A product quality inspection system for thermally insulated color steel sandwich composite panels based on digital twins.

Abstract

This invention discloses a product quality inspection system for thermally insulated color steel sandwich composite panels based on digital twins, specifically in the field of quality control. The system includes a full-dimensional three-dimensional digital model construction module for the composite panel, a production digital twin creation module, a real-time data acquisition module for all elements, a data preprocessing module, a twin data-driven and model calibration module, a quality intelligent analysis and decision-making module, and an optimization execution module. The model construction module builds a benchmark model and embeds process standards and quality thresholds; the twin creation module builds a virtual production line, standardizes data interfaces, and assigns IDs to the panels; the data acquisition module collects real-time data; the preprocessing module removes outliers, standardizes data, and correlates it; the calibration module updates the twin, calculates deviations, and calibrates the model; the analysis module performs compliance judgment, defect analysis, and quality prediction; and the optimization module generates and executes instructions, solving the problems of low efficiency, fragmented data, and lack of closed-loop processing in traditional testing, thereby improving the accuracy and efficiency of quality control.

Description

Technical Field

[0001] This invention relates to the field of quality control technology, and more specifically, to a product quality inspection system for thermally insulated color steel sandwich composite panels based on digital twins. Background Technology

[0002] Insulated color steel sandwich composite panels, as lightweight, heat-insulating, and easy-to-install building materials, are widely used in industrial plants, temporary buildings, cold chain warehouses, and other scenarios. Their quality directly determines the structural safety, heat insulation performance, and service life of the building. The core quality indicators of the composite panels include the flatness of the color steel panel, the uniformity of the sandwich layer thickness, the bonding strength of the adhesive layer, the heat insulation coefficient of the sandwich layer, and the area of ​​appearance defects. These indicators need to be strictly controlled during the production process.

[0003] Currently, the industry mainly relies on three traditional methods for quality control of composite panels: First, manual sampling and offline measurement, such as taking 3-5 samples from each batch and sending them to the laboratory to test the bonding strength, and measuring the thickness of the sandwich layer by manual calipers; Second, the application of single-point automated testing equipment, with some companies introducing laser thickness gauges to monitor the thickness of the sandwich layer and surface imagers to identify surface scratches; Third, the initial exploration of digital twin technology.

[0004] However, in practical use, it still has some drawbacks. For example, the sample size of manual sampling in existing technologies is small, which cannot represent the quality of the entire batch of products and is prone to missing hidden defects; the offline testing process is time-consuming, and by the time the test results are issued, the unqualified products have already been mass-produced, resulting in waste of raw materials and labor time and increased production costs; the production parameters of existing technologies are stored in the PLC control system, the quality inspection data is recorded in an independent quality inspection terminal, and the environmental parameters are collected by scattered sensors. The data are stored in isolation and there is no correlation mechanism. When quality defects occur, production and environmental parameters need to be checked manually one by one, which cannot quickly locate the root cause and the investigation cycle can be as long as several hours; existing testing equipment can only output a binary judgment result of pass / fail, which cannot be based on historical data. Historical data can predict quality trends, but it cannot automatically feed adjustment instructions back to the production line. For example, when the thickness deviation of the sandwich layer is detected to exceed the limit, the operator needs to adjust the opening of the feed valve based on experience. The adjustment accuracy is low, and the same problem is prone to recurrence. It is impossible to form a closed loop of detection-analysis-adjustment-verification. The composite panel is a three-layer heterogeneous structure of color steel panel, adhesive layer and sandwich layer. The traditional twin modeling method of single part cannot accurately represent the physical properties of each layer at the same time. The production data is multi-source and heterogeneous and lacks a unified interface, so the virtual model cannot reflect the physical production line status in real time. There is no dedicated simulation model for the quality defects of composite panels. It is impossible to reproduce defects and verify the root cause in the virtual scene. The digital twin only stays at the visualization level and has not played a control value role. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a product quality inspection system for thermally insulated color steel sandwich composite panels based on digital twins, which solves the problems mentioned in the background art through the following solutions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a product quality inspection system for thermally insulated color steel sandwich composite panels based on digital twins, comprising a three-dimensional digital model construction module for the composite panel: used to construct an integrated benchmark model of the composite panel;

[0007] Digital twin creation module: Constructs dynamic digital twins based on an integrated baseline model;

[0008] Real-time data acquisition module: Used to collect full-dimensional data through a distributed acquisition network, providing data support for twin calibration and quality analysis;

[0009] Data preprocessing module: Used to preprocess raw full-dimensional data at edge computing nodes to eliminate noise and outliers;

[0010] Twin data-driven and model calibration module: used to update the dynamic digital twin Mt through real-time data, calculate the virtual-real deviation and perform dynamic calibration;

[0011] The calculation and calibration method for the virtual-real deviation is as follows:

[0012] Virtual prediction: Based on the current twin, run the simulation module to predict the quality indicators of the composite board, including predicting the bond strength σpre and the thermal insulation coefficient Kpre;

[0013] Deviation calculation: Calculate the absolute deviation between the predicted value and the measured value. The formula is:

[0014] , Where Δσ is the bond strength deviation and ΔK is the thermal insulation coefficient deviation;

[0015] Model calibration: If the deviation exceeds the allowable range, calibrate the curing kinetic model parameters of the virtual adhesive layer. The initial model for the curing degree α of the virtual adhesive layer and the pressing temperature and pressing time is as follows: Where k is the initial baseline coefficient, a is the initial temperature influence coefficient, and b is the initial time influence coefficient. The calibration process uses the least squares method to fit historical data and update the model parameters. , , Where 0.1 is the parameter adjustment coefficient, and knew, anew, and bnew are the calibrated parameters;

[0016] Quality Intelligent Analysis and Decision Module: Used for quality analysis based on calibrated twins and outputting decision results;

[0017] The method for quality analysis is as follows:

[0018] S2.1 Automatic Compliance Determination: The measured quality indicators of physical testing are compared with the acceptable range defined by the composite board full-dimensional three-dimensional digital model construction module, and a veto system is used to determine the product grade.

[0019] Superior grade: All measured indicators are within the qualified range, and the deviation of the key indicators σactual and Kactual is ≤0.05 times the design value, that is, |σactual-σ0|≤0.05σ0, |Kactual-K0|≤0.05K0, where σactual is the measured bonding strength, Kactual is the measured thermal insulation coefficient, σ0 is the design bonding strength, and K0 is the design thermal insulation coefficient;

[0020] Qualified product: All measured indicators are within the qualified range, but the deviation of key indicators is >0.05 times the design value;

[0021] Non-conforming products: If any measured index exceeds the acceptable range, a quality judgment report containing ID, measured values ​​of each index, and grade will be generated for each composite board and synchronized to the twin, where it will be marked with different colors in the virtual scene.

[0022] S2.2 Defect Root Cause Analysis: When a product is determined to be non-conforming or conforming but with significant deviations in key indicators, the root cause is located through parameter correlation analysis and virtual simulation verification.

[0023] The parameter correlation analysis is as follows: The Pearson correlation coefficient r between the quality defect index and each production parameter is calculated to quantify the correlation. The formula is:

[0024] , where n is the number of samples, xi is the measured value of the i-th production parameter, yi is the measured value of the defect index of the corresponding sample, and the closer the absolute value of r is to 1, the stronger the correlation between the production parameter and the defect index.

[0025] The virtual simulation verification is as follows: In the twin, other parameters are fixed, and only the 2-3 production parameters with the strongest correlation are adjusted. The simulation module is run, and the defects of the virtual composite board are observed to be reproduced. If they are reproduced, the parameter is confirmed as the root cause.

[0026] S2.3 Quality Trend Prediction: Based on historical data, an LSTM model is trained to predict the quality trend for the next 1-2 hours, providing early warning of potential defects.

[0027] Model input: Time-series production parameter data divided into 10-minute time windows, with each time window containing 600 data points;

[0028] Model output: Predicted mean and fluctuation range of key quality indicators σ, K, and Δd3 for the next hour, where σ is the bonding strength, K is the thermal insulation coefficient, and Δd3 is the uniformity of the sandwich layer thickness.

[0029] Prediction accuracy measurement: The root mean square error (RMSE) is used to measure the deviation between the predicted and measured values. The formula is: Where yi is the i-th measured quality index value. The corresponding predicted value is n, which is the number of predicted samples. The model training objective is RMSE < 0.03 × design value.

[0030] Warning Trigger: If the predicted quality index fluctuation range reaches the qualified threshold boundary, a warning will be triggered, and a prompt message indicating insufficient potential bond strength will be output.

[0031] Optimization Execution Module: Generates optimization instructions based on the analysis and decision results of the quality intelligent analysis and decision module, and executes the optimization instructions as production adjustment actions.

[0032] The technical effects and advantages of this invention are as follows:

[0033] This invention enables real-time data collection across the entire production process through a distributed acquisition network. Each composite board is assigned a unique RFID tag, allowing for one-to-one inspection. The inspection covers all core quality indicators, improving inspection efficiency. Data is transmitted to the digital twin in real time, eliminating the need for offline waiting. Defective products can be identified instantly during the composite board production process, avoiding batch waste.

[0034] This invention cleans and standardizes multi-source data through a data preprocessing module, and binds the entire lifecycle data of the same composite board based on RFID tags and stores it in a unified database; the quality intelligent analysis and decision-making module can directly call the associated data, calculate the correlation between quality defects and production parameters through Pearson correlation coefficient, and combine virtual simulation verification to shorten the root cause location time from several hours to 5-10 minutes and improve the location accuracy.

[0035] This invention trains an LSTM model based on historical data, which can predict the trend and fluctuation range of key quality indicators in the next 1-2 hours, enabling proactive prevention. The optimization execution module outputs two types of instructions: one type of instruction is automatically sent to the PLC through the OPCUA interface to control the equipment to adaptively adjust parameters; the other type of instruction pushes an optimization report containing root causes, measures, and expected effects to the operator. After execution, the effect is verified by comparing the quality data before and after the adjustment through a twin. The qualified data is stored in the knowledge base to iterate the model, continuously improving the prediction accuracy and forming a complete closed loop.

[0036] This invention employs layered parametric modeling, constructing and fusing sub-models for the color steel panel, adhesive layer, and sandwich layer respectively, to accurately characterize the physical properties of each layer. It achieves unified mapping of multi-source heterogeneous data through the OPCUA standardized interface, and combines virtual-real deviation calculation and model calibration to ensure high-fidelity consistency between the digital twin and the physical production line. It also includes a built-in simulation model for composite panel defects, which can reproduce issues such as thickness deviation and appearance defects in a virtual scene, verify the root cause, and upgrade the digital twin from visualization to a control tool, adapting to the needs of composite panel production lines. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the overall structure of the present invention;

[0038] Figure 2 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0039] 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.

[0040] As attached Figure 1-2 The digital twin-based product quality inspection system for insulated color steel sandwich composite panels shown includes:

[0041] Composite Panel Full-Dimensional 3D Digital Model Building Module: Used to build an integrated benchmark model of the composite panel;

[0042] The method for constructing the integrated benchmark model M0 is as follows:

[0043] Layered modeling: For the three-layer structure of the composite panel (A), adhesive layer (B), and core layer (C), a parametric modeling method is used to construct sub-models for each layer.

[0044] Color steel panel (A): Define geometric parameters (length L1, width W1, thickness d1) and physical properties (elastic modulus E1, Poisson's ratio μ1, thermal conductivity λ1).

[0045] Adhesive layer (B): Define geometric parameters (thickness d2) and physical properties, including design bond strength σ0 and curing temperature range Tsolid.

[0046] Sandwich layer (C): Define the geometric parameters thickness d3 and physical properties including design thermal insulation coefficient K0 and design density ρ0. The reason for using parametric modeling is that composite board orders need to be customized in size (such as length L1 can be set to 6m, 9m or 12m). By modifying the geometric parameters, a new specification digital model can be quickly generated to adapt to flexible production.

[0047] The process standards embedded in the integrated benchmark model are as follows: Ideal process parameters (qualified production baseline) are associated with the benchmark model M0. These ideal process parameters include:

[0048] Feeding stage: tension of color steel coil F0 (2-3kN), design moisture content of core layer raw material ω0 (≤5%);

[0049] Pressing stage: Design pressing temperature T0 (12-15℃), design pressing pressure P0 (1.5-2.MPa), design pressing time t0 (3-6s);

[0050] Forming stage: Design conveying speed V0 (1-1.5m / s), design cooling temperature T_cooling0 (25-3℃).

[0051] Quality threshold definition: The acceptable range of each quality indicator is embedded in the baseline model M0 as the basis for subsequent quality judgment.

[0052] The flatness of the color steel panel ΔL (the deviation between the measured flatness and the design value): the acceptable range is ΔL∈[-1mm,+1mm];

[0053] The uniformity of the sandwich layer thickness Δd3 (the deviation between the measured thickness and the designed thickness d3): the acceptable range is Δd3∈[-0.5mm,+0.5mm];

[0054] Bond strength σ (measured value): The acceptable range is σ≥0.8σ0;

[0055] Thermal insulation coefficient K (measured value): The acceptable range is K∈[0.95K0,1.5K0];

[0056] The area of ​​the appearance defect, S_defect (measured value): The acceptable range is S_defect ≤ 5mm².

[0057] Digital twin creation module: Constructs dynamic digital twins based on an integrated baseline model;

[0058] The method for constructing the dynamic digital twin is as follows:

[0059] S1.1 Virtual Production Line Setup: In a digital twin platform (such as Unity or TwinBuilder), a virtual scene is built according to the layout of the physical production line (feeding unit → pressing unit → forming unit → inspection unit → stacking unit), including virtual equipment (virtual roller press, virtual laser detector), virtual materials (digital model of composite board), and virtual environment (virtual workshop temperature Tvirtual, virtual workshop humidity Hvirtual).

[0060] S1.2 Data Interface Standardization: Develop standardized data interfaces (using the OPCUA protocol) for PLCs, sensors, and testing instruments in the physical production line, enabling the twin Mt to read real-time data from physical equipment (such as the measured pressing temperature Tactual and measured pressing pressure Pactual of the roller press) through the interface, ensuring the real-time performance and compatibility of data transmission.

[0061] S1.3 Unique Identification Assignment: Assign a unique identification ID (carried by an RFID tag) to each composite board entering the production line. At the same time, assign the same ID to the corresponding virtual composite board in the twin Mt to achieve the binding of physical board and virtual board. The reason for choosing RFID tags is that they can withstand the high temperature of 150℃ during the pressing stage and can be quickly identified by the readers at each production node to update the board position and status data in real time.

[0062] Real-time data acquisition module: Used to collect full-dimensional data through a distributed acquisition network, providing data support for twin calibration and quality analysis;

[0063] The data acquisition objects and functions of each device in the distributed data acquisition network are as follows:

[0064] Laser thickness gauge: Deployed at the core layer feed inlet, it collects the actual thickness d3 of the core layer with a measurement accuracy of ±0.01mm and a collection frequency of once per board;

[0065] Moisture content sensor: Deployed at the outlet of the core layer raw material silo, it collects the measured moisture content ωactual of the core layer raw material, with a collection frequency of once per roll of raw material;

[0066] Thermocouple sensor: installed on the surface of the pressing roller of the roller press, to collect the measured pressing temperature Tactual, with a sampling frequency of 10 times / second;

[0067] Pressure sensor: installed in the hydraulic system of the roller press, to collect the measured pressing pressure P_actual, with a sampling frequency of 10 times / second;

[0068] Encoder: Installed on the end of the conveyor roller motor shaft, it collects the actual conveying speed Vactual at a frequency of 5 times / second;

[0069] 3D laser scanner: Deployed at the outlet of the molding unit, it collects the actual flatness ΔL of the composite board, with a scanning speed of 500,000 points / second;

[0070] Industrial camera: Deployed at the appearance inspection station in conjunction with a ring light source to capture images of the composite board's appearance. The actual area of ​​the appearance defect, S_defect, is calculated using an image recognition algorithm. The camera has a resolution of 20 megapixels and a frame rate of 15fps.

[0071] Infrared thermal imager: Deployed in the detection unit, it collects the surface temperature distribution of the composite board, calculates the measured thermal insulation coefficient Kactual, and has a detection accuracy of ±0.001 W / (m·K).

[0072] Human-Machine Interface (HMI): Deployed at the workstation, it records operator ID and parameter adjustment records (such as manually adjusted pressing temperature values). The data recording method is event-triggered (automatically recorded when an adjustment operation occurs).

[0073] Data preprocessing module: Used to preprocess raw full-dimensional data at edge computing nodes, eliminating noise and outliers, specifically including:

[0074] Outlier removal: The 3σ criterion is used. If the measured value x_actual of a parameter exceeds the interval [μ-3σ, μ+3σ] (where μ is the historical mean of the parameter and σ is the historical standard deviation of the parameter), it is identified as an outlier and removed. For example, the historical mean of the compression pressure P_actual is μ=1.8MPa and the standard deviation is σ=0.1MPa. If the measured value of P_actual is 10MPa, which exceeds the interval [1.5MPa, 2.1MPa], it is identified as an outlier and removed.

[0075] Data standardization: Converting parameters of different dimensions to the [0,1] interval using standardization formulas;

[0076] Data Association: By using the unique identification ID of the composite board, the material data (d3 real, ω real), equipment data (T real, P real), and quality data (ΔL real, S missing real, K real) corresponding to the same ID are bound together to form a full life cycle data archive for a single board.

[0077] Twin data-driven and model calibration module: used to update the dynamic digital twin Mt through real-time data, calculate the virtual-to-real deviation and dynamically calibrate to ensure high-fidelity consistency between Mt and the physical production line Pt.

[0078] The data mapping rules for the twin data-driven and model calibration module are as follows:

[0079] Based on the principle of ID matching and parameter correspondence, the preprocessed real-time data is mapped to the twin Mt:

[0080] Geometric parameter mapping: The measured flatness ΔL and measured length L1 collected by the 3D laser scanner are updated to the corresponding geometric parameters of the virtual composite plate model;

[0081] Physical property mapping: The measured moisture content ωactual collected by the moisture content sensor is used to correct the density of the virtual core layer model—because an increase in moisture content will lead to an increase in the density of the core layer, the correction formula is: , where ρvirtual is the calibrated density of the virtual sandwich layer, ρ0 is the designed density of the sandwich layer, and 0.02 is the experimentally measured moisture content-density correlation coefficient;

[0082] Process parameter mapping: The actual pressing temperature Tactual and actual pressing pressure Pactual collected by the PLC are synchronized to the operating parameters of the virtual roller press, so that the temperature and pressure of the virtual roller press are consistent with those of the physical equipment.

[0083] The calculation and calibration method for the virtual-real deviation is as follows:

[0084] Virtual prediction: Based on the current twin Mt, run the simulation module to predict the quality indicators of the composite board, including the predicted bond strength σpre and the predicted thermal insulation coefficient Kpre;

[0085] Deviation calculation: Calculate the absolute deviation between the predicted value and the measured value. The formula is:

[0086] , Where Δσ is the bond strength deviation and ΔK is the thermal insulation coefficient deviation;

[0087] Model calibration: If the deviation exceeds the allowable range (e.g., Δσ>0.05σ0, where σ0 is the design bond strength), then calibrate the curing kinetic model parameters of the virtual bond layer. The initial model for the degree of curing αpre of the virtual bond layer (degree of curing is positively correlated with bond strength, σpre=σ0×αpre) and the pressing temperature and pressing time is as follows: Where k is the initial base coefficient, a is the initial temperature influence coefficient, and b is the initial time influence coefficient. The calibration process uses the least squares method to fit historical data (historical measured pressing temperature T_history, historical measured pressing time t_history, and historical measured degree of cure α_history) and updates the model parameters. , , , where 0.1 is the parameter adjustment coefficient (verified through a large number of experiments to balance calibration speed and stability), k new, a new, and b new are the calibrated parameters, which can be substituted into the curing degree model to make the next predicted σ prediction closer to σ actual.

[0088] Quality Intelligent Analysis and Decision Module: Used for quality analysis based on calibrated twins and outputting decision results;

[0089] The method for quality analysis is as follows:

[0090] S2.1 Automatic Compliance Determination: The measured quality indicators of physical testing are compared with the acceptable range defined by the composite board full-dimensional three-dimensional digital model construction module, and a veto system is used to determine the product grade.

[0091] Superior grade: All measured indicators are within the qualified range, and the deviation of key indicators (σactual, Kactual) is ≤0.05 times the design value (i.e., |σactual-σ0|≤0.05σ0, |Kactual-K0|≤0.05K0), where σactual is the measured bonding strength and Kactual is the measured thermal insulation coefficient;

[0092] Qualified product: All measured indicators are within the qualified range, but the deviation of key indicators is >0.05 times the design value;

[0093] Non-conforming products: Any measured index exceeds the acceptable range (e.g., ΔLactual = 1.2mm > 1mm, or σactual = 0.7σ0 < 0.8σ0). After judgment, a quality judgment report containing ID, measured values ​​of each index, and grade is generated for each composite board and synchronized to the twin entity Mt. In the virtual scene, it is marked with different colors (green = superior product, yellow = qualified product, red = non-conforming product).

[0094] S2.2 Defect Root Cause Analysis: When a product is determined to be non-conforming or conforming but with significant deviations in key indicators, the root cause is located through parameter correlation analysis and virtual simulation verification.

[0095] The parameter correlation analysis is as follows: The Pearson correlation coefficient r between the quality defect index and each production parameter is calculated to quantify the correlation. The formula is:

[0096] , where n is the number of samples, xi is the measured value of the i-th production parameter (such as T_actual, P_actual), yi is the measured value of the defect index of the corresponding sample (such as Δd3_actual), and the closer the absolute value of r is to 1, the stronger the correlation between the production parameter and the defect index.

[0097] The virtual simulation verification is as follows: In the twin Mt, other parameters are fixed, and only the 2-3 production parameters with the strongest correlation are adjusted (such as reducing T from 120℃ to 100℃). The simulation module is run, and the defects of the virtual composite board are observed to see if they are reproduced (such as whether Δd3 virtual increases from 0.3mm to 1.2mm). If they are reproduced, the parameter is confirmed as the root cause.

[0098] S2.3 Quality Trend Prediction: Based on historical data (production parameters and quality indicators from the past 3 months), an LSTM (Long Short-Term Memory) model is trained to predict quality trends for the next 1-2 hours, providing early warnings of potential defects.

[0099] Model input: Time series production parameter data divided into 10-minute time windows (such as T_real, P_real, V_real, ω_real, T_loop_real, H_real), with each time window containing 600 data points (10 times / second × 60 seconds × 10 minutes).

[0100] Model output: Predicted mean and fluctuation range of key quality indicators (σ, K, Δd3) for the next hour;

[0101] Prediction accuracy is measured using the root mean square error (RMSE), which measures the deviation between the predicted and measured values. The formula is as follows: Where yi is the i-th measured quality index value. The corresponding predicted value is n, which is the number of predicted samples. The model training objective is RMSE < 0.03 × design value (to ensure prediction accuracy).

[0102] Warning Trigger: If the predicted quality index fluctuation range touches the qualified threshold boundary (e.g., the predicted minimum value of σ = 0.81σ0, close to the unqualified line of 0.8σ0), an warning will be triggered, and a prompt message indicating insufficient potential bonding strength will be output.

[0103] Optimization Execution Module: Generates optimization instructions based on the analysis and decision results of the quality intelligent analysis and decision module, and executes the optimization instructions as production adjustment actions;

[0104] The method for generating the optimization instructions is as follows: Based on the analysis results of the quality intelligent analysis and decision-making module, optimization instructions are generated in two categories:

[0105] Type 1 instruction (adaptive adjustment): For simple linear deviations (such as Tactual = 130℃, which is lower than the design value of 140℃, and Tactual is confirmed to be the root cause), the system automatically generates parameter adjustment instructions (such as raising the pressing temperature to 145℃, holding it for 5 minutes, and then lowering it back to 140℃), without the need for manual intervention;

[0106] Type II instructions (human decision support): For complex nonlinear problems (such as the simultaneous existence of Δd3 actual deviation and S missing actual deviation, the root cause being uneven particle size of the sandwich layer raw material + uneven pressure distribution of the roller press), the system generates an optimization suggestion report, including a problem description, root cause analysis, suggested measures (such as changing the sandwich layer raw material supplier and performing dynamic balancing calibration of the roller press), and expected results (such as Δd3 actual deviation can be controlled within ±0.3mm after adjustment, and S missing actual deviation ≤30mm²), and pushes it to the operator's HMI terminal;

[0107] The method for executing the optimization instructions is as follows:

[0108] Command issuance: One type of command is directly issued to the production line PLC via the OPCUA interface to control the corresponding equipment to perform adjustments (e.g., the temperature control module of the roller press receives a temperature adjustment command and adjusts the power of the electric heating tube to raise the temperature from 130℃ to 145℃); the other type of command is manually triggered and executed via the HMI terminal after being confirmed by the operator (e.g., starting the roller press motor balancing calibration program).

[0109] Effect data acquisition: After the command is executed, the acquisition network continuously collects the adjusted production parameters (such as the adjusted T actual and P actual) and quality indicators (such as the adjusted σ actual and Δd3 actual) and synchronizes them to the twin;

[0110] Verification and Iteration: The digital twin engine compares the quality data before and after adjustment (e.g., σactual = 0.7σ0 before adjustment, σactual = 0.95σ0 after adjustment) to determine whether the optimization effect meets expectations (e.g., whether σactual ≥ 0.8σ0). If it meets expectations, the data is stored in the knowledge base for subsequent iterative training of the LSTM model. If it does not meet expectations, the system returns to the quality intelligent analysis and decision-making module to re-perform root cause analysis until the problem is solved, forming a complete closed loop.

[0111] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0112] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A heat-insulating color steel sandwich composite board product quality detection system based on digital twinning, characterized in that, The method comprises the following steps: A composite board full-dimensional three-dimensional digital model construction module is used to construct an integrated benchmark model of the composite board; A production digital twin creation module is used to construct a dynamic digital twin based on the integrated benchmark model; A full-element real-time data acquisition module is used to acquire full-dimensional data through a distributed acquisition network to provide data support for twin calibration and quality analysis; A data preprocessing module is used to preprocess the original full-dimensional data on the edge computing node to eliminate noise and outliers; A twin data driving and model calibration module is used to update the dynamic digital twin Mt through real-time data, calculate the virtual-real deviation, and dynamically calibrate; The calculation and calibration method of the virtual-real deviation is as follows: Virtual prediction: based on the current twin, run the simulation module to predict the quality indicators of the composite board, including predicting the bonding strength σpre and the thermal insulation coefficient Kpre; Deviation calculation: calculate the absolute deviation of the predicted value and the measured value, the formula is: , where Δσ is the deviation in bond strength and ΔK is the deviation in thermal insulation coefficient; Model calibration: if the deviation exceeds the allowable range, calibrate the curing kinetics model parameters of the virtual adhesive layer, the curing degree of the virtual adhesive layer αpre and the initial model of the pressing temperature and pressing time are: , where k is the initial basic coefficient, a is the initial temperature influence coefficient, and b is the initial time influence coefficient. The calibration process fits the historical data by least squares method to update the model parameters: , , , where 0.1 is the parameter adjustment coefficient, knew, anew, and bnew are the calibrated parameters; A quality intelligent analysis decision module is used to perform quality analysis based on the calibrated twin and output a decision result; The quality analysis method is as follows: S2.1 Compliance automatic judgment: compare the measured quality indicators of physical detection with the qualified range defined by the composite board full-dimensional three-dimensional digital model construction module, and use a one-vote veto system to determine the product grade: Excellent product: all measured indicators are within the qualified range, and the deviation of key indicators σreal and Kreal is less than or equal to 0.05 times the design value, i.e., |σreal-σ0|≤0.05σ0 and |Kreal-K0|≤0.05K0, wherein σreal is the measured bonding strength, Kreal is the measured thermal insulation coefficient, σ0 is the designed bonding strength, and K0 is the designed thermal insulation coefficient; Qualified product: all measured indicators are within the qualified range, but the deviation of the key indicators is greater than 0.05 times the design value; Unqualified product: any one measured indicator exceeds the qualified range, a quality judgment report containing ID, measured values of each indicator, and grade is generated for each composite board, and is synchronized to the twin, which is marked with different colors in the virtual scene; S2.2 Defect root cause analysis: when it is determined to be an unqualified product or a qualified product but with a large deviation of key indicators, the root cause is located through parameter correlation analysis and virtual simulation verification: The parameter correlation analysis is as follows: calculate the Pearson correlation coefficient r of the quality defect indicators and each production parameter to quantify the correlation, and the formula is: Wherein n is the sample number, xi is the measured value of the i th production parameter, yi is the measured value of the defect index of the corresponding sample, and the absolute value of r is closer to 1, the stronger the correlation between the production parameter and the defect index. The virtual simulation verification is as follows: in the twin, fix other parameters, adjust only the 2-3 production parameters with the strongest correlation, run the simulation module, and observe whether the virtual composite board defect is reproduced, if reproduced, the parameter is confirmed as the root cause; S2.3 Quality trend prediction: based on historical data, an LSTM model is trained to predict the quality trend in the next 1-2 hours to early warn potential defects: Model input: time series production parameter data divided by 10-minute time windows, each time window contains 600 data points; Model output: predicted mean and fluctuation range of key quality indicators σ, K, and Δd3 in the next 1 hour, wherein σ is the bonding strength, K is the thermal insulation coefficient, and Δd3 is the core layer thickness uniformity; Prediction accuracy measurement: the root mean square error is used to measure the deviation between the predicted value and the measured value, the formula is: Where yi is the i-th measured quality index value, is the corresponding predicted value, n is the number of prediction samples, and the model training target is RMSE < 0.03 × design value; Warning Trigger: If the predicted quality index fluctuation range reaches the qualified threshold boundary, a warning will be triggered, and a prompt message indicating insufficient potential bond strength will be output. Optimization Execution Module: Generates optimization instructions based on the analysis and decision results of the quality intelligent analysis and decision module, and executes the optimization instructions as production adjustment actions.

2. The digital-twin-based heat-insulation color-steel sandwich composite board product quality detection system according to claim 1, characterized in that: The method for constructing the integrated benchmark model M0 is as follows: Layered modeling: For the three-layer structure of the composite panel color steel panel A, adhesive layer B, and sandwich layer C, a parametric modeling method is used to construct sub-models for each layer. Color steel panel A: Define geometric parameters including length L1, width W1, thickness d1 and physical properties including elastic modulus E1, Poisson's ratio μ1, thermal conductivity λ1; Adhesive layer B: Defines geometric parameters, thickness d2, and physical properties including design bond strength σ0 and curing temperature range Tsolid; Sandwich layer C: Defines geometric parameters, thickness d3, and physical properties including design thermal insulation coefficient K0 and design density ρ0; The process standard embedding of the integrated benchmark model is as follows: The ideal process parameters of the production process are associated with the benchmark model M0. The ideal process parameters include: Feeding stage: tension of color steel coil F0, design moisture content of core layer raw material ω0; Pressing stage: Design pressing temperature T0, design pressing pressure P0, design pressing time t0; Forming stage: Design conveying speed V0, design cooling temperature Tcooling0; Quality threshold definition: The acceptable range of each quality indicator is embedded in the baseline model M0 as the basis for subsequent quality judgment. The flatness of the color steel panel ΔL: the acceptable range is ΔL∈[-1mm,+1mm]; The uniformity of the sandwich layer thickness Δd3: the acceptable range is Δd3∈[-0.5mm,+0.5mm]; Bond strength σ: The acceptable range is σ≥0.8σ0; Thermal insulation coefficient K: The acceptable range is K∈[0.95K0,1.5K0]; The area of ​​the appearance defect, S_defect, is within acceptable limits: S_defect ≤ 5mm².

3. The digital-twin-based heat-insulation color-steel sandwich composite board product quality detection system according to claim 1, characterized in that: The method for constructing the dynamic digital twin is as follows: S1.1 Virtual Production Line Setup: In the digital twin platform, a virtual scene is built according to the layout of the physical production line, including virtual equipment, virtual materials, and a virtual environment; S1.2 Data Interface Standardization: Develop standardized data interfaces for PLCs, sensors, and testing instruments in physical production lines, enabling twins to read real-time data from physical equipment through the interfaces; S1.3 Unique Identifier Assignment: Assign a unique identifier (ID) to each composite board entering the production line, and assign the same ID to the corresponding virtual composite board in the twin entity Mt.

4. The digital-twin-based heat-insulation color-steel sandwich composite board product quality detection system according to claim 1, characterized in that: The data acquisition objects and functions of each device in the distributed data acquisition network are as follows: Laser thickness gauge: Deployed at the core layer feed inlet, it collects the actual thickness d3 of the core layer with a measurement accuracy of ±0.01mm and a collection frequency of once per board; Moisture content sensor: Deployed at the outlet of the core layer raw material silo, it collects the measured moisture content ωactual of the core layer raw material, with a collection frequency of once per roll of raw material; Thermocouple sensor: installed on the surface of the pressing roller of the roller press, to collect the measured pressing temperature Tactual, with a sampling frequency of 10 times / second; Pressure sensor: installed in the hydraulic system of the roller press, collecting the measured pressing pressure Preal, with a collection frequency of 10 times per second; Encoder: installed at the shaft end of the conveying roller motor, collecting the measured conveying speed Vreal, with a collection frequency of 5 times per second; 3D laser scanner: deployed at the outlet of the forming unit, collecting the measured flatness ΔLreal of the composite panel, with a scanning speed of 500,000 points per second; Industrial camera: deployed at the appearance detection station with a ring-shaped light source, collecting the appearance image of the composite panel, and calculating the measured appearance defect area Sdefact through image recognition algorithm, with a camera resolution of 20 million pixels and a frame rate of 15 fps; Infrared thermal imager: deployed at the detection unit, collecting the surface temperature distribution of the composite panel, and calculating the measured thermal insulation coefficient Kreal, with a detection accuracy of ±0.001 W / (m·K); Human-machine interaction terminal: deployed at the operator station, recording the operator ID and parameter adjustment record, with an event-triggered data recording method.

5. The digital-twin-based heat-insulation color-steel sandwich composite board product quality detection system according to claim 1, characterized in that: The data mapping rules of the twin data-driven and model calibration module are as follows: According to the ID matching and parameter corresponding principle, the preprocessed real-time data is mapped to the twin Mt: Geometric parameter mapping: the measured flatness ΔLreal and the measured length L1real collected by the 3D laser scanner are updated to the corresponding geometric parameters of the virtual composite panel model; Physical property mapping: The measured moisture content ωreal collected by the moisture content sensor is used to correct the density of the virtual sandwich model, and the correction formula is: where ρvirt is the calibrated density of the virtual sandwich, ρ0 is the design density of the sandwich, and 0.02 is the correlation coefficient between moisture content and density measured in the experiment; Process parameter mapping: the measured pressing temperature Treal and the measured pressing pressure Preal collected by the PLC are synchronized to the operating parameters of the virtual roller press, so that the temperature and pressure of the virtual roller press are consistent with those of the physical equipment.

6. The digital-twin-based heat-insulation color-steel sandwich composite board product quality detection system according to claim 1, characterized in that: The classification generation method of the optimization instruction is as follows: according to the analysis results of the quality intelligent analysis and decision module, the optimization instructions are generated in two categories: One type of instruction: for linear deviation, the system automatically generates parameter adjustment instructions without human intervention; The second type of instruction: for non-linear problems, the system generates an optimization suggestion report containing problem description, root cause analysis, suggested measures, and expected effect, and pushes it to the operator's HMI terminal.

7. The digital-twin-based heat-insulation color-steel sandwich panel product quality detection system according to claim 1, characterized in that: The execution method of the optimization instruction is as follows: Instruction issuance: one type of instruction is directly issued to the production line PLC through the OPCUA interface to control the corresponding equipment to execute the adjustment; the second type of instruction is executed manually through the HMI terminal after being confirmed by the operator; Effect data collection: after the execution of the instruction, the network continuously collects the adjusted production parameters and quality indicators, and synchronizes them to the twin; Verification and iteration: the digital twin engine compares the quality data before and after the adjustment to determine whether the optimization effect meets the expectation, if it meets the expectation, the data of this time is stored in the knowledge base for subsequent iteration training of the LSTM model; if it does not meet the expectation, the quality intelligent analysis and decision module is returned to perform root cause analysis again until the problem is solved.

Images