caused by network delay. In addition, the module integrates an online learning function. When it is detected that the actual production quality index continuously deviates from the predicted value, the model incremental training process is automatically triggered, and the neural network weights are updated using real-time production data to achieve the continuous evolution of the process optimization ability.

[0026] Such as Figure 1As shown, the defect correlation engine adopts multi-modal data fusion technology to process the surface defect images through a multi-scale convolutional neural network: the first layer of convolutional kernels extracts the bubble edge gradient features, the second layer uses dilated convolution to expand the receptive field to identify the defect distribution pattern, and the third layer applies an attention mechanism to focus on the high stress concentration areas. After the wall thickness laser scanning data is denoised by Kalman filtering, it is converted into an axial distribution histogram and the local thickness variation coefficient is calculated. When the variation coefficients of three consecutive measurement points are detected to exceed the corresponding threshold of the material's Poisson ratio, a thickness anomaly mark is triggered. The defect traceability process establishes a cross-level causal graph model, aligns the detected bubble morphological features (such as sphericity, estimated internal pressure value) with the historical sequence of process parameters stored in the digital twin module in terms of time, and calculates the posterior probability of process node anomalies through a Bayesian network to locate the hot air temperature fluctuation event of the raw material dryer or the sudden drop in the die head pressure of the extruder. The quality traceability report generation module uses a decision tree rule engine to match the predefined process calibration scheme library according to the defect cause probability distribution matrix. For example, when the analysis of the bubble internal pressure shows that the gas source is water vaporization in the raw material, it recommends extending the drying time and increasing the hot air temperature setting value by 5-10°C. When the fourth signal is transmitted to the digital twin module, it triggers the model re-calibration service: first, freeze the current simulation process, perform residual analysis on the actual detection results and the virtual pipe mechanical property data. If the crystallinity prediction error exceeds the safety threshold, start the online identification algorithm for material characteristic parameters, update the polyethylene melt index compensation coefficient, and synchronously adjust the forgetting gate weight parameters of the LSTM neural network in the analysis and optimization module. The blockchain evidence storage subsystem adopts a hierarchical architecture design: at the bottom layer, the raw material supplier code, process parameter version number, and the grayscale image of the detection results in the quality traceability report are processed into a Merkle tree structure, and after generating a data fingerprint, it is written into the permissioned chain node; in the middle layer, access control of data is realized through smart contracts to ensure that different roles such as manufacturers, quality inspection institutions, and customers obtain differentiated data views; the electronic quality certificate at the application layer integrates a verifiable digital signature and a Beidou time stamp. When the pipe enters the logistics link, the full life cycle data on the blockchain is associated through the invisible two-dimensional code laser-engraved on the pipe wall. This two-dimensional code is printed with weather-resistant ceramic substrate ink and can withstand temperature cycle shocks from -40°C to 120°C. The scanning device can verify the integrity of the production data layer by layer through parsing the hash pointer and detect whether there are traces of tampering.

[0027] Specifically, the Actor-Critic network in the Deep Deterministic Policy Gradient (DDPG) framework (a reinforcement learning method that combines policy optimization and value function estimation) adopts a dual-delay update mechanism: the input state space of the Actor network includes the third derivative of the temperature difference between the inlet and outlet of the cooling water, the spectral entropy value of the pressure oscillation signal of the vacuum sizing box, and the sliding window variance of the harmonic distortion rate of the traction motor current; the Critic network additionally receives the historical trajectory data of the action space to evaluate the long-term value function. The action space is designed as a multi-dimensional continuous control vector, including the frequency adjustment step of the frequency converter output (resolution 0.1 Hz), the logarithmic change of the proportional coefficient of the PID control loop, and the start-stop duty cycle of the vacuum pump. The reward function is constructed by weighted synthesis of three dimensions using the Analytic Hierarchy Process: the quality dimension weight accounts for 60% (including the reduction of the pipe ovality deviation and the improvement of the wall thickness uniformity), the energy efficiency dimension accounts for 30% (including the reduction rate of the energy consumption per unit output and the recycling rate of the cooling water), and the equipment maintenance dimension accounts for 10% (including the penalty term for the number of motor starts and stops and the bearing vibration severity warning coefficient). The Monte Carlo Tree Search algorithm is introduced in the virtual twin verification link to perform multi-step look-ahead simulation on the control instructions to be executed. When the predicted reduction of the pipe pressure resistance level exceeds the material safety factor, a multi-objective Pareto front search is started: first, the quality index is set as a hard constraint by the ε-constraint method, and the simulated annealing algorithm is used to find the optimal energy efficiency solution in the feasible solution space; if there is no feasible solution, it switches to the compromise decision-making mode, and the compromise coefficients of each objective are calculated using the fuzzy membership function to generate an alternative control strategy. The instruction issuing system adopts a priority label management mechanism, classifying the control instructions into three levels: emergency correction (such as the melt temperature exceeding the limit), optimization adjustment (such as energy efficiency improvement), and preventive maintenance (such as vibration warning), and ensuring the timeliness of instruction execution through a timeout rollback mechanism - if a certain instruction does not receive feedback on the device status within the set time, the instruction is automatically revoked and the fault diagnosis process is triggered.

[0028] Furthermore, when the relative error between the predicted value of the virtual pipe stiffness and the actual pressure test results exceeds the threshold for three consecutive cycles, a three-level correction process is initiated: for the primary correction, a least squares support vector machine (LS-SVM) is used to establish an error compensation model to adjust the melt flow rate compensation coefficient online; for the intermediate correction, the particle swarm optimization algorithm is used to search for the optimal solution in the material property parameter space, and the crystallization activation energy parameter in the non-steady state thermodynamics equation is updated synchronously; for the advanced correction, a full model reconstruction is triggered, and the multi-physics field coupling simulation engine of the digital twin is retrained using real-time production data. During the correction process, the system establishes a version control log to record the amplitude, timestamp, and trigger conditions of each parameter adjustment, forming a traceable model evolution path. The updated LSTM neural network weight matrix maintains parameter orthogonality through the orthogonal initialization method to prevent the problem of gradient disappearance. At the same time, the knowledge distillation technology is used to transfer the knowledge of the corrected model to the edge computing node to achieve distributed intelligent optimization. The process parameter backtracking adjustment recommendation list generation module applies the association rule mining algorithm to analyze the frequent itemsets in the historical adjustment records. For example, when the vacuum sizing pressure decreases and the traction speed increases simultaneously, it is recommended to synchronously adjust the cooling water flow rate to prevent the pipe ovality from exceeding the standard. When these recommendations are pushed to the MES system dashboard through the RESTful API (an architecture style based on the HTTP protocol), they are accompanied by a confidence rating (a Bayesian estimate based on the historical adjustment success rate) and an expected benefit prediction (calculating the quality improvement range and energy consumption savings using a random forest regression model).

[0029] As Figure 1As shown, the improvement of the NSGA-II algorithm is reflected in three aspects: First, the reference point-guided non-dominated sorting is introduced, taking the industry standard values of the environmental stress cracking index of the pipe material and the energy consumption per unit output as reference points to accelerate the convergence of the Pareto front; Second, the process constraint knowledge is embedded in the crossover operator to ensure that the generated solution vectors do not exceed physical limits such as the safety threshold of the extruder torque and the limit value of the heat exchange efficiency of the cooling tower; Finally, the surrogate-assisted selection based on the Kriging model is adopted to reduce the number of evaluations of the true objective function. The construction of the fuzzy comprehensive evaluation system includes five dimensions: quality stability (weight 40%), energy efficiency (30%), equipment loss rate (15%), continuity of production rhythm (10%), and environmental protection indicators (5%). The membership functions of each dimension are designed with a trapezoidal distribution. The feedforward control component in the temperature compensation instruction is dynamically generated by the extrusion rate change prediction model: an ARIMA time series model is established to predict the raw material feeding fluctuation trend within the next 5 minutes, and the required heating power compensation amount is calculated in advance. This compensation amount is fused with the feedback control amount in the frequency domain and the low-frequency and high-frequency components are separated by a Butterworth filter for separate processing. The traction speed dynamic compensation algorithm introduces an axial shrinkage rate real-time observer: based on the deviation between the actual length and the theoretical value of the pipe material collected by the laser length measuring instrument, combined with the material creep characteristic model, a Luenberger observer is constructed to estimate the real-time shrinkage rate, and then the compensation coefficient of the linear speed of the traction roller is deduced. This coefficient is limited to prevent speed overshoot.

[0030] Furthermore, the Merkle tree construction uses the SHA-3 algorithm to generate the hash values of the leaf nodes. Each leaf node corresponds to the following data elements: raw material batch number, drying process parameter set, set values of each temperature zone of the extruder, cooling water circulation parameters, and fingerprint of the detection result image. The permission chain node deployment adopts a hybrid architecture: the production end nodes are deployed on the enterprise internal server, the quality inspection agency nodes are deployed in the cloud trusted execution environment (TEE), and the client nodes adopt lightweight blockchain clients. The zero-knowledge proof mechanism is implemented in the data uploading process: before writing sensitive process parameters (such as the enterprise-specific temperature curve) into the blockchain, they are first converted into homomorphic encrypted ciphertext to ensure that the data on the chain is available but invisible. The invisible QR code encoding scheme uses the Reed-Solomon error correction code and information dispersal storage technology: the complete hash value is divided into multiple segments and stored in QR codes at different physical locations. When a single QR code is damaged, the complete information can still be restored through the remaining segments. The QR code scanning verification system integrates multiple verification mechanisms: First, the data timeliness is verified through the Beidou time stamp (to prevent replay attacks), second, the authenticity of the digital signature is verified using the national cryptography SM2 algorithm, and finally, the hash value of the grayscale image stored on the chain is compared with the real-time calculation result of the actual pipe material detection image to ensure data integrity.

[0031] As Figure 1As shown, when a large - area failure of sensors or a communication interruption is detected, the degraded operation mode is activated: the data acquisition module switches to the virtual sensing state based on the mechanism model, estimates the melt pressure using the extruder current signal, and derives the estimated value of the cooling water flow through the energy conservation equation; the digital twin module enables the simulation prediction using the historical optimal parameter set cached offline; the analysis and optimization module switches to the expert system based on case - based reasoning, and retrieves the process parameter combinations under similar working conditions from the historical successful case library; the integration module adopts the preset safety control strategy to limit the equipment operation parameters within a conservative range. The fault self - healing system locates the fault source through multi - dimensional feature analysis: conducts wavelet packet energy spectrum analysis on the sensor signals to distinguish hardware faults from process anomalies, identifies the status of the communication module through the industrial bus diagnostic message, and uses redundant channels to achieve multi - path transmission of key data. During the system recovery stage, a progressive restart strategy is implemented: first, the basic sensing function of the data acquisition module is restored, second, a simplified version of the digital twin model is gradually loaded, and finally, the intelligent optimization algorithm is activated in stages to ensure a smooth transition of the production line to the normal state. All data generated during the fault - tolerance process are recorded in a dedicated slice of the blockchain, forming an immutable fault - handling knowledge base, providing a traceability basis for subsequent system optimization.

[0032] As Figure 2 shown, it is the digital management method for the whole process of intelligent PE pipe manufacturing of the present invention. It includes S1: Collect various data of raw materials through sensors deployed on detection equipment, generate real - time process data streams through edge - computing nodes, and form the first signal; S2: According to the first signal and construct a digital twin model, fuse material characteristic parameters and environmental data to simulate the pipe crystallization process, and form the second signal; S3: Based on the second signal and analyze the correlation between process parameters and pipe properties through a neural network model, generate temperature compensation and drawing speed correction instructions, and form the third signal; S4: Receive the defect images and wall - thickness data of the detection instrument, associate defects with process anomalies and trace them back to the raw material dryness, generate a quality traceability report, form the fourth signal and update the digital twin model; S5: Integrate the optimization instructions of the third signal and the quality constraints of the fourth signal, dynamically adjust the cooling water pump frequency and drawing power through reinforcement learning, generate equipment control instructions and send them to the production line.

[0033] As Figure 2As shown, in step S1, the system uses a multi-dimensional sensor network to collect data throughout the entire process. During raw material pretreatment, a vibration sensor embedded in the dryer bearing housing monitors rotor imbalance. An infrared spectrometer scans the near-infrared absorption spectrum of polyethylene granules in real time. The absorption peak intensity in the characteristic wavelength range of 1650-1850nm reflects the degree of molecular chain branching. In the extrusion section, a melt pressure sensor is installed in the fifth heating zone of the barrel, and a fiber-optic temperature probe monitors the melt temperature gradient with a resolution of 0.1°C. A laser diameter gauge is deployed in conjunction with an encoder at the traction and cutting station, capturing pipe outer diameter fluctuations at a sampling rate of 200Hz. Edge computing nodes synchronize the clocks of each sensor using the Time-Sensitive Networking (TSN) protocol and preprocess the raw data. Wavelet packet decomposition is performed on the vibration signal to extract characteristic energy values in the 1-5kHz frequency band, reflecting the state of screw wear. Dynamic Time Warping (DTW) is performed on the infrared spectrum data to match the standard curve in the raw material database to calculate the branching deviation index. A Kalman filter is applied to the melt pressure data to eliminate mechanical vibration noise from the equipment. The processed data stream is encapsulated as the first signal through the OPC UA over TSN protocol, and each data packet is attached with a digital signature based on the elliptic curve cryptography system to ensure the integrity and tamper-resistance of the data transmitted to the digital twin module.

[0034] The digital twin model construction process in step S2 embodies multi-physics coupling. First, the polyethylene grade parameters for the current production order, including melt index history, density fluctuation range, and additive ratio, are retrieved from the company's MES system. Data on workshop air dew point temperature, cooling tower inlet and outlet water temperature difference, and circulating water flow rate, provided by the environmental monitoring unit, are also simultaneously accessed. The core model utilizes a modified crystallization kinetics equation, decomposing the pipe crystallization process into two stages: nucleation and growth. The nucleation rate calculation incorporates the Ziabicki non-isothermal crystallization theory, accounting for the influence of cooling rate on critical nucleus size. The crystal growth rate model incorporates the Hoffman-Lauritzen theory (the non-isothermal crystallization behavior of polylactic acid) and incorporates a screw shear rate correction factor to account for molecular chain orientation effects. The simulation engine updates the three-dimensional temperature field distribution every 50 milliseconds. The crystallinity gradient across the pipe cross section is calculated using the finite volume method. Ring stiffness prediction utilizes the Voigt model to superimpose the contribution of anisotropic crystallization zones. When the cooling water temperature rise rate exceeds 3°C / min, a dynamic compensation node is automatically inserted into the virtual model. The node coordinates are adaptively adjusted based on the pipe surface temperature field captured by the infrared thermal imager. The generated second signal contains the crystallinity distribution matrix, predicted bursting pressure value and compensation node topology data, which are pushed to the message queue of the analysis and optimization module and the cache database of the detection module through the MQTT protocol.

[0035] The process parameter optimization process of step S3 is realized relying on deep learning methods: the input layer of the bidirectional LSTM neural network is designed as a three-channel structure - the first channel receives the melt temperature time series data output by the digital twin module, and extracts the trend term, periodic term, and residual term features through a sliding window; the second channel processes the screw speed fluctuation curve and uses the fast Fourier transform to identify the harmonic components that coincide with the natural frequency of the gearbox; the third channel analyzes the influence of environmental humidity on the cooling rate and calculates the equivalent crystallization driving factor. The hidden layer adopts a gated recurrent unit (GRU) structure to reduce the parameter scale, and the output layer generates process adjustment instructions through a fully connected network: the temperature compensation value of the third heating zone of the extruder is calculated according to the melt flow front prediction model, and a feedforward-feedback composite control strategy is adopted. The feedforward component adjusts the heating power in advance based on the predicted value of the raw material feeding speed, and the feedback component performs PID correction according to the real-time reading of the melt pressure sensor; the traction speed correction coefficient combines the outer diameter data fed back by the laser diameter gauge, is smoothed using the exponentially weighted moving average method, and ensures the convergence of the speed adjustment process through Lyapunov stability analysis. When constructing the training data set, a generative adversarial network (GAN) is used to expand the sample diversity, generate virtual pipe mechanical property data under different process disturbance scenarios, and improve the generalization ability of the model under abnormal working conditions. The generated third signal is synchronized with the device control cycle through the IEEE 1588 precise time protocol to ensure that the optimization instructions are issued and executed within the specified phase window.

[0036] The defect traceability mechanism in step S4 realizes cross - link causal association: The surface defect detection uses an improved YOLOv5 algorithm, enhances the bubble edge detection ability through the attention mechanism, and uses transfer learning to achieve a recall rate of over 95% with a small number of labeled samples; The wall thickness analysis module converts the laser scanning data into an axial distribution spectrogram and uses variational mode decomposition (VMD) to separate the equipment vibration noise from the real thickness fluctuation signal. When the coefficient of variation of the wall thickness at three consecutive measurement points exceeds the threshold corresponding to the material yield strength, a multi - level traceability process is initiated: First, align the defect timestamp with the process log of the digital twin module and extract the key parameters during the abnormal period (such as sudden drops in melt pressure events, fluctuations in cooling water flow); Second, calculate the abnormal contribution degree of each process node through the causal graph model, and use the Shapley value allocation method to quantify the influence weights of factors such as raw material dryness deviation, extrusion temperature fluctuation, etc.; The finally generated quality traceability report includes a tree - shaped diagram of the probability distribution of defect causes and parameter calibration suggestions. For example, when the internal pressure analysis of the bubble shows that the gas source is excessive moisture content in the raw material, it is recommended to extend the drying time and increase the set value of the hot air temperature. The fourth signal triggers the online correction of the digital twin model: By comparing the predicted value of the virtual pipe burst pressure with the actual hydrostatic test result, the Bayesian optimization algorithm is used to adjust the activation energy parameter in the crystallization kinetics model, and the corrected parameter is synchronized to the neural network weight matrix of the analysis and optimization module. The blockchain evidence - storing subsystem adopts a hierarchical encryption architecture: The raw material batch information, process parameter set, and test result images are encrypted by the national secret SM4 algorithm to generate the hash values of the leaf nodes of the Merkle tree, and are written into the distributed ledger through the PBFT consensus mechanism; The electronic quality certificate integrates the time - service information of Beidou navigation satellites and digital watermark technology to ensure the spatio - temporal credibility of the data.

[0037] The reinforcement learning control strategy in step S5 achieves multi-objective dynamic balance: In the Deep Deterministic Policy Gradient (DDPG) algorithm, the input state space of the Actor network includes 12-dimensional features such as the second derivative of the temperature difference between the inlet and outlet of the cooling water, the sample entropy of the vacuum sizing pressure signal, and the total harmonic distortion rate (THD) of the traction motor current. When the Critic network evaluates the value function, a long-term energy efficiency return discount factor is introduced. The action space is designed as a 7-dimensional continuous vector, including the adjustment amount of the frequency converter output frequency (step size of ±2Hz), the logarithmic change amount of the PID proportional coefficient (range of ±0.3), and the adjustment amplitude of the vacuum pump duty cycle (±15%). The reward function is constructed by synthesizing three dimensions using the Analytic Hierarchy Process: quality indicators (weight 55%, including the reduction amount of ovality deviation and wall thickness uniformity index), energy efficiency indicators (30%, including the reduction rate of power consumption per ton and the utilization rate of cooling water circulation), and equipment health indicators (15%, including bearing vibration intensity and motor start-stop frequency). In the virtual twin verification link, Monte Carlo tree search is used for multi-step look-ahead simulation. When it is predicted that the execution of the instruction causes the pressure resistance level of the pipe to drop by more than the material safety factor, the search for the Pareto optimal solution is started: First, the quality indicator is set as a hard boundary through the ε-constraint method, and the particle swarm optimization is used to find the optimal energy efficiency solution in the feasible solution space; if there is no feasible solution, it switches to the compromise mode, and the fuzzy membership function is applied to calculate the compromise coefficients of each target to generate an alternative strategy. The control instruction issuing system adopts a priority management mechanism, classifying the instructions into three levels: emergency correction (red label), optimization adjustment (yellow label), and preventive maintenance (blue label), and ensuring the execution reliability through the timeout rollback mechanism - any instruction that does not receive a device feedback signal within the set time will be automatically revoked and the fault diagnosis process will be triggered. The execution result data is transmitted back to the data acquisition module in real time, forming a closed-loop control link of "perception - decision - execution - feedback", enabling the system to have the ability of continuous self-optimization.

[0038] The intelligent digital management system and method for the whole process of PE pipe manufacturing of the present invention eliminates information islands through the penetration of data in the whole process, constructs a digital twin model to predict quality defects, dynamically optimizes process parameters based on the LSTM neural network, traces the root cause of defects through causal reasoning, solidifies production data using blockchain, and realizes the collaborative optimization of quality and energy efficiency through the closed-loop regulation of equipment operation by reinforcement learning.

[0039] Therefore, through the intelligent digital management system and method for the whole process of PE pipe manufacturing of the present invention, the problems of data islands, quality fluctuations, energy consumption waste, and difficult quality traceability in PE pipe manufacturing are solved.

[0040] The above embodiments are only illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical idea disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. Digital management system for the whole process of intelligent PE pipe manufacturing, characterized in that, Including: A data acquisition module that collects various data of raw materials through sensors deployed on detection equipment, processes them through an edge computing node to generate a real-time process data stream, and forms a first signal; A digital twin module that receives the first signal and constructs a digital twin model, fuses material characteristic parameters and environmental data to simulate the pipe crystallization process, and forms a second signal; An analysis and optimization module that receives the simulation data of the second signal, analyzes the correlation between process parameters and pipe performance through a neural network model, generates temperature compensation and draw speed correction instructions, and forms a third signal; A detection module that receives the defect image and wall thickness data of the detection instrument, correlates defects with process anomalies and traces them back to the dryness of the raw materials, generates a quality traceability report, forms a fourth signal and transmits it to the digital twin module to update the digital twin model; An integration module that integrates the optimization instructions of the third signal and the quality constraints of the fourth signal, dynamically adjusts the cooling water pump frequency and draw power through reinforcement learning, generates equipment control instructions and issues them to the production line.

2. The digital management system for the whole process of intelligent PE pipe manufacturing according to claim 1, characterized in that, The data acquisition module includes a vibration sensor and an infrared spectrometer. The vibration sensor is embedded in the screw bearing seat of the extruder to monitor the axial vibration spectrum. The infrared spectrometer is installed at the entrance of the raw material conveyor belt to analyze the molecular chain branching degree of polyethylene particles in real time. The edge computing node performs wavelet noise reduction processing on the vibration spectrum data, and matches the molecular chain branching degree with the preset raw material quality standard. When it detects that the branching degree deviates from the threshold range, it triggers an early warning signal. The real-time process data stream includes vibration feature vectors, branching degree indices, and early warning status identifiers. When the first signal is transmitted to the digital twin module through the industrial Ethernet protocol, an additional data integrity check code is added.

3. The digital management system for the whole process of intelligent PE pipe manufacturing according to claim 1, characterized in that In the digital twin module, the material characteristic parameters are obtained by docking with the enterprise MES system interface to obtain the historical batch data of the melt index corresponding to the polyethylene grade. The environmental data includes the dew point temperature of the workshop air collected by the humidity sensor array and the real-time temperature rise gradient of the cooling water circulation pipeline. When simulating the pipe crystallization process, the non-steady-state thermodynamics equation is used to calculate the crystallization rate, and a virtual temperature compensation node is automatically inserted when it detects that the cooling water temperature rise exceeds the preset rate. The second signal includes the crystallinity distribution cloud map, predicted ring stiffness value, and compensation node coordinate data. The digital twin model receives the equipment status feedback from the integration module through the OPC UA protocol to synchronously update the virtual production line topology.

4. The digital management system for the whole process of intelligent PE pipe manufacturing according to claim 1, characterized in that, The neural network model of the analysis and optimization module adopts a bidirectional LSTM architecture. The input layer receives the melt temperature time series data, the screw speed fluctuation curve, and the environmental humidity change gradient output by the digital twin module. The output layer generates the temperature compensation value for the third heating zone of the extruder and the dynamic correction coefficient of the traction speed. The training data set of the neural network model contains the mapping relationship between the pipe burst pressure test results and the corresponding process parameters in the historical production process. The temperature compensation value for the third heating zone includes a gradient heating-up strategy based on the prediction of the melt flow front. The traction speed correction coefficient is processed by sliding window mean filtering according to the real-time measured value of the pipe outer diameter. The third signal is synchronized with the device control cycle received by the integration module through the timestamp alignment mechanism.

5. The digital management system for the whole process of intelligent PE pipe manufacturing according to claim 1, characterized in that The defect association process of the detection module includes: performing multi-scale convolutional feature extraction on the surface defect image to identify the morphological features of bubbles, converting the wall thickness laser scanning data into an axial distribution histogram and detecting thickness mutation points. When the bubble diameter exceeds one-fifth of the pipe wall thickness or the thickness mutation gradient reaches the critical value of the material yield strength, start the cross-link traceability process. The traceability process uses the matching abnormal detection timestamp and the historical curve of the process parameters recorded by the digital twin module to locate the hot air temperature fluctuation event in the raw material drying stage or the sudden drop event of the die head pressure of the extruder through the historical curve of the process parameters. The quality traceability report contains the defect cause probability distribution matrix and the recommended process parameter calibration scheme. When the fourth signal is transmitted to the digital twin module, it triggers the model recalibration service.

6. The digital management system for the whole process of intelligent PE pipe manufacturing according to claim 1, characterized in that The reinforcement learning algorithm of the integration module adopts the deep deterministic policy gradient framework. The state space is defined as the temperature difference between the inlet and outlet of the cooling water, the pressure oscillation amplitude of the vacuum sizing tank, and the harmonic distortion rate of the traction motor current. The action space is the adjustment step of the inverter output frequency and the proportional coefficient of the PID control loop. The reward function comprehensively considers the reduction of the pipe ovality deviation, the reduction rate of unit energy consumption, and the penalty term for the number of equipment start-stop times. After the device control instruction is generated, it is verified through the virtual twin simulation. When the predicted execution result causes the pipe pressure resistance level to drop by more than the safety threshold, start the multi-objective Pareto front search to generate an alternative control strategy. When the instruction is issued, an execution priority label and a timeout rollback mechanism are attached.

7. The digital management system for the whole process of intelligent PE pipe manufacturing according to claim 1, characterized in that, The model correction mechanism of the digital twin module includes: when the relative error between the predicted value of the virtual pipe ring stiffness and the actual pressure test result feedback by the detection module continuously exceeds 8% for three cycles, start the iterative optimization of the melt flow rate compensation coefficient, search for the optimal compensation vector in the material characteristic parameter space through the particle swarm algorithm, and synchronously update the hidden layer weight matrix of the LSTM neural network in the analysis and optimization module. The corrected digital twin model recalculates the key quality indicators of the production batches in the past two hours and generates a list of recommended process parameter backtracking adjustments and pushes it to the MES system dashboard.

8. The digital management system for the whole process of intelligent PE pipe manufacturing according to claim 4, wherein, The analysis and optimization module further integrates a multi-objective optimization function. The function takes maximizing the environmental stress cracking index of the pipe and minimizing the energy consumption per unit output as parallel optimization objectives. The constraint conditions include the safety threshold of the extruder torque, the limit value of the heat exchange efficiency of the cooling tower, and the avoidance interval of the mechanical resonance frequency of the traction system. The NSGA-II algorithm is used to generate the Pareto optimal solution set, and the fuzzy comprehensive evaluation is used to select the process parameter combination that takes into account both quality and energy efficiency. The temperature compensation value in the third heating zone includes a feedforward control component predicted based on the change rate of the extrusion volume. The traction speed correction coefficient calculates the dynamic compensation amount in real time according to the axial shrinkage rate of the pipe.

9. The digital management system for the whole process of intelligent PE pipe manufacturing according to claim 8, wherein The blockchain evidence storage process of the detection module includes: structurally processing the raw material supplier code, process parameter version number, and grayscale image of the detection result in the quality traceability report through a Merkle tree to generate a data fingerprint and then writing it into the permissioned chain node. The permissioned chain node includes an electronic quality certificate, and the electronic quality certificate contains a verifiable digital signature and a Beidou time stamp. When the pipe enters the logistics link, the whole life cycle data on the blockchain is associated through the invisible two-dimensional code sprayed on the pipe wall. The two-dimensional code is printed with a ceramic substrate ink that resists high and low temperature deformation. The scanning device can verify the integrity of the production data and the tampering trace layer by layer by parsing the hash pointer.

10. A digital management method for the whole process of manufacturing intelligent PE pipes using any one of claims 1-9, characterized in that, including: S1: Collect various data of raw materials through sensors deployed on the detection equipment, generate a real-time process data stream through edge computing nodes, and form a first signal; S2: According to the first signal and construct a digital twin model, fuse the material characteristic parameters and environmental data to simulate the pipe crystallization process, and form a second signal; S3: Based on the second signal and analyze the correlation between process parameters and pipe performance through a neural network model, generate temperature compensation and traction speed correction instructions, and form a third signal; S4: Receive the defect image and wall thickness data of the detection instrument, associate the defect with the process anomaly and trace it back to the dryness of the raw material, generate a quality traceability report, form a fourth signal and update the digital twin model; S5: Integrate the optimization instructions of the third signal and the quality constraints of the fourth signal, dynamically adjust the cooling water pump frequency and traction power through reinforcement learning, generate equipment control instructions and send them to the production line.

Images