Process optimization method of film coating production line

Abstract

The invention discloses a process optimization method of a film coating production line, aims to construct a closed-loop system of digital twinborn precise modeling-AI dynamic decision-automatic execution, solves the problem of dependence on efficiency, quality and manpower in traditional film coating production, and comprises the following steps: constructing a digital twinborn model of the film coating production line; synchronizing physical production line data in real time; an AI decision model is trained, and coating speed, temperature and pressure parameters are dynamically optimized; and controlling the automatic equipment to execute the process instruction, and performing online feedback calibration. Through deep coupling of the digital twinning and the AI decision in the steps, virtual-real collaboration, accurate control and resource efficiency are realized, so that the trial and error cost of the digital twinning rehearsal is reduced, and the response speed of the AI decision is improved; micron-sized thickness regulation and nano-filler dispersion optimization are carried out; the energy consumption is reduced by 25%, and the glue utilization rate is increased to 98%.

Description

Technical Field

[0001] This invention relates to the field of pipeline manufacturing technology, and in particular to a process optimization method for a coating production line. Background Technology

[0002] Krah pipe is a reinforced structural wall pipe made of high-density polyethylene (HDPE) through spiral winding and welding processes. It is mainly used in gravity flow systems such as drainage and sewage discharge. The coating process for Krah pipes is a crucial step in improving their weather resistance and functionality. Traditional Krah pipe coating production technology mainly relies on the following methods: Human experience-driven: Process parameters (such as coating speed, temperature, and pressure) rely on manual settings, which are slow to adjust and have low accuracy (thickness deviation > ±15%).

[0003] Semi-automated equipment: coating machines, slitting machines and other single-machine equipment operate independently, lacking global coordination, and mold changing takes as long as 4-6 hours.

[0004] Quality inspection is lagging behind: it relies on manual sampling (sampling rate < 5%) and cannot intercept defects in real time (missed detection rate > 20%).

[0005] The disadvantages of existing technology are as follows: Low efficiency: Manual mold changing and parameter adjustment result in an overall equipment efficiency (OEE) of less than 60%. Large quality fluctuations: Uneven coating thickness, frequent defects such as interface peeling (scrap rate > 3%). High labor intensity: High-temperature and high-risk processes rely on manual operation (such as flame treatment and defect re-inspection); Lack of intelligence: The digital model is disconnected from the physical production line and cannot respond to disturbances (such as raw material fluctuations or equipment malfunctions) in real time. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of the existing technology mentioned in the background section by proposing a process optimization method for a kraft tube coating production line that integrates digital twins and artificial intelligence. This method achieves automated, high-precision, and low-energy-consumption operation of the entire production process through virtual-real collaboration and intelligent decision-making.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A process optimization method for a coating production line includes the following steps: Build a digital twin model of the film coating production line and synchronize physical production line data in real time; Train an AI decision-making model to dynamically optimize coating speed, temperature, and pressure parameters; Control automated equipment to execute process instructions and provide online calibration feedback.

[0008] As a further step in the method of the present invention, the digital twin model is constructed in the following manner: Laser scanning acquires 3D point cloud data of the production line (accuracy 0.01mm). Embedded multiphysics simulation engine (ANSYS Fluent / Mechanical); Real-time mapping of equipment status data (vibration, temperature, pressure).

[0009] As a further step in the method of the present invention, the AI ​​decision model is a deep reinforcement learning (DRL) architecture, comprising: State space: coating speed, adhesive viscosity, substrate temperature, equipment load; Motion space: parameter adjustment amount (±10% of the baseline value); Reward function: R = 0.6 × quality score + 0.3 × efficiency score - 0.1 × energy consumption score.

[0010] As a further step in the method of the present invention, the automated equipment includes: Six-axis robotic arm (repeat positioning accuracy ±0.02mm); AGV transportation system (laser SLAM navigation, positioning error ≤1mm); Online inspection unit (X-ray thickness gauge + hyperspectral camera).

[0011] As a further step in the method of the present invention, an exception handling mechanism is also included, which includes: Real-time monitoring of coating pressure fluctuations; triggering the die head self-cleaning program when the deviation is > ±5%. After AI identifies coating defects, it automatically segments and marks abnormal segments.

[0012] As a further step in the method of the present invention, the online feedback calibration includes: Digital twins are compared with actual production data to generate error compensation parameters; Update the weights of the AI ​​decision-making model and optimize the instructions for the next cycle.

[0013] As a further step in the method of the present invention, the method supports rapid changeover for multiple specifications: RFID identifies pipe specifications and automatically retrieves the corresponding process parameter library; The digital twin simulates the model change path, and the robotic arm completes the model change within 30 minutes.

[0014] As a further step in the method of the present invention, the use of a federated learning framework includes: Each production line trains its model locally and then uploads the gradient parameters to the cloud in encrypted form. The global model is aggregated, updated weekly, and distributed to all nodes.

[0015] As a further step in the method of the present invention, an energy consumption optimization module is also included: Digital twins simulate energy consumption curves under different parameters, and AI recommends the lowest energy consumption combination; The waste heat recovery system uses the waste heat from the drying section for substrate preheating.

[0016] As a further step in the method of the present invention, the method of the present invention is applied to the following scenarios: High abrasion resistant mining corrugated tubing (TPU coated); Photocatalytic self-cleaning drainage pipe (TiO2 coating).

[0017] Compared with the prior art, the beneficial effects of the present invention are at least as follows: This invention proposes a process optimization method for a lamination production line, aiming to construct a closed-loop system of "precise digital twin modeling → AI dynamic decision-making → automated execution," which solves the efficiency, quality, and manual labor dependence problems in traditional lamination production and achieves the following objectives: Production speed increased to ≥10m / min, OEE >85%; Coating thickness deviation ≤ ±3%, defect rate < 0.1%; Human intervention is reduced by 70%, and high-risk processes are 100% automated.

[0018] Based on the above objectives, the following advantages can be achieved: Virtual-real collaboration: Digital twin simulations reduce trial-and-error costs, and AI decision-making improves response speed; Precise control: micron-level thickness regulation and nanofiller dispersion optimization; Resource efficiency: Energy consumption is reduced by 25%, and glue utilization rate is increased to 98%. Detailed Implementation

[0019] The invention will be more readily understood by referring to the following detailed description of preferred embodiments and included examples. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. In case of conflict, the definitions in this specification shall prevail.

[0020] In some instances, approximate terms may correspond to the precision of the instrument used to measure the value. In this specification and claims, scope definitions may be combined and / or interchanged, unless otherwise stated, these scopes include all sub-scopes contained therein.

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0022] A process optimization method for a coating production line, the core of which is to achieve adaptive optimization of process parameters through deep coupling of digital twins and AI decision-making, roughly includes the following steps: Step 1: Multi-source data acquisition and twin modeling Sensor network: Deploy vibration, temperature, pressure, and vision sensors to collect equipment status and process data in real time, with a sampling rate ≥1kHz; Digital twin construction: Based on laser scanning and CAD modeling, a high-fidelity production line model is generated and embedded with physical rules, including heat conduction equations and fluid dynamics models.

[0023] Step 2: AI-driven process optimization Dynamic parameter recommendation: Train a deep reinforcement learning (DRL) model, input real-time production data, and output the optimal combination of process parameters, such as coating speed and glue flow rate; Abnormal operating condition handling: AI identifies equipment abnormalities, such as a sudden increase in extruder pressure of >10%, triggering a self-repair program, such as reducing speed and cleaning the die head.

[0024] Step 3: Automated Execution and Feedback Intelligent equipment control: Through the collaboration of PLC and robotic arm, automatic mold changing, coating, slitting and other operations are completed (positioning accuracy ±0.05mm). Closed-loop feedback calibration: Online detection data is transmitted back to the digital twin in real time to iteratively optimize the model prediction accuracy, with an accuracy error rate of <1%.

[0025] Step 4: Iterative Upgrade of the Knowledge Base Federated learning: Encrypted data sharing across multiple production lines, weekly updates to the global model, and improved decision robustness in small sample scenarios; Fault Case Library: Records historical anomaly data and constructs fault trees (FTAs) to assist AI diagnosis, with a diagnostic accuracy rate of >95%.

[0026] A process optimization method for a coating production line, more specifically: 1. Process optimization methods based on digital twins and AI decision-making, including: (1) Digital twin modeling: A 3D model of the production line is generated by laser scanning (accuracy 0.01mm) and CAD modeling, integrating the physical parameters of the equipment (such as the flow channel size of the coating machine die head and the heat capacity of the hot press roller). Embedded multiphysics simulation engines (such as ANSYS Fluent for simulating glue rheological properties and Mechanical for calculating structural stress).

[0027] (2) AI decision-making: The Deep Reinforcement Learning (DRL) algorithm is used to input real-time sensor data (temperature, pressure, speed, etc.) and output parameter optimization instructions (such as coating speed adjustment ±10%).

[0028] (3) Automated execution: Instructions are sent to the PLC via industrial Ethernet (Profinet protocol) to drive the robotic arm (such as KUKA KR1000) to perform coating and slitting actions.

[0029] (4) Feedback calibration: Online detection data (such as X-ray thickness measurement results) are compared with twin predictions. If the error is greater than 3%, the model parameters are iteratively updated (gradient descent method).

[0030] 2. Methods for constructing digital twin models, including: (1) Laser scanning and modeling: Use a 3D laser scanner (such as FARO Focus) to collect geometric data from the production line and generate a point cloud model (density > 1 million points / m²). 3 ); Reconstruct it into a parametric CAD model in Siemens NX and annotate dynamic parts (such as robotic arm joints and coating head lip).

[0031] (2) Embedding of physical rules: Thermodynamics module: Calculates the temperature field distribution of the hot press roller based on Fourier's law (grid size 1mm); Fluid dynamics module: The Carreau model is used to simulate the non-Newtonian fluid properties of glue (shear thinning index n=0.3).

[0032] (3) Real-time data mapping: Sensor data (vibration, temperature, etc.) is synchronized to the twin via the OPC UA protocol, with a refresh rate ≥100Hz.

[0033] 3. AI decision-making model architecture, including: (1) DRL framework design: ①State space: Static parameters: substrate type (HDPE / TPU), adhesive viscosity (3000-5000cP); Dynamic parameters: real-time coating speed (m / min), hot press roller temperature (°C), ambient humidity (%RH).

[0034] ②Motion space: Discrete actions: coating speed adjusted by ±5% / ±10%, glue flow rate adjusted by ±3% / ±5%; Continuous operation: fine adjustment of hot press roller pressure (0.1-0.5MPa step).

[0035] ③Reward function: Quality score: Thickness deviation (weight 0.6), number of defects (weight 0.3). Efficiency score: Production speed (weight 0.3); Energy consumption score: Energy consumption per unit (weight -0.1).

[0036] (2) Training process: Initial stage: Supervised pre-training using historical data (100,000 sets) (cross-entropy loss function); Online reinforcement learning: Optimize the policy network through real-time interactive data from the production line (ε-greedy exploration strategy).

[0037] 4. The automated execution equipment consists of: (1) Six-axis robotic arm: Model: KUKA KR 1000 Titan, load capacity 1 ton, repeatability ±0.03mm; Task: Automatically change the coating die head (complete the DN300 to DN3000 changeover within 30 minutes) and accurately spray the adhesive (trajectory error <0.1mm).

[0038] (2) AGV transportation system: Navigation method: Laser SLAM (positioning error ±1mm) + QR code-assisted positioning; Functions: Transporting base material rolls (maximum load capacity 3 tons), waste recycling (automatic docking with waste bin).

[0039] (3) Online detection unit: X-ray thickness gauge: measuring range 0.1-5mm, scanning frequency 100Hz; Hyperspectral camera: wavelength range 400-2500nm, identifies coating components (e.g., alarm for TiO2 concentration deviation > 5%).

[0040] 5. Exception handling mechanism, including: (1) Coating pressure monitoring: Sensor: Piezoresistive pressure transmitter (range 0-20MPa, accuracy ±0.1%FS); Triggering condition: Pressure fluctuation > ±5% for 10 seconds; Response action: Reduce speed to 50% and start ultrasonic cleaning of the mold head (frequency 28kHz, power 500W).

[0041] (2) Defect segmentation and marking: AI recognition: YOLOv8 model (mAP@0.5>95%) is used to detect bubbles and scratches; Execution process: Femtosecond laser cuts out abnormal segments (cut width 0.05mm), inkjet printer marks the location (QR code traceability).

[0042] 6. Feedback calibration methods, including: (1) Error compensation generation: Comparison: Digital twin predicted thickness distribution vs. X-ray measured thickness; Compensation algorithm: The PID controller calculates the glue flow correction (integral time constant Ti = 2s).

[0043] (2) Model update mechanism: Incremental learning: Collect 10,000 new data sets every week and fine-tune the DRL model (learning rate 0.0001).

[0044] Hot update: Deploy new models via edge computing nodes (NVIDIA Jetson AGX) without production downtime.

[0045] 7. Quick changeover to multiple specifications, including: (1) RFID identification: Tag type: UHF RFID (frequency 860-960MHz), storing tube diameter, material, and process parameters; Reading method: Fixed reader (reading distance 3m, speed 200 tags / second).

[0046] (2) Mold change path rehearsal: Digital twin simulation: generating collision-free paths based on the RRT* algorithm (validated by 100,000 Monte Carlo samplings); Robotic arm execution: 6-DOF coordinated motion (acceleration 0.5g), mold change time ≤30 minutes.

[0047] 8. Federated learning framework, including: (1) Gradient encryption upload: Encryption algorithm: Homomorphic encryption (Paillier algorithm), local model gradient parameters are encrypted before being uploaded to the cloud; Aggregation method: FedAvg algorithm, weighted average gradient of each node (weights are allocated according to the amount of data).

[0048] (2) Model update cycle: Global model: Data is aggregated every Friday morning, trained, and then distributed to various production lines; Compatibility Guarantee: The API interface supports ONNX format models and is compatible with different brands of devices.

[0049] 9. Energy consumption optimization module, including: (1) Digital twin energy consumption simulation: Input parameters: coating speed, drying temperature, ambient temperature and humidity; Output: Energy consumption curve per unit (kWh / m³), recommended minimum energy consumption parameter combination.

[0050] (2) Waste heat recovery system: Heat exchanger: Plate heat exchanger (efficiency > 80%), recovering heat from the waste gas in the drying section (120℃ → 60℃); Application scenario: Preheat the substrate to 50-80℃ to reduce the energy consumption of adhesive heating by 30%.

[0051] 10. Application scenario expansion, including: (1) High abrasion-resistant mining pipe: Coating material: TPU + 30% silicon carbide (3 times better abrasion resistance); Process parameters: Coating speed 8m / min, hot pressing temperature 130℃.

[0052] (2) Photocatalytic self-cleaning tube: Coating formulation: TiO2 nanoparticles (20nm particle size) + UV-cured epoxy resin; Curing conditions: UV-LED light source (wavelength 365nm, energy density 800mJ / cm²) 2 ).

[0053] Further specific implementations of the method of the present invention include: 1. System Hardware Configuration Sensor networks: Vibration monitoring: PCB 352C33 ICP accelerometer (frequency response 0.5Hz-10kHz); Temperature monitoring: FLIR A655sc infrared thermal imager (resolution 640×480); Thickness measurement: NDC X-ray thickness gauge (accuracy ±0.1μm).

[0054] Implementing agency: Coating robotic arm: KUKA KR 1000 Titan (load capacity 1t, repeatability ±0.03mm). AGV: Hikrobot lurking lifting type (load capacity 3t, navigation accuracy ±1mm).

[0055] 2. Software Algorithm Implementation Digital twin platform: Modeling tools: Siemens NX + ANSYS Twin Builder; Real-time rendering: NVIDIA Omniverse (supports 8K visualization).

[0056] AI decision engine: Framework: PyTorch 1.12 + RLlib; Training data: 100,000 sets of historical production records (including normal / abnormal operating conditions); Hyperparameters: LSTM hidden layer with 256 nodes, learning rate 0.001, discount factor γ=0.99.

[0057] Example of implementation process (coating DN1200 carat tube) Step 1: Data Acquisition and Twin Synchronization The sensor collects real-time data on coating machine pressure (15MPa), substrate temperature (80℃), and ambient humidity (45%RH). The digital twin model maps the data and simulates and predicts the coating thickness distribution (target value 1.2mm ± 0.03mm).

[0058] Step 2: AI Dynamic Decision Making The DRL model analyzes the current situation and recommends increasing the coating speed to 10 m / min and increasing the adhesive flow rate by 5%. The digital twin simulation confirmed that there were no quality risks, and the instructions were sent to the PLC for execution.

[0059] Step 3: Automated Execution The robotic arm sprays adhesive along a trajectory, while the AGV simultaneously transports the pipes to the curing area. Online detection revealed a local thickness deviation (1.15mm), triggering compensation spraying (additional adhesive amount 0.05mm).

[0060] Step 4: Closed-loop feedback Actual production data is fed back to the twin model to calibrate the thermal conductivity error (from 5% to 1%). Update the AI ​​model weights to improve the accuracy of decisions in the next batch.

[0061] The effects of the above implementation are as follows: Production efficiency increased from 2 m / min to 12 m / min, representing a 500% increase. The uniformity of coating thickness has been improved from ±15% to ±2%, representing a 7.5-fold increase in accuracy. The frequency of manual intervention was reduced from 30 times / shift to ≤5 times / shift, a reduction of 83% in human involvement; Energy consumption per unit area decreased from 5.8 kWh / m² to 4.2 kWh / m², a reduction of 27.6%. The defect rate decreased from 3.2% to 0.08%, a reduction of 97.5%.

[0062] The examples described herein are merely illustrative, intended to explain some features of the methods described herein. The appended claims are intended to claim the broadest possible scope, and the embodiments presented herein are merely illustrative of selected implementations based on combinations of all possible embodiments. Therefore, the applicant intends that the appended claims are not limited by the selection of examples illustrating the features of the invention. Some numerical ranges used in the claims also include sub-ranges within them, and variations within these ranges should be interpreted, where possible, as covered by the appended claims.

Claims

1. A process optimization method for a coating production line, characterized in that, Includes the following steps: S1. Construct a digital twin model of the film coating production line and synchronize physical production line data in real time; S2. Train the AI ​​decision-making model to dynamically optimize coating speed, temperature, and pressure parameters; S3 controls automated equipment to execute process instructions and provides online calibration feedback.

2. The process optimization method for a coating production line according to claim 1, characterized in that, In S1, the digital twin model is constructed in the following way: Laser scanning acquires 3D point cloud data of the production line; Embedded multiphysics simulation engine; Real-time mapping of device status data.

3. The process optimization method for a coating production line according to claim 1, characterized in that, In S2, the AI ​​decision-making model is based on a deep reinforcement learning architecture, including: State space: coating speed, adhesive viscosity, substrate temperature, equipment load; Motion space: The parameter adjustment range, with an adjustment base value of ±10%; Reward function: R = 0.6 × quality score + 0.3 × efficiency score - 0.1 × energy consumption score.

4. The process optimization method for a coating production line according to claim 1, characterized in that, In S3, the automated equipment includes a six-axis robotic arm, an AGV transportation system, and an online inspection unit.

5. The fully automated steel die replacement method for carat tube production according to claim 1, characterized in that, S3 also includes an exception handling mechanism, which includes: Real-time monitoring of coating pressure fluctuations; triggering the die head self-cleaning program when the deviation is > ±5%. After AI identifies coating defects, it automatically segments and marks abnormal segments.

6. The process optimization method for a coating production line according to claim 1, characterized in that, In S3, online feedback calibration includes: Digital twins are compared with actual production data to generate error compensation parameters; Update the weights of the AI ​​decision-making model and optimize the instructions for the next cycle.

7. The process optimization method for a coating production line according to claim 1, characterized in that, This method supports rapid changeover across multiple specifications: RFID identifies pipe specifications and automatically retrieves the corresponding process parameter library; The digital twin simulates the model change path, and the robotic arm completes the model change within 30 minutes.

8. The process optimization method for a coating production line according to claim 1, characterized in that, This method employs a federated learning framework, including: Each production line trains its model locally and then uploads the gradient parameters to the cloud in encrypted form. The global model is aggregated, updated weekly, and distributed to all nodes.

9. The process optimization method for a coating production line according to claim 1, characterized in that, The method also includes an energy consumption optimization module: Digital twins simulate energy consumption curves under different parameters, and AI recommends the lowest energy consumption combination; The waste heat recovery system uses the waste heat from the drying section for substrate preheating.

10. The process optimization method for a coating production line according to claim 1, characterized in that, This method is applicable to the following scenarios: High abrasion resistant mining corrugated tubing with TPU coating; Photocatalytic self-cleaning drainage pipe TiO coating.