A method for optimizing a double-layer composite guardrail profile extrusion molding die and temperature control parameters in cooperation

Abstract

This invention relates to the field of polymer material extrusion molding technology, and proposes a method for the coordinated optimization of extrusion molds and temperature control parameters for double-layer composite guardrail profiles. This method first uses a coaxially integrated linear array camera and infrared thermal imager to simultaneously acquire images of the profile contour and temperature field. These images are then synchronized and fused with process parameters such as extruder screw speed, melt pressure, and temperatures in various mold zones to extract geometric, image, and temperature feature vectors directly related to product quality. Next, these feature vectors are input into a pre-trained hybrid neural network model to predict the interface peel strength and dimensional comprehensive score in real time. Furthermore, based on the current quality prediction value, a rolling time-domain optimization problem aimed at achieving optimal quality in the future is constructed and solved. The optimal coordinated adjustment commands for the set temperatures in each mold zone and the main screw speed are dynamically calculated and executed.

Description

Technical Field

[0001] This invention relates to the field of polymer material extrusion molding technology, specifically to a method for co-optimizing the extrusion molding die and temperature control parameters of a double-layer composite guardrail profile. Background Technology

[0002] In the field of plastic extrusion molding, especially in the production of double-layer composite guardrail profiles, ensuring the strong bonding between the inner and outer layers and the accuracy of product dimensions has always been a core technological challenge that the industry urgently needs to solve. Traditional production methods heavily rely on the experience of operators, controlling quality by adjusting mold temperature and screw speed. However, this manual control mode has significant drawbacks: First, for the key intrinsic quality indicator of interfacial peel strength, there is a lack of effective online detection methods, and judgment can only be made through subsequent destructive sampling, which cannot achieve real-time control; second, the adjustment of multiple variables such as temperature and speed is often isolated and trial-and-error, lacking collaborative optimization, making it difficult to stably ensure product consistency under complex working conditions; finally, existing technologies cannot predict and proactively compensate for quality changes during the production process, resulting in a high scrap rate and severely restricting production efficiency and product quality stability.

[0003] Therefore, a collaborative optimization method for extrusion molding dies and temperature control parameters of double-layer composite guardrail profiles is proposed. Technically, firstly, multi-source data on the profile contour and temperature field are simultaneously collected using coaxially integrated vision and thermal imaging sensors. Combined with process parameters, key feature vectors directly related to the final quality are extracted. Then, a trained hybrid neural network model is used to predict the interface peel strength and dimensional score online in real time based on these feature vectors. Finally, based on this quality prediction, a rolling time-domain optimization problem is constructed with the goal of achieving optimal quality in future time periods. The optimal adjustment commands for the temperature of each temperature zone of the die and the screw speed of the extruder are dynamically and collaboratively calculated and output, thereby achieving precise, forward-looking, and adaptive control of the production process and improving the automation level and reliability of quality control. Summary of the Invention

[0004] To address the challenges of existing extrusion production technologies that rely on manual experience and cannot predict and coordinate the intrinsic quality, such as interfacial bonding strength, online, this invention provides a method for coordinating the optimization of extrusion molding dies and temperature control parameters for double-layer composite guardrail profiles, comprising the following steps: S1. Hardware Deployment and Synchronous Acquisition of Multi-Source Data: A rigid measuring bracket with an integrated sensor head is installed at the entrance of the shaping section of the extrusion production line. The sensor head includes a linear array camera and an infrared thermal imager, enabling synchronous image acquisition of the same observation area on the profile surface by both. The image data and key process parameters from the programmable logic controller of the extrusion production line are acquired synchronously. The key process parameters include at least the extruder screw speed and die temperature. The image data and process parameter data are hardware-level synchronized and timestamped through an edge computing gateway, and then packaged and sent to the central industrial control computer.

[0005] S2. Data Acquisition Processing and Key Feature Extraction: The central industrial control computer processes the received linear array image, identifies the profile outline and calculates the thickness features; calculates the image texture features in the identified composite interface area; maps the interface coordinates to the synchronously acquired thermal image and extracts the temperature features of the interface area; combines and normalizes the extracted physical features with process parameters to form a feature vector representing the production status.

[0006] S3. Quality Prediction Model Construction and Adaptive Learning: A neural network model is used as an online quality predictor. Its input is the feature vector, and its output is the predicted value of the combined score of interface peeling strength and size. The model is obtained through offline training on historical datasets and updated using an online incremental learning strategy based on cached delayed data pairs.

[0007] S4. Temperature control and speed coordinated optimization and rolling control: Rolling optimization is performed with a fixed control cycle. At each control moment, the set temperature adjustment amount of the mold temperature zone and the screw speed adjustment amount within the future finite time domain are used as decision variables to construct a constrained optimization problem for solution, and the optimal control command obtained from the solution is executed to realize closed-loop rolling control.

[0008] S5. System Workflow Control and Security Assurance: The system follows a hierarchical state machine process. In the steady-state control loop, it sequentially executes data acquisition, feature extraction, quality prediction, rolling optimization solution, and instruction issuance. The system is equipped with a multi-level anomaly handling mechanism and adopts corresponding security control strategies according to different anomaly levels.

[0009] Preferably, in step S1, the integrated sensor head is a coaxial vision and thermal imaging integrated sensor head, which combines the optical paths of the linear array camera and the infrared thermal imager through an optical beam splitter to achieve synchronous acquisition of the same observation area; the key process parameters specifically include the main extruder screw speed, the auxiliary extruder screw speed, the mold inlet melt pressure, and the set temperature and measured temperature of multiple temperature zones of the mold.

[0010] Preferably, in step S1, the rigid measuring bracket is fixed at a predetermined distance downstream of the extrusion die lip; the linear array camera is equipped with a telecentric lens with a focal length of 50 mm, and the infrared thermal imager has a temperature measurement range of 50 degrees Celsius to 300 degrees Celsius; spatial pixel-level alignment between the linear array image and the thermal image is achieved through calibration, with an alignment error not exceeding one pixel; the process parameter data acquisition cycle is 50 milliseconds, and the acquisition delay does not exceed 10 milliseconds.

[0011] Preferably, in step S2, identifying the profile outline and calculating the thickness features includes: sequentially performing Gaussian filtering, contrast-limited adaptive histogram equalization, edge detection, and binarization on the linear array image to identify the outer and inner profiles of the profile, and calculating the outer thickness, inner thickness, and thickness ratio.

[0012] Preferably, in step S2, the calculation of image texture features includes: cutting a strip region of a predetermined pixel width on the identified composite interface line, calculating the gray-level gradient magnitude of all pixels in the region and averaging them to obtain the average gray-level gradient magnitude; and calculating the Shannon entropy of the gray-level distribution in the region.

[0013] Preferably, in step S2, the extraction of interface region temperature features includes: mapping the interface line coordinates in the visual image to an infrared thermogram; extracting a temperature analysis band with a predetermined pixel width along the mapped interface line; calculating the average temperature and temperature standard deviation of the analysis band; and calculating the area-weighted average temperature of the inner main body region and the area-weighted average temperature of the outer main body region to obtain the interlayer temperature difference.

[0014] Preferably, in step S3, the neural network model is a hybrid architecture of convolutional neural network and multilayer perceptron; the online incremental learning strategy is as follows: the system maintains a fixed-capacity first-in-first-out cache pool to store delayed data pairs; when the amount of data in the cache pool reaches a predetermined threshold, the current model is fine-tuned with a low learning rate; if the loss of the model on the validation set after fine-tuning is reduced by a predetermined proportion compared with the original model, the model is updated; otherwise, the original model is retained.

[0015] Preferably, in step S4, the control period of the rolling optimization is ten seconds, and the prediction time domain is three periods; the control period and prediction time domain of the rolling optimization are predetermined values; the objective function includes a quality tracking term and a control action penalty term, the quality tracking term is used to characterize the deviation between the predicted quality and the target quality, and the control action penalty term is used to suppress abrupt changes in the control parameters.

[0016] Preferably, in step S4, the operational constraints of the optimization problem include: the set temperature adjustment amount constraint for each temperature zone, the main screw speed change rate constraint, and the safe temperature range constraint after the actual temperature adjustment for each temperature zone.

[0017] Preferably, in step S5, the multi-level exception handling mechanism includes: Level 1 anomaly handling: When any critical data source loses data for multiple consecutive periods or the data exceeds the physically reasonable range, a Level 1 alarm is triggered, and the control module switches to a safe mode that uses a preset set of safety parameters.

[0018] Level 2 anomaly handling: When the quality prediction value exceeds the empirical range of the model training data for multiple consecutive cycles, it is determined that the working condition exceeds the applicable range of the model, the optimization control is suspended, the control mode is downgraded to the basic control mode that operates according to fixed process parameters, and a level 2 alarm is issued.

[0019] Level 3 anomaly handling: When the feedback deviation of the actuator exceeds the set threshold for multiple cycles and cannot be automatically corrected, the actuator is judged to be faulty, the control loop is blocked, the parameters before the fault are maintained, and a level 3 alarm is issued.

[0020] Compared with the prior art, the present invention provides a method for co-optimizing the extrusion molding die and temperature control parameters of a double-layer composite guardrail profile, which has the following beneficial effects: 1. In this solution, by deeply integrating multi-source information perception, online intelligent prediction and rolling optimization control, a fully closed-loop active quality control system of perception, prediction, decision-making and execution is constructed. Compared with the existing traditional methods that rely on human experience, single-variable feedback or offline sampling inspection, this invention realizes real-time and forward-looking prediction and coordinated control of key quality indicators of the complex dynamic process of composite extrusion molding, namely interface bonding strength and dimensional accuracy.

[0021] 2. In this solution, by synchronizing and fusing multi-sensor data in time and space and feature fusion, it is possible to accurately capture and quantify the deep state information that characterizes product quality, thereby fundamentally solving the inherent defects of low control accuracy and poor stability caused by the lag in perception and the isolation of variables in traditional methods.

[0022] 3. In this solution, a rolling time-domain optimization model with quality prediction as the direct objective is constructed to dynamically and collaboratively optimize multiple key process parameters such as the die temperature field and extrusion speed, achieving a shift from passive correction to proactive defect prevention. This not only improves the consistency and reliability of product quality but also enhances the adaptability and robustness of the production process to disturbances such as raw material fluctuations and equipment aging. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the hardware deployment of the present invention; Figure 2 This is a flowchart of the feature extraction process of the present invention; Figure 3 This is a diagram of the predictive model architecture of the present invention; Figure 4 This is a schematic diagram of the rolling control principle of the present invention; Figure 5 This is a flowchart of the system state machine of the present invention. Detailed Implementation

[0024] The technical solutions of 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.

[0025] Please see Figures 1-5 A method for co-optimizing the extrusion molding die and temperature control parameters of a double-layer composite guardrail profile is disclosed in this embodiment. This method relies on an integrated hardware data acquisition and control system, and the specific steps are as follows: Step 1: Hardware Deployment and Synchronous Acquisition of Multi-Source Data On the extrusion production line, downstream of the extrusion die lip At the inlet of the shaping section, a rigid measuring bracket perpendicular to the extrusion direction is fixedly installed. The bracket is made of a high-temperature resistant alloy to ensure structural stability under production temperature conditions.

[0026] The bracket integrates a coaxial vision and thermal imaging sensor head, containing a line scan camera and an infrared thermal imager. The line scan camera is configured with a focal length... The telecentric lens has its optical axis strictly perpendicular to the extrusion direction and a pixel resolution of no less than 2048 pixels, ensuring high-precision acquisition of the profile outline.

[0027] The infrared thermal imager has a temperature measurement range of 50℃ to 300℃ and a temperature measurement accuracy of ±0.5℃. Its optical path is combined with the optical path of the linear array camera through a high-transmittance optical beam splitter placed at a 45° angle, with a beam splitting ratio of 1:1 and applicable to the visible to mid-infrared bands. This enables both to simultaneously acquire data on the same observation area of ​​the profile surface, covering the full width of the profile cross-section.

[0028] The sensor head connects to the downstream central industrial control computer via Gigabit Ethernet with a communication rate of ≥1000Mbps. After system installation, a strict calibration procedure must be performed: a high-temperature resistant precision checkerboard calibration plate with an accuracy ≤0.01mm and a temperature resistance ≥300℃ is placed on the moving plane of the profile, and linear array images and thermal images are simultaneously acquired under actual production temperature conditions. A precise mapping from the pixel coordinate system to the physical world coordinate system is established through Zhang's calibration method, ultimately achieving spatial pixel-level alignment of the two image data, with an alignment error ≤1 pixel.

[0029] Inside the electrical control cabinet, a real-time data subscription link is established by accessing the server of the main programmable logic controller (PLC) of the extrusion production line via the PROFINET industrial communication protocol. This is done at fixed intervals. Stable reading of key process parameter variables with a data acquisition delay of ≤10ms.

[0030] The parameters collected include: main extruder screw speed. (Unit: rpm, measurement accuracy ±0.1 rpm), Auxiliary extruder screw speed (Unit: rpm, measurement accuracy ±0.1 rpm), melt pressure sensor reading at mold inlet. (Unit: bar, measurement range 0~200 bar, accuracy ±0.5%FS).

[0031] The set temperature of the three independent temperature control zones on the mold body , , (Unit: °C, setting accuracy ±0.1 °C), from the feed end to the lip end, temperature zones are defined as zone 1, zone 2, and zone 3 respectively; the measured temperature of the embedded thermocouple in each temperature control zone is... , , (Unit: °C, temperature range: 0~400 °C, accuracy: ±0.3 °C).

[0032] A multi-functional industrial-grade data acquisition device is used as the edge computing gateway, which supports multi-protocol access and hardware-level synchronization triggering. The external trigger signals from the line scan camera, the external trigger signal from the infrared thermal imager, and the system clock signal from the PLC are all connected to this gateway, which provides a globally unified hardware triggering and time synchronization mechanism.

[0033] The gateway assigns a uniform and accurate timestamp to each set of synchronously collected data. (Time precision) The parameter data and image data are encapsulated into standard data packets conforming to JSON format and sent to the central industrial control computer for further processing via a PCIe high-speed bus with a transmission rate of ≥2GB / s.

[0034] Step 2: Data Processing and Key Feature Extraction The central industrial control computer processes the received data packets according to a standardized process. It performs a 5×5 Gaussian filter on the original linear array image to suppress random noise. Then, it uses the contrast-limited adaptive histogram equalization (CLAHE) algorithm to enhance the image contrast. The CLAHE cropping limit coefficient for the composite interface region is set to 2.0.

[0035] Next, edge detection is performed using the Canny edge detection operator with a high threshold of 150 and a low threshold of 50. Then, the binarization threshold is determined by the Otsu's method (OTSU) to convert the image into a binary image. Using the contour finding algorithm in OpenCV, combined with the preset geometric constraints of the profile, the outermost contour of the profile and the contour of the inner material are identified.

[0036] The thickness of the outer layer of the profile is calculated based on the pixel coordinates of the contour and the calibrated physical mapping relationship. Inner layer thickness (Unit: mm, calculation accuracy ±0.01 mm), and further calculate the thickness ratio of the two. .

[0037] On the identified composite interface line, a line with a width perpendicular to the interface direction is cut. A strip-shaped region of pixels. The gray-level gradient magnitude of all pixels within this strip-shaped region is calculated using a 3×3 Sobel operator, and the average value is taken as the average gray-level gradient magnitude. .

[0038] Simultaneously calculate the Shannon entropy of the grayscale distribution in this region. The formula for calculating Shannon entropy is: ,in grayscale value The probability of its occurrence.

[0039] The system calls the pre-stored calibration file and uses affine transformation to accurately map the physical coordinates of the composite interface lines identified in the visual image to the pixel coordinates of the synchronously acquired infrared thermal image, with a mapping error of ≤1 pixel.

[0040] On the infrared thermal image, along the mapped interface line, extract a region with a width of Temperature analysis band for pixels, covering the composite interface and adjacent areas on both sides.

[0041] Calculate the average temperature of all pixels within the temperature analysis band. (Unit: °C, accuracy ±0.1 °C), characterizing the real-time temperature level of the interface region; calculating the standard deviation of the temperature of all pixels within the temperature analysis band. (Unit: °C)

[0042] Select the main body areas of the inner and outer layers respectively, avoiding the edges by 5 pixels, and calculate the area-weighted average temperature of each area. (Inner layer) and (Outer layer), interlayer temperature difference (Unit: °C)

[0043] The physical features extracted in real time and the process parameters read synchronously are combined to form a characterization. Feature vector of complete production state at any given time The vector form is as follows:

[0044] superscript This represents the transpose of a vector, where the values ​​of each element are normalized to the range [0,1].

[0045] Step 3: Quality Prediction Model Construction and Adaptive Learning Using a hybrid neural network model As an online quality predictor, this model is a hybrid architecture of convolutional neural networks (CNN) and multilayer perceptrons (MLP).

[0046] The model input is the normalized feature vector from step two. (Dimension 14×1), the output is a two-dimensional prediction vector. .in The value is the predicted value of the interfacial peel strength (unit: N / cm, prediction accuracy ±0.5N / cm). This is the predicted value for the overall size score.

[0047] The CNN module of the model contains two convolutional layers and one pooling layer. The convolutional kernel sizes are 3×1 and 2×1, and the number of convolutional kernels are 32 and 64, respectively. The pooling layer uses max pooling with a pooling kernel size of 2×1, which is used to extract local correlation features from the feature vector.

[0048] The MLP module contains three fully connected layers with 128, 64, and 2 neurons respectively. The ReLU activation function is used in all layers except the output layer. The output layer uses a Sigmoid function and a linear function, respectively, for... Normalized output and Output directly.

[0049] Offline model training uses historical datasets Conduct supervised training, among which and , This is the true value measured in the laboratory. Measured using a universal tensile testing machine After measurement using a 3D profilometer, the results are calculated according to preset scoring rules.

[0050] Training to minimize the loss function For the goal, among which These are all trainable parameters for the model.

[0051] The Adam optimizer is used, with an initial learning rate set to The number of iterations was set to 100, and the batch size was set to 32. Training was stopped using early stopping, where the validation set loss did not decrease for 10 consecutive iterations, ultimately yielding the basic prediction model. .

[0052] To cope with slow changes in production conditions and ensure the long-term accuracy of the prediction model, the system employs an online update strategy. The system records... Feature vector at time step and the corresponding process scenario information, after a fixed physical delay time After hours, obtain the corresponding laboratory measured true value: , constituting delayed data pairs .

[0053] The system maintains a fixed capacity. First-In-First-Out (FIFO) Cache Pool It is used to store the latest delayed data pairs. When the amount of data in the cache pool reaches a threshold... Incremental learning will be automatically initiated at that time.

[0054] With a low learning rate Load the current model , cache pool The data pairs were divided into a fine-tuning set and a validation set in a 7:3 ratio, and gradient descent updates were performed for 10 iterations. The updated model parameters are as follows: ,in For the model The loss function value on the fine-tuning set.

[0055] If the loss of the fine-tuned model on the validation set is reduced by ≥5% compared to the original model, then the new model should be used. Replace the original model Otherwise, discard the new model, keep the original model unchanged, and record the failure log for this fine-tuning.

[0056] Step 4: Temperature control and speed synergistic optimization and rolling control The core of the system control is based on a cycle The rolling optimization process achieves precise control of quality indicators by dynamically adjusting key process parameters.

[0057] At each control moment The system constructs and solves a finite-time optimization problem. The decision variables are defined as future... The sequence of control input adjustments within each control cycle, in the total time domain. .

[0058] This includes the sequence of temperature adjustment values ​​for the three temperature zones of the mold: , express Time prediction Time of the first The set temperature adjustment amount for each temperature zone.

[0059] It also includes the main screw speed adjustment sequence. , express Time prediction Adjustment amount of main screw speed at all times.

[0060] The objective function is to minimize the overall performance index. This indicator is determined by the quality tracking item. and control effect penalty items Composed of two parts, balancing quality tracking accuracy and control stability: 1. Quality Tracking Items : Characterizes the deviation between the predicted quality and the target quality, expressed as:

[0061] 2. Control effect penalty item To suppress sudden changes in control parameters and ensure stable production, the expression is:

[0062] 3. Overall objective function:

[0063] in, , The preset target values ​​are the interface peel strength and the comprehensive size score. , For prediction models Future calculated based on predicted state Step quality prediction value. Weighting coefficient. , , , .

[0064] Operational constraints include: temperature adjustment amplitude constraints. ,in , Speed ​​change rate constraint ,in Absolute temperature safety constraints ,in , .

[0065] Using convergence accuracy The Sequential Quadratic Programming (SQP) algorithm with a maximum of 50 iterations is used to solve the above-mentioned constrained nonlinear optimization problem.

[0066] After the solution is obtained, the system only adopts the instruction of the first control step in the optimal solution sequence, i.e., the rolling time-domain control strategy. The optimal control instruction is:

[0067] Will The signals are transmitted to the corresponding actuators via industrial communication protocols. The temperature control actuator is used to adjust the temperature of the temperature zone, and the screw speed driver is used to adjust the speed of the main screw.

[0068] In the next control cycle Based on the latest collected sensor data and process parameters, the system re-executes the above optimization process to achieve closed-loop rolling control.

[0069] Step 5: System Workflow Control and Security Assurance The system's software execution follows a hierarchical state machine process, ensuring safety, reliability, and continuity from startup to stable operation.

[0070] After the system is powered on, it automatically executes the initialization process, sequentially checking the communication status of the line scan camera and infrared thermal imager, the PLC connection link (communication delay ≤20ms), and the response status of the actuator.

[0071] After the self-test passes, the latest camera-thermal imager calibration parameter file will be automatically loaded; if the calibration file is missing or expired, the system will issue an alarm and pause the startup.

[0072] Then, the latest quality prediction model is loaded, including the model structure file and parameter weight file; if it is a brand new system without a historical model, a pre-trained general model is loaded.

[0073] The system then enters "data monitoring mode," continuously collecting data and inputting it into the model for prediction, but without outputting control commands. The warm-up phase ends and the system automatically enters the steady-state control cycle phase once the coefficient of variation of the model's predicted values ​​is ≤3% for 10 consecutive cycles.

[0074] The system is based on To execute a steady-state control loop with a fixed period, each loop performs the following operations in sequence: 1. The edge gateway, in accordance with the requirements of step one, completes the synchronous acquisition of linear array images, infrared thermal images, and process parameters, timestamps them, encapsulates them into standard data packets, and sends them to the central industrial control computer.

[0075] 2. After receiving the data packet, the central industrial control computer completes image processing and temperature field analysis according to the process in step two, and calculates the feature vector at the current moment. .

[0076] 3. Normalize the Input quality prediction model Real-time output of quality prediction values .

[0077] 4. Based on the current forecast value Using the process conditions as initial conditions, an optimization problem is constructed according to the relevant requirements in step four, and the optimal control command is obtained by solving it using the SQP algorithm. .

[0078] 5. The optimization results enter the safety constraint verification module, where dual verification is performed: ① Command amplitude verification, checking... , Does it meet the constraints of step four? ② Safety boundary verification, calculate the actual temperature of the adjusted temperature zone. Confirm that it is in Within the range.

[0079] 6. If the verification passes, then... The command is sent to the executing agency; if the verification fails, the system automatically adopts a backup strategy, prioritizing the maintenance of the control command from the previous cycle. If three consecutive verifications fail, the system switches to a preset safety parameter set. At the same time, it will issue an audible and visual alarm.

[0080] The background process manages the delayed data cache pool in real time, stores delayed data pairs according to the relevant requirements in step three, and automatically triggers model fine-tuning when the incremental learning conditions are met, and automatically backs up the updated model parameters to local storage.

[0081] The system is equipped with a three-level exception handling mechanism to ensure production safety under various abnormal scenarios: 1. Level 1 Anomaly (Data Flow Anomaly): If any critical data source (line scan camera, infrared thermal imager, PLC parameters) continuously... If data is lost within a single cycle, or if the data exceeds a physically reasonable range, the system immediately triggers a Level 1 alarm. The control module automatically switches to a safe mode, using a preset conservative fixed parameter set. To ensure proper operation and prevent production accidents.

[0082] 2. Secondary anomaly (anomaly in prediction confidence): If the quality prediction value... or Five consecutive cycles exceeding the empirical range of the model training data , , Based on the minimum and maximum values ​​of historical measured data, the system determines that the current operating condition exceeds the applicable range of the model. It automatically pauses optimized control, downgrades to basic control mode, operates only according to fixed process parameters without dynamic adjustments, issues a level-two alarm, and prompts operator intervention.

[0083] 3. Level 3 Anomaly (Execution Feedback Anomaly): The system collects the actual feedback values ​​of the actuator in real time. Calculate feedback deviation Temperature deviation is measured in °C, and speed deviation is measured in rpm. If the deviation exceeds the set threshold for three consecutive cycles... Temperature deviation threshold Speed ​​deviation threshold If the fault cannot be eliminated through automatic correction, the system determines that the actuator is faulty. Control of the affected control loop is immediately disabled, pre-fault parameters are maintained, a level-three alarm is issued, and maintenance personnel are notified for repair.

[0084] Finally, it should be noted that the above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Any obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for co-optimizing the extrusion molding die and temperature control parameters of a double-layer composite guardrail profile, characterized in that: Includes the following steps: S1. Hardware Deployment and Synchronous Acquisition of Multi-Source Data: A rigid measuring bracket with an integrated sensor head is installed at the entrance of the shaping section of the extrusion production line. The sensor head includes a linear array camera and an infrared thermal imager, enabling synchronous image acquisition of the same observation area on the profile surface by both. The image data and key process parameters from the programmable logic controller of the extrusion production line are acquired synchronously. The key process parameters include at least the extruder screw speed and the die temperature. The image data and process parameter data are synchronized at the hardware level through the edge computing gateway and stamped with a unified timestamp. After encapsulation, the data is sent to the central industrial control computer. S2. Data Acquisition Processing and Key Feature Extraction: The central industrial control computer processes the received linear array image, identifies the profile outline and calculates the thickness features; and calculates the image texture features in the identified composite interface area. The interface coordinates are mapped to the synchronously acquired thermal image to extract the temperature features of the interface area. The extracted physical features are combined with process parameters and normalized to form a feature vector representing the production status. S3. Quality Prediction Model Construction and Adaptive Learning: A neural network model is used as an online quality predictor. Its input is the feature vector, and its output is the predicted value of the interface peeling strength and size comprehensive score. The model is obtained through offline training on historical datasets and updated using an online incremental learning strategy based on cached delayed data pairs; S4. Temperature control and speed synergistic optimization and rolling control: Rolling optimization is performed with a fixed control cycle. At each control moment, the set temperature adjustment amount of the mold temperature zone and the screw speed adjustment amount within the future finite time domain are used as decision variables to construct a constrained optimization problem for solution, and the optimal control command obtained from the solution is executed to achieve closed-loop rolling control. S5. System Workflow Control and Security Assurance: The system follows a hierarchical state machine process. In the steady-state control loop, it sequentially executes data acquisition, feature extraction, quality prediction, rolling optimization solution, and instruction issuance. The system is equipped with a multi-level anomaly handling mechanism and adopts corresponding security control strategies according to different anomaly levels.

2. The method for co-optimizing the extrusion molding die and temperature control parameters of a double-layer composite guardrail profile according to claim 1, characterized in that: In step S1, the integrated sensor head is a coaxial vision and thermal imaging integrated sensor head. The optical paths of the linear array camera and the infrared thermal imager are combined through an optical beam splitter to achieve synchronous acquisition of the same observation area. The key process parameters specifically include the main extruder screw speed, the auxiliary extruder screw speed, the die inlet melt pressure, and the set temperature and measured temperature of multiple temperature zones of the die.

3. The method for co-optimizing the extrusion molding die and temperature control parameters of a double-layer composite guardrail profile according to claim 2, characterized in that: In step S1, the rigid measuring bracket is fixed at a predetermined distance downstream of the extrusion die lip; the line array camera is equipped with a telecentric lens with a focal length of 50 mm, and the infrared thermal imager has a temperature measurement range of 50 degrees Celsius to 300 degrees Celsius. The linear array image and the thermal image are aligned at the spatial pixel level through calibration, with an alignment error of no more than one pixel; the process parameter data acquisition cycle is fifty milliseconds, and the acquisition delay is no more than ten milliseconds.

4. The method for co-optimizing the extrusion molding die and temperature control parameters of a double-layer composite guardrail profile according to claim 1, characterized in that: In step S2, identifying the profile outline and calculating the thickness features includes: sequentially performing Gaussian filtering, contrast-limited adaptive histogram equalization, edge detection, and binarization on the linear array image to identify the outer and inner profiles of the profile, and calculating the outer thickness, inner thickness, and thickness ratio.

5. The method for co-optimizing the extrusion molding die and temperature control parameters of a double-layer composite guardrail profile according to claim 4, characterized in that: In step S2, the calculation of image texture features includes: cutting a strip region of a predetermined pixel width on the identified composite interface line, calculating the gray-level gradient magnitude of all pixels in the region and averaging them to obtain the average gray-level gradient magnitude; and calculating the Shannon entropy of the gray-level distribution in the region.

6. The method for co-optimizing the extrusion molding die and temperature control parameters of a double-layer composite guardrail profile according to claim 1, characterized in that: In step S2, the extraction of interface region temperature features includes: mapping the interface line coordinates in the visual image to an infrared thermogram; extracting a temperature analysis band with a predetermined pixel width along the mapped interface line; calculating the average temperature and temperature standard deviation of the analysis band; and calculating the area-weighted average temperature of the inner main body region and the area-weighted average temperature of the outer main body region to obtain the interlayer temperature difference.

7. The method for co-optimizing the extrusion molding die and temperature control parameters of a double-layer composite guardrail profile according to claim 1, characterized in that: In step S3, the neural network model is a hybrid architecture of convolutional neural network and multilayer perceptron; the online incremental learning strategy is as follows: the system maintains a fixed-capacity first-in-first-out cache pool to store delayed data pairs; when the amount of data in the cache pool reaches a predetermined threshold, the current model is fine-tuned with a low learning rate; if the loss of the model on the validation set after fine-tuning is reduced by a predetermined proportion compared with the original model, the model is updated; otherwise, the original model is retained.

8. The method for co-optimizing the extrusion molding die and temperature control parameters of a double-layer composite guardrail profile according to claim 1, characterized in that: In step S4, the control period of the rolling optimization is ten seconds, and the prediction time domain is three periods; the control period and prediction time domain of the rolling optimization are predetermined values; the objective function includes a quality tracking term and a control action penalty term, the quality tracking term is used to characterize the deviation between the predicted quality and the target quality, and the control action penalty term is used to suppress abrupt changes in the control parameters.

9. The method for co-optimizing the extrusion molding die and temperature control parameters of a double-layer composite guardrail profile according to claim 8, characterized in that: In step S4, the operational constraints of the optimization problem include: the set temperature adjustment amount constraint for each temperature zone, the main screw speed change rate constraint, and the safe temperature range constraint after the actual temperature adjustment for each temperature zone.

10. The method for co-optimizing the extrusion molding die and temperature control parameters of a double-layer composite guardrail profile according to claim 1, characterized in that: In step S5, the multi-level exception handling mechanism includes: Level 1 anomaly handling: When any critical data source loses data for multiple consecutive cycles or the data exceeds the physically reasonable range, a Level 1 alarm is triggered, and the control module switches to a safe mode that uses a preset set of safety parameters. Level 2 anomaly handling: When the quality prediction value exceeds the empirical range of the model training data for multiple consecutive cycles, it is determined that the working condition exceeds the applicable range of the model, the optimization control is suspended, the control mode is downgraded to the basic control mode that operates according to fixed process parameters, and a level 2 alarm is issued. Level 3 anomaly handling: When the feedback deviation of the actuator exceeds the set threshold for multiple cycles and cannot be automatically corrected, the actuator is judged to be faulty, the control loop is blocked, the parameters before the fault are maintained, and a level 3 alarm is issued.

Images