Intelligent self-adaptive control method of multi-size compatible oven processing system

Abstract

The invention provides an intelligent self-adaptive control method of a multi-size compatible oven processing system. The method comprises the following steps: collecting material size data, real-time state data in a processing process and environment data; material sizes are calculated, materials are classified, and a mapping relation between the sizes and process parameters is established in combination with material types and target processing requirements; analog simulation is conducted on the machining process under different technological parameters, and an optimal technological parameter set is predicted; matching adaptive process parameters in the process parameter set, and performing dynamic adjustment in combination with real-time state data; and generating a control instruction according to the matched or adjusted process parameters, driving execution equipment to act based on the control instruction, and feeding back execution effect dynamic optimization model parameters. The problems that the material state cannot be monitored in real time during heating, heating parameters need to be manually adjusted to adapt to materials of different sizes, and then the production efficiency and the material percent of pass are affected can be solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent oven control technology, and more particularly to an intelligent adaptive control method for a multi-size compatible oven processing system. Background Technology

[0002] Oven processing systems generate heat through electric heating elements or hot air circulation, uniformly heating materials using heat conduction, convection, and radiation. They are suitable for industrial or civilian applications such as food processing, electronic component curing, and material drying. However, existing technologies have the following limitations when heating materials: Firstly, when the size of the processed material changes, operators need to manually measure the material size and then manually adjust key parameters such as the position of the support rack inside the oven, the power distribution of the heating unit, the hot air circulation path, and the processing time. This operation is cumbersome and inefficient. When processing multiple batches of materials of different sizes continuously, manually adjusting the support structure and process parameters takes up a lot of production time, which seriously affects production efficiency. Furthermore, manual debugging relies on experience and judgment, which can easily lead to parameter setting deviations, resulting in a decrease in the material qualification rate. Secondly, the lack of real-time monitoring means in the processing process makes it impossible to detect abnormalities in the material status in a timely manner. If the actual size of the material deviates due to thermal expansion and contraction or changes in stacking state during the processing, the traditional system cannot sense and adjust the parameters in real time, which can easily lead to processing quality problems.

[0003] Therefore, there is an urgent need for an oven processing control system that can monitor the heating status of materials in real time and intelligently and dynamically adjust process parameters in combination with changes in material size, so as to improve production efficiency and material qualification rate. Summary of the Invention

[0004] This invention provides an intelligent adaptive control method for a multi-size compatible oven processing system, which solves the problems of not being able to monitor the material status in real time during heating and having to manually adjust heating parameters to adapt to materials of different sizes, thereby affecting production efficiency and material qualification rate.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: This invention provides an intelligent adaptive control method for a multi-size compatible oven processing system, comprising the following steps: S1: Collect material size data, real-time status data during processing, and environmental data, and preprocess the collected data; S2: Calculate material dimensions using a dynamic dimension recognition algorithm and classify the materials. Establish a mapping relationship between dimensions and process parameters by combining material type and target processing requirements. S3: Based on digital twin technology, simulate the processing under different process parameters and predict the optimal set of process parameters; S4: Based on the material size, the process parameters are automatically matched and adapted to the process parameters. The process parameters are then dynamically adjusted in combination with real-time status data during the processing. S5: Generate control commands based on the matched or adjusted process parameters, drive the execution of equipment actions based on the control commands, and dynamically optimize model parameters based on the execution effect.

[0006] Preferably, based on the calculated material dimensions and dimensional classification, and combining the material type with the target processing requirements, a mapping relationship between dimensions and process parameters is established, including the following steps: Define the specific dimensions of input features and output parameters, and construct a three-dimensional database of input features, output parameters, and quality results; Correlation analysis is performed on the input features to retain those that are strongly correlated with the output parameters and quality results. Redundant features are removed based on RFE to obtain the core input feature set. A hybrid model of gradient boosting tree and neural network is adopted to take into account both linear and nonlinear mapping relationships and predict output parameters based on input features; The predicted output parameters are corrected by taking into account the physical constraints, process constraints and environmental fluctuations of the actual processing scenario.

[0007] Preferably, the process constraints include size and parameter correlation constraints, material type constraints, and target requirement constraints.

[0008] S3 includes the following steps: Construct 3D models of the oven body, sensors, and materials, and use LOD technology to refine the core components; A multiphysics coupling model was constructed to simulate key processes in oven processing, and the virtual model parameters were calibrated through physical experiments to eliminate the deviation between simulation and reality. Establish a communication interface between the industrial IoT gateway and the digital twin platform to synchronize physical device data and virtual models in real time; Based on the material size, type, and target quality requirements, parameter boundaries are set according to the material characteristics, and Latin hypercube sampling is used to generate multiple sets of candidate parameters within the parameter range; The entire processing of candidate parameters is simulated one by one in the digital twin model, and detailed simulation data is output. The Analytic Hierarchy Process (AHP) was used to quantitatively score the simulation results of each set of candidate parameters based on evaluation indicators, and the set with the highest score was selected as the optimal parameter set.

[0009] Preferably, constructing a multiphysics coupled model to simulate key processes in oven processing includes the following steps: Define the heat-fluid, heat-mass, and heat-reaction coupling relationships for the oven and materials; Build a geometric model and mesh the oven body and materials; Import material properties and set initial and boundary conditions; Configure the solver for field decoupling and set the solution parameters; Key data collected during the oven processing are used as a benchmark to calibrate model parameters.

[0010] Preferably, the evaluation indicators include quality compliance, processing uniformity, processing efficiency, and energy consumption cost.

[0011] S4 includes the following steps: Construct a four-dimensional relational database of size, type, target requirements, and process parameters, and filter out a set of candidate parameters that meet the conditions through multi-condition retrieval; Calculate the dimensional deviations corresponding to the candidate parameters, and select the parameter with the smallest deviation as the basic parameter; Adjust the basic parameters based on real-time environmental data, and use the environmentally corrected parameters as the initial execution parameters. Continuously collect key status data during the processing, compare real-time data with theoretical data predicted by the digital twin model to calculate the deviation value, and trigger the parameter adjustment mechanism according to the preset deviation threshold; PID control algorithm is used to handle linear deviation, and fuzzy control algorithm is used to handle nonlinear and multi-factor coupled deviation, and the parameter adjustment is calculated. The dynamically adjusted parameters are converted into control commands that the device can execute, driving the actions of the execution components and monitoring the execution effect.

[0012] The execution equipment includes a zone heating unit, a hot air circulation unit, a load adjustment unit, and a material conveying unit; The partitioned heating unit uses PWM pulse width modulation and a current sensor feedback closed-loop control to distribute the heating power of the top, middle and bottom of the oven. The hot air circulation unit controls the fan speed and guide plate angle through frequency conversion control and stepper motor drive, respectively. The load-bearing adjustment unit is driven by a servo motor and adjusts the height of the load-bearing frame based on closed-loop control of displacement sensor feedback. The material conveying unit controls the start and stop of the conveyor belt and adjusts its speed via frequency conversion.

[0013] The beneficial effects of this invention are as follows: This invention utilizes a multi-sensor fusion-based dynamic size recognition algorithm to accurately calculate and classify the three-dimensional dimensions of materials within 3 seconds, adapting to both regular and irregular materials. During continuous processing of multiple batches and multi-size materials, it automatically matches process parameters with a switching time of ≤5 seconds, eliminating the need for manual adjustment of parameters such as the support frame position and heating power, thus solving the pain points of time-consuming and labor-intensive traditional manual adjustments. An XGBoost-NN hybrid model establishes a mapping relationship between dimensions and process parameters, combined with digital twin technology to simulate and predict the optimal parameter set, achieving a parameter matching accuracy of ≥95%. Based on a multi-physics coupling model and a PID+fuzzy control dynamic adjustment mechanism, it responds in real-time to changes in material thermal expansion and contraction, environmental fluctuations, etc., ensuring processing uniformity deviation ≤±2℃. Digital twin simulation reduces the cost of physical trial and error, while the dynamic adjustment mechanism avoids quality problems such as localized overheating and undercooking, effectively improving product quality and production efficiency. It achieves compatibility with multi-size oven processing systems and dynamic adjustment of process parameters. Attached Figure Description

[0014] Figure 1 This is a flowchart of an intelligent adaptive control method for a multi-size compatible oven processing system according to the present invention. Detailed Implementation

[0015] To further illustrate the technical means and effects of the present invention in achieving the intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0016] Please see Figure 1 As shown, the present invention relates to an intelligent adaptive control method for a multi-size compatible oven processing system, comprising the following steps: S1: Collect material size data, real-time status data during processing, and environmental data, and preprocess the collected data; Dimensional data: Two sets of horizontal laser rangefinders are installed on the left and right sides of the oven feed inlet to measure the material length L, with a measurement range of 0-50cm and an accuracy of ±0.1mm; two sets of vertical laser rangefinders are installed on the top and bottom of the oven feed inlet to measure the material height H, with a measurement range of 0-30cm and an accuracy of ±0.1mm; a matrix infrared rangefinder is installed above the support frame inside the oven to measure the material width W and the actual placement area S, in a 10×8 array (30cm spacing), with a measurement range of 0-40cm; a vision sensor (industrial camera) is also configured to acquire two-dimensional images of the material to assist in correcting the dimensional data.

[0017] When the material enters the feed inlet and the material blocks the sensor beam for ≥0.5s, the laser rangefinder is triggered to continuously collect 10 sets of L and H data. After the material is completely placed on the support frame, the matrix infrared rangefinder scans the surface of the material to obtain the distance between each measuring point and the sensor, and calculates W and the placement area S. The vision sensor simultaneously captures a top view of the material and extracts the contour features.

[0018] Processing status data: Temperature distribution sensors, infrared temperature sensors, humidity sensors, weight sensors, and current sensors are deployed to collect data on furnace zone temperature, material surface temperature, furnace humidity, material weight changes (reflecting moisture evaporation / curing progress), and equipment operating status (heating power, fan load). The collected data is linked to material batch numbers and timestamps for easy traceability and analysis.

[0019] Environmental data: Temperature and humidity are collected by environmental temperature and humidity sensors, and voltage fluctuations of the power grid are collected by voltage sensors.

[0020] For dimensional data, a Kalman filter algorithm was used to process 10 consecutively acquired laser / infrared data sets. Color images were converted to grayscale, Gaussian noise reduction was applied to smooth image noise, and the Canny algorithm was used to extract material contours, removing background interference and obtaining clear material edge features. For status and environmental data, outliers were removed, linear interpolation was used to complete the data, and normalization was performed. Since different sensors have different acquisition frequencies, all data needed to be aligned to a 1-second time axis. High-frequency data (e.g., 10Hz for laser ranging) was averaged within 1 second; low-frequency data (e.g., environmental data every 5 seconds) was extended to every 1 second using nearest-neighbor interpolation.

[0021] Based on a dual communication architecture combining CAN bus and Ethernet, real-time data (such as temperature and equipment status) is transmitted via CAN bus with a latency of ≤10ms, while batch data (such as visual images and historical parameters) is transmitted via Ethernet, ensuring real-time and stable data transmission. Collected data is uploaded to the local edge computing gateway in real time and simultaneously synchronized to the cloud database in JSON format, with fields including "data type, collection timestamp, sensor ID, material batch number, and value," facilitating cross-device access and long-term traceability.

[0022] S2: Calculate material dimensions using a dynamic dimension recognition algorithm and classify the materials. Establish a mapping relationship between dimensions and process parameters by combining material type and target processing requirements. Before calculating material dimensions using a dynamic dimension recognition algorithm, it is necessary to calibrate the sensor distance and installation spacing, as well as the image pixels. Calibration establishes a mapping relationship between sensor measurement data and actual data, serving as a baseline parameter for dimension calculation. The ratio coefficient between pixels and actual dimensions is then calculated. Calculations are performed based on 10 continuously acquired sets of dimension data. Length L = S_L - (mean L1 + mean L2); where S_L is the installation spacing of the horizontal laser sensors, L1 is the distance from the horizontal laser sensor 1 to the left end of the material, and L2 is the distance from the horizontal laser sensor 2 to the right end of the material.

[0023] Height H = S_H - (mean H1 + mean H2); where S_H is the installation spacing of the vertical laser sensors, H1 is the distance from vertical laser sensor 1 to the top of the material, and H2 is the distance from vertical laser sensor 2 to the bottom of the material.

[0024] Width W: The measurement points corresponding to the material edge are identified by clustering algorithm. The maximum horizontal distance between the edge measurement points is the width. Based on the infrared ranging data D_xy, measurement points with a distance ≤ the height of the support frame (i.e., measurement points blocked by the material) are selected. The x-coordinate (horizontal direction) of the blocked measurement points is clustered (using K-means algorithm, K=2, corresponding to the left and right edges of the material). The extreme values ​​of the x-coordinates (x_min, x_max) of the two clusters are taken to calculate the width W=(x_max-x_min)×sensor array spacing (30cm).

[0025] The volume of a regular material is V = L × W × H (directly multiplied by the three-dimensional dimensions); The volume V of irregular materials is calculated based on the visual image, specifically the pixel area S_pixel of the material outline (by counting the pixels of the outline), which is then converted into the actual placement area: S = S_pixel × pixel ratio factor. 2 (The scaling factor is the calibrated "actual size / number of pixels"), estimated volume: V = S × H × shape correction factor.

[0026] The calculated dimensions are verified and corrected by extracting the material's outline dimensions (L_vis, W_vis, H_vis) from the visual image. Based on the pixel ratio coefficient, the pixel length, width, and height of the outline are converted into actual dimensions, and the deviation is calculated. ΔL = |L - L_vis| / L × 100%; ΔW = |W - W_vis| / W × 100%; ΔH = |H - H_vis| / H × 100%; If ΔL, ΔW, and ΔH are all ≤3%, then the L, W, and H calculated by the distance measurement method are used directly as the final dimensions. If any deviation is greater than 3%, a weighted fusion algorithm is used for correction, with a weight of 0.7 for ranging data (high ranging accuracy) and a weight of 0.3 for visual data (good contour integrity). The formula is as follows: Finally, L = 0.7 × L + 0.3 × L_vis; Finally, W = 0.7 × W + 0.3 × W_vis; Finally, H = 0.7 × H + 0.3 × H_vis.

[0027] Materials are automatically categorized based on preset size ranges, with small materials having a volume (V) of less than 5L or a length (L×W×H) of less than 5000cm. 3 For medium-sized materials, 5L ≤ V ≤ 20L; for large-sized materials, V > 20L.

[0028] Based on the calculated material dimensions and dimensional classifications, and combining the material type with the target processing requirements, a mapping relationship between dimensions and process parameters is established, including the following steps: Define the specific dimensions of input features and output parameters, and construct a three-dimensional database of input features, output parameters, and quality results; Input features: Core size features, including three-dimensional dimensions (L / W / H), volume V, and size classification; Material type features, including material properties / attributes and thermophysical parameters (such as specific heat capacity c, thermal conductivity λ, and density ρ); Target processing requirements, including core quality indicators (such as food moisture content ≤15%) and auxiliary requirements (such as processing efficiency priority: high / medium / low; energy consumption limit: ≤5kWh / batch); Environmental correction features, i.e., room temperature / ambient humidity.

[0029] Output process parameters: Temperature parameters, including furnace set temperature T and zone heating power ratio (top P1, middle P2, bottom P3); Time parameters, including total processing time t and segmented heating time ratio (preheating section t1, constant temperature section t2, cooling section t3); Hot air parameters, including fan speed v and hot air guide plate angle θ; Bearing parameter is the support frame height h, which is adjusted according to the material height H to ensure that the material center coincides with the furnace temperature field center.

[0030] Collect processing data containing different sizes (small / medium / large), different material types, and different target requirements to ensure sample diversity, and label the quality results to clearly indicate whether the finished product is qualified, the achievement of core quality indicators (such as actual moisture content and actual curing degree), and abnormal situations (such as local overheating and incomplete curing), and establish a three-dimensional database.

[0031] Correlation analysis is performed on the input features to retain those that are strongly correlated with the output parameters and quality results. Redundant features are removed based on RFE to obtain the core input feature set. Continuous features (L, W, H, V, specific heat capacity, room temperature, etc.) are mapped to the [0,1] interval to avoid model weight bias due to differences in numerical ranges. Discrete features (material type, size classification, efficiency priority) are uniquely encoded to facilitate model identification of category differences. New effective features are generated based on existing features, such as "volume-thermal conductivity product (V×λ)" reflecting the overall heat conduction requirements of the material. The Pearson correlation coefficient between each input feature and core quality indicators (such as pass rate, degree of curing) is calculated, and features with an absolute correlation coefficient ≥ 0.6 are retained. Redundant features are eliminated using the RFE algorithm to finally determine the core input feature set.

[0032] A hybrid model combining gradient boosting tree (XGBoost) and neural network (NN) is adopted to balance linear and nonlinear mapping relationships and predict output parameters based on input features. Input layer: Receives the core input features after feature engineering processing; The first-level XGBoost basic model (coarse matching) is used to mine the linear and simple nonlinear relationships between input features and process parameters, output basic parameter combinations, and set hyperparameters, such as tree depth 5~8, learning rate 0.1~0.3, and number of iterations 100~200, to avoid model overfitting; the loss function is the quality result compliance rate (such as curing degree ≥98%), minimizing the proportion of non-compliant samples.

[0033] The second-level neural network model (fine-tuning): Based on the basic parameters output by XGBoost, fine-grained features are combined to optimize parameter accuracy. The network structure includes an input layer, hidden layers, and an output layer. The input layer receives the basic parameters and fine-grained features. Two hidden layers are used. The first hidden layer uses the ReLU activation function to alleviate gradient vanishing and improve fitting ability, and is combined with batch normalization (BN) layers to stabilize the training process. The second hidden layer further extracts the complex nonlinear relationship between features and parameters using the ReLU activation function. L1 and L2 regularization are added to the hidden layers to remove redundant weights and improve the model's generalization ability. A Dropout layer is embedded between the two hidden layers to suppress overfitting. The output layer uses the Sigmoid activation function to output the normalized parameter values. The mean squared error (MSE) between the actual and predicted parameters and the deviation of the quality results are used as dual objective functions to ensure parameter accuracy and quality compliance.

[0034] Fusion layer: The basic parameters of XGBoost and the optimized parameters of the neural network are fused by weighting (XGBoost weight 0.6, neural network weight 0.4) to output the final combination of prediction parameters.

[0035] The predicted output parameters are corrected by taking into account the physical constraints, process constraints and environmental fluctuations of the actual processing scenario.

[0036] Physical constraints: Set hard boundaries for output parameters to avoid exceeding the equipment's capacity range, such as furnace temperature T (50℃≤T≤300℃, oven rated temperature range); heating power ratio (power ratio between any two zones ≤1.2, to avoid local power overload leading to equipment damage); fan speed v (500r / min≤v≤3000r / min, fan rated speed range); processing time t (t≥5min, to avoid incomplete parameter execution due to short processing time, t≤180min, to avoid overprocessing).

[0037] Process constraints include: dimensional and parameter-related constraints; for large materials (V > 20L), the fan speed v ≥ 2000 r / min to enhance hot air circulation and avoid uneven heating; for small materials (V < 5L), the guide plate angle θ ≥ 50° to focus hot air and avoid heat waste. Material type constraints: for food materials, the maximum set temperature T ≤ 250℃ to avoid scorching; for electronic materials, the zone power ratio deviation ≤ 10% to avoid localized overheating and component damage. Target requirement constraints: for materials with high moisture content requirements (e.g., food moisture content ≤ 10%), the processing time t is extended by 10%~15%, and the fan speed v is increased by 5%~10% to accelerate moisture evaporation.

[0038] Environmental Correction: Predictive parameters are adjusted based on real-time environmental data to offset the impact of environmental fluctuations. For example, for every 5°C increase in room temperature (T) above the standard room temperature (25°C), the furnace set temperature (T) decreases by 3°C; for every 5°C decrease, T increases by 3°C; for every 10% increase in ambient humidity (RH) above the standard humidity (50% RH), the processing time (t) is extended by 5% (for processing scenarios requiring humidity control); for every 10% decrease in voltage fluctuation (ΔU) below the rated voltage (220V), the heating power is increased by 5% (to compensate for the decrease in heating efficiency caused by insufficient voltage).

[0039] S3: Based on digital twin technology, simulate the processing under different process parameters and predict the optimal set of process parameters; S3 includes the following steps: 3D models of the oven body, sensors, and materials are constructed, and LOD (Level of Detail) technology is used to refine the core components; A multiphysics coupling model was constructed to simulate key processes in oven processing, and the virtual model parameters were calibrated through physical experiments to eliminate the deviation between simulation and reality. The construction of a multiphysics coupling model to simulate key processes in oven processing includes the following steps: Define the heat-fluid, heat-mass, and heat-reaction coupling relationships for the oven and materials; Specifically, for the oven, it is necessary to simulate solid heat conduction (simulating heat generation from heating elements and heat transfer within the oven chamber), fluid flow (simulating the flow field between the hot air fan, guide plate, and oven chamber), and thermal radiation (simulating radiative heat transfer from the heating elements to the material). For the material, it is necessary to simulate unsteady-state heat transfer (simulating the internal temperature gradient of the material), mass transfer (simulating moisture migration within the material), and reaction kinetics (simulating the curing of electronic components / food baking reaction). The coupling relationships are defined as follows: Heat-flow coupling: Hot air flow carries heat to the material surface. The velocity distribution of the flow field directly affects the convective heat transfer coefficient. By setting up bidirectional coupling between the flow field and the temperature field, the flow field provides the convective heat transfer boundary for the temperature field. The temperature field affects the air density and thus changes the flow field. Thermo-mass coupling: The rise in material temperature drives moisture evaporation, and the moisture migration process is accompanied by latent heat exchange. By setting up temperature field-mass transfer field coupling, the temperature gradient serves as the driving force for moisture diffusion, and the moisture evaporation carries away heat to correct the temperature field. Thermal-reaction coupling: When the temperature reaches the reaction threshold, the curing / baking reaction is triggered. The exothermic reaction further affects the temperature field. Temperature field-reaction field coupling is set up so that the reaction rate changes with the temperature and the exothermic reaction acts as the internal heat source of the temperature field.

[0040] Build a geometric model and mesh the oven body and materials; Import oven CAD drawings, including the furnace chamber, heating elements (top / middle / bottom zones, 3 groups per zone), hot air fan (blade angle 30°), guide plate (adjustable 0°~90°), and support frame, with geometric errors ≤0.1%; simplify non-core structures while retaining the dimensions of core components. For regular materials, use SolidWorks to create cylindrical models; for irregular materials (such as irregularly shaped electronic components), use Blender to scan the physical outline, export it as an STL format, and import it into the simulation tool; annotate key material structures. Create tetrahedral meshes for the furnace chamber / hot air channel area, boundary layer meshes for the heating elements / material surfaces, and structured meshes for the material interior.

[0041] Import material properties and set initial and boundary conditions; For example, the heating element of the oven component is made of stainless steel, and the thermal property parameter of the material is thermal conductivity. Density 7850kg / m³ 3 The initial conditions are set as follows: oven initial temperature 25℃, material initial temperature 25℃ (food) / -18℃ (frozen food), and initial moisture content 45% (cake base).

[0042] The boundary conditions for the thermal field are set as follows: The power density of the heating element is set (e.g., 1000 W / m²). 2 Matching the rated power of the physical heating tube), furnace wall insulation boundary (heat flux = 0), and material surface convective heat transfer. (Provided by flow field simulation results), reconstructing the heat generation of the heating tube, the heat preservation of the furnace, and the heat transfer path of the hot air to the material; The boundary conditions of the flow field are set as follows: set the inlet wind speed of the fan (e.g., 5 m / s, matching the rated speed of the physical fan), set the guide plate angle (e.g., 45°, consistent with the physical adjustment), and set the furnace outlet pressure boundary to ensure that the flow field is consistent with the physical hot air circulation and avoid wind speed / pressure deviation from affecting heat transfer. The mass transfer boundary condition is set as follows: the material surface moisture evaporation boundary (the relative humidity is set to 60%, matching the actual humidity inside the oven), which simulates the diffusion process of material moisture into the hot air and is consistent with the actual moisture evaporation rate.

[0043] Configure the solver for field decoupling and set the solution parameters; Steady-state solution (preheating stage): First, run a steady-state simulation of pure heat conduction and flow field to solve the stable temperature distribution and hot air velocity distribution in the furnace, which serve as the initial conditions for transient simulation to avoid initial oscillations; Transient solution (processing stage): Use a piecewise solver to break down the thermal field, flow field, and mass transfer field into independent sub-step iterations: solve the flow field (fix the temperature field and calculate the wind speed distribution), the temperature field (calculate the temperature distribution based on the convective heat transfer coefficient provided by the flow field), the mass transfer field (calculate the moisture migration based on the gradient provided by the temperature field), and the reaction field (determine the reaction state based on the temperature field and calculate the reaction rate and heat release).

[0044] The initial time step is set to 1e-3s, and the maximum step is 0.1s. For transient problems involving thermal-mass coupling, an excessively large step can lead to errors in the calculation of water evaporation. A small step ensures that the rapid changes are captured. The upper limit of the number of iterations (nonlinear iteration) is set to 50 to improve the convergence probability. The relative error is set to 1e-4, and the absolute error is 1e-6 to balance the calculation accuracy and efficiency. Excessive errors in the thermal field / mass transfer field can lead to errors in temperature / water prediction. The damping coefficient is configured to be automatically damped. When enabled, it automatically reduces the iteration step when the values ​​oscillate, avoiding divergence in the solution. This is especially suitable for scenarios with intense exothermic reactions.

[0045] Key data collected during the oven processing are used as a benchmark to calibrate model parameters.

[0046] Establish a communication interface between the Industrial Internet of Things (IIoT) gateway and the digital twin platform to synchronize physical device data and virtual models in real time; The synchronized data includes physical equipment status data, material size / type data, and historical processing data, and is transmitted using CAN bus, Ethernet, and HTTP / HTTPS methods.

[0047] Based on the material size, type, and target quality requirements, parameter boundaries are set according to the material characteristics, and Latin hypercube sampling is used to generate multiple sets of candidate parameters within the parameter range; The entire processing of candidate parameters is simulated one by one in the digital twin model, and detailed simulation data is output. First, the simulation is initialized by loading the corresponding material model, inputting candidate parameters, and setting initial environmental conditions consistent with the physical processing environment. Then, a multi-physics co-simulation is performed: for temperature field simulation, the process of heat generation from the heating element → heat transfer from the hot air → heat absorption by the material is calculated, and a furnace temperature distribution cloud map is output every 10 seconds, recording the temperature changes on the material surface / interior; for flow field simulation, the path of hot air circulation from the fan outlet → guide plate → material surface is simulated, and the uniformity of wind speed on the material surface is output; for material reaction simulation, based on the reaction kinetic model, the degree of solidification (electronic components) or moisture content (food) of the material is calculated every 10 seconds, and the reaction progress curve is recorded. The temperature field, material state, and equipment operating data are displayed in real time on the virtual monitoring interface. If abnormalities such as localized overheating (material surface temperature > target temperature 5℃) or uneven solidification (internal temperature difference > 8℃) occur, the parameter is automatically marked as needing optimization, and the cause of the abnormality is recorded.

[0048] After the simulation, a "parameter-process-quality" related report is automatically generated, including process curves (temperature versus time curve, moisture evaporation rate curve, degree of cure curve), quality indicators (final degree of cure / moisture content, temperature uniformity deviation, virtual pass rate), and energy consumption data (total energy consumption during processing = heating power × time + fan power × time). Multiple GPUs can be used to simultaneously simulate 10-15 sets of candidate parameters, shortening the overall simulation time.

[0049] The Analytic Hierarchy Process (AHP) was used to quantitatively score the simulation results of each set of candidate parameters based on evaluation metrics, and the set with the highest score was selected as the optimal parameter set.

[0050] The evaluation indicators include quality compliance (i.e., the achievement rate of core quality indicators), processing uniformity (material surface temperature deviation, internal temperature gradient), processing efficiency (i.e., the total processing time compared to the target time), and energy consumption cost (i.e., the total energy consumption compared to the benchmark energy consumption). The comprehensive score of a single set of parameters is: indicator score = Σ (indicator score × weight).

[0051] The parameters are sorted in descending order of comprehensive score. The top 5 sets of parameters are selected, and outliers are removed for secondary simulation. The set of parameters with the highest comprehensive score and best stability after the secondary simulation is selected as the optimal process parameter set. At the same time, the top 3 sets of parameters are retained as a candidate set to deal with sudden equipment failures. The optimal parameters are applied to physical oven processing to verify the actual effect, and the feedback data optimizes the digital twin model, forming a closed loop.

[0052] S4: Based on the material size, the process parameters are automatically matched and adapted to the process parameters. The process parameters are then dynamically adjusted in combination with real-time status data during the processing. S4 includes the following steps: Construct a four-dimensional relational database of size, type, target requirements, and process parameters, and filter out a set of candidate parameters that meet the conditions through multi-condition retrieval; The database includes: a material size classification table including classification ID, volume range, size interval (L / W / H), and compatible material type; an optimal process parameter table including parameter ID, classification ID, material type, target requirements, temperature parameter (T / P1:P2:P3), time parameter (t / t1:t2:t3), hot air parameter (v / θ), and load parameter (h); and an environmental correction rule table including rule ID, room temperature range, humidity range, temperature correction value, and time correction value.

[0053] Calculate the dimensional deviation (the deviation between the actual size of the material and the size range in the parameter table) corresponding to the candidate parameters, and select the parameter with the smallest deviation as the basic parameter; The basic parameters are adjusted based on real-time environmental data (room temperature, humidity, voltage), and the environmentally corrected parameters are used as the initial execution parameters. The specific amendment rules are as follows: Room temperature T_ring: For every 5°C that T_ring exceeds the standard room temperature (25°C), the furnace temperature T_ring decreases by 3°C; for every 5°C that T_ring falls below the standard room temperature (25°C), T_ring increases by 3°C. Ambient humidity RH loop: For every 10% increase in RH loop above the standard humidity (50%), the processing time t is extended by 5% (for food products requiring humidity control). For every 10% decrease in grid voltage U compared to the rated voltage (220V), the heating power increases by 5% to compensate for the decrease in heating efficiency caused by insufficient voltage.

[0054] Continuously collect key status data during the processing, compare real-time data with theoretical data predicted by the digital twin model to calculate the deviation value, and trigger the parameter adjustment mechanism according to the preset deviation threshold; The key status data include temperature status (furnace zone temperature, material surface temperature, and material internal temperature), humidity / mass status (furnace humidity and material weight changes reflecting moisture evaporation), and equipment operating status (heating power, fan speed, and guide plate angle).

[0055] Set multi-dimensional deviation thresholds. When the real-time data deviation exceeds the threshold, trigger the parameter adjustment mechanism to avoid over-adjustment or adjustment lag. When multiple deviations are triggered simultaneously, they are sorted according to their impact on quality, and key deviations are adjusted first. The priority order from high to low is as follows: material temperature deviation (directly affects curing / baking effect), furnace temperature deviation (affects the overall thermal environment), humidity / weight deviation (indirectly affects moisture content), and equipment operation deviation (such as fan speed deviation).

[0056] PID control algorithm is used to handle linear deviation, and fuzzy control algorithm is used to handle nonlinear and multi-factor coupled deviation, and the parameter adjustment is calculated. Specifically, PID control is used for single linear deviations. For example, during the constant temperature stage, if only the furnace temperature deviates from the set value while humidity / wind speed is normal, the material surface temperature deviation is used as the input variable, and the heating power adjustment is used as the output variable. The adjustment range is set as follows: ΔT: -10℃ to +10℃; ΔP: -20% to +20% (power adjustment boundary). The core PID parameters include the proportional coefficient Kp, integral coefficient Ki, and derivative coefficient Kd. If the temperature deviation exceeds the threshold and is linear during actual operation, the adjustment amount is calculated based on the coefficients.

[0057] Fuzzy control is employed to address nonlinear coupling deviations. For example, in the preheating stage, if the temperature is low and the humidity is high, the combined effect of multiple factors leads to slow moisture evaporation. In this case, temperature and humidity deviations are used as input variables, while heating power and fan speed adjustments are used as output variables. The adjustment ranges are set as follows: ΔRH: -10% to +10%; Δv_F: -15% to +15% (fan speed adjustment boundary). The input and output variables are fuzzified, i.e., quantified into levels. Based on expert experience and historical processing data, rules are established for IF input conditions and THEN output results. The rules are matched according to the actual operating level, and the final adjustment amount is determined through fuzzy inference.

[0058] In scenarios where both linear and nonlinear deviations exist, PID control is used as the primary method, supplemented by fuzzy control, and the final adjustment amount is obtained through weighted calculation.

[0059] The dynamically adjusted parameters are converted into control commands that the device can execute, driving the actions of the execution components and monitoring the execution effect.

[0060] Specifically, the parameters are first adjusted according to the type of actuator. For example, to increase the heating unit power by 10%, the PWM duty cycle is adjusted from 50% to 55% (through closed-loop control with feedback from a current sensor). Commands are sent in real-time via the CAN bus with a delay of ≤10ms to ensure rapid response from the actuators. Real-time data on the actuators' actions (such as heating current and fan speed) is collected to verify the accuracy of the command execution. For example, if the command requires the heating current to increase from 10A to 11A (corresponding to a 10% power increase), and the actual collected current is 10.8A, the deviation is 1.8% ≤ ±5%, indicating successful execution. If the deviation is > ±5%, the command is resent, and the equipment status is checked. The "timestamp, reason for adjustment, adjustment amount, and execution effect" of each adjustment are recorded and synchronized to the cloud database.

[0061] By adjusting the frequency to adaptively balance real-time performance and stability, the dynamic adjustment process is optimized. In the early stage of processing (preheating stage, when the temperature changes rapidly), the frequency is adjusted once every 30 seconds to quickly respond to temperature deviations; in the middle stage of processing (constant temperature stage, when the state is stable), the frequency is adjusted once every 2 minutes to avoid excessive adjustment that could cause parameter fluctuations; in the later stage of processing (cooling stage, when the quality is finalized), the frequency is adjusted once every 5 minutes to correct only critical deviations (such as sudden temperature drops).

[0062] When multiple parameters need to be adjusted simultaneously, they are ordered according to their weight in terms of their impact on quality, with the priority from high to low being temperature parameters, hot air parameters, time parameters, and load parameters.

[0063] S5: Generate control commands based on the matched or adjusted process parameters, drive the execution of equipment actions based on the control commands, and dynamically optimize model parameters based on the execution effect.

[0064] The execution equipment includes a zone heating unit, a hot air circulation unit, a load adjustment unit, and a material conveying unit; The partitioned heating unit uses PWM pulse width modulation and a current sensor feedback closed-loop control to distribute the heating power of the top, middle and bottom of the oven. The hot air circulation unit controls the fan speed and guide plate angle through frequency conversion control (fan) and stepper motor drive (guide plate), respectively. The load-bearing adjustment unit is driven by a servo motor and adjusts the height of the load-bearing frame based on closed-loop control of displacement sensor feedback. The material conveying unit controls the start and stop of the conveyor belt and adjusts its speed via frequency conversion.

[0065] The optimized process parameters are broken down into specific control commands according to the type of actuator, such as heating power → PWM duty cycle, fan speed → frequency converter frequency, and verified through equipment security permissions (e.g., remote control requires administrator authorization) to avoid misoperation. Real-time monitoring of actuator feedback (e.g., motor speed, heating current) is conducted; if the execution deviation is > ±5%, an audible and visual alarm is immediately issued, and the command is reissued.

[0066] Record all data during the processing to achieve full traceability from material entering the furnace to finished product leaving the warehouse. Assign a unique identifier to each batch of materials using RFID tags or QR codes, recording the material name, batch number, initial state (e.g., moisture content, weight), and furnace entry time. Continuously collect and store process parameters (temperature, time, humidity), equipment operating data (heating power, current, vibration), and quality inspection data (e.g., finished product moisture content, degree of curing). Support for querying all data in the corresponding processing flow by finished product identifier and batch number, including process parameter curves, equipment operating status, and quality inspection results, facilitating problem traceability.

[0067] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. An intelligent adaptive control method for a multi-size compatible oven processing system, characterized in that, Includes the following steps: S1: Collect material size data, real-time status data during processing, and environmental data, and preprocess the collected data; S2: Calculate material dimensions using a dynamic dimension recognition algorithm and classify the materials. Establish a mapping relationship between dimensions and process parameters by combining material type and target processing requirements. S3: Based on digital twin technology, simulate the processing under different process parameters and predict the optimal set of process parameters; S4: Based on the material size, the process parameters are automatically matched and adapted to the process parameters. The process parameters are then dynamically adjusted in combination with real-time status data during the processing. S5: Generate control commands based on the matched or adjusted process parameters, drive the execution of equipment actions based on the control commands, and dynamically optimize model parameters based on the execution effect.

2. The intelligent adaptive control method for a multi-size compatible oven processing system according to claim 1, characterized in that, Based on the calculated material dimensions and dimensional classifications, and combining the material type with the target processing requirements, a mapping relationship between dimensions and process parameters is established, including the following steps: Define the specific dimensions of input features and output parameters, and construct a three-dimensional database of input features, output parameters, and quality results; Correlation analysis is performed on the input features to retain those that are strongly correlated with the output parameters and quality results. Redundant features are removed based on RFE to obtain the core input feature set. A hybrid model of gradient boosting tree and neural network is adopted to take into account both linear and nonlinear mapping relationships and predict output parameters based on input features; The predicted output parameters are corrected by taking into account the physical constraints, process constraints and environmental fluctuations of the actual processing scenario.

3. The intelligent adaptive control method for a multi-size compatible oven processing system according to claim 2, characterized in that, The process constraints include size and parameter correlation constraints, material type constraints, and target requirement constraints.

4. The intelligent adaptive control method for a multi-size compatible oven processing system according to claim 1, characterized in that, S3 includes the following steps: Construct 3D models of the oven body, sensors, and materials, and use LOD technology to refine the core components; A multiphysics coupling model was constructed to simulate key processes in oven processing, and the virtual model parameters were calibrated through physical experiments to eliminate the deviation between simulation and reality. Establish a communication interface between the industrial IoT gateway and the digital twin platform to synchronize physical device data and virtual models in real time; Based on the material size, type, and target quality requirements, parameter boundaries are set according to the material characteristics, and Latin hypercube sampling is used to generate multiple sets of candidate parameters within the parameter range; The entire processing of candidate parameters is simulated one by one in the digital twin model, and detailed simulation data is output. The Analytic Hierarchy Process (AHP) was used to quantitatively score the simulation results of each set of candidate parameters based on evaluation indicators, and the set with the highest score was selected as the optimal parameter set.

5. The intelligent adaptive control method for a multi-size compatible oven processing system according to claim 4, characterized in that, A multiphysics coupling model is constructed to simulate key processes in oven processing, including the following steps: Define the heat-fluid, heat-mass, and heat-reaction coupling relationships for the oven and materials; Build a geometric model and mesh the oven body and materials; Import material properties and set initial and boundary conditions; Configure the solver for field decoupling and set the solution parameters; Key data collected during the oven processing are used as a benchmark to calibrate model parameters.

6. The intelligent adaptive control method for a multi-size compatible oven processing system according to claim 4, characterized in that, The evaluation indicators include quality compliance, processing uniformity, processing efficiency, and energy consumption cost.

7. The intelligent adaptive control method for a multi-size compatible oven processing system according to claim 1, characterized in that, S4 includes the following steps: Construct a four-dimensional relational database of size, type, target requirements, and process parameters, and filter out a set of candidate parameters that meet the conditions through multi-condition retrieval; Calculate the dimensional deviations corresponding to the candidate parameters, and select the parameter with the smallest deviation as the basic parameter; Adjust the basic parameters based on real-time environmental data, and use the environmentally corrected parameters as the initial execution parameters. Continuously collect key status data during the processing, compare real-time data with theoretical data predicted by the digital twin model to calculate the deviation value, and trigger the parameter adjustment mechanism according to the preset deviation threshold; PID control algorithm is used to handle linear deviation, and fuzzy control algorithm is used to handle nonlinear and multi-factor coupled deviation, and the parameter adjustment is calculated. The dynamically adjusted parameters are converted into control commands that the device can execute, driving the actions of the execution components and monitoring the execution effect.

8. The intelligent adaptive control method for a multi-size compatible oven processing system according to claim 1, characterized in that, The execution equipment includes a zone heating unit, a hot air circulation unit, a load adjustment unit, and a material conveying unit; The partitioned heating unit uses PWM pulse width modulation and a current sensor feedback closed-loop control to distribute the heating power of the top, middle and bottom of the oven. The hot air circulation unit controls the fan speed and guide plate angle through frequency conversion control and stepper motor drive, respectively. The load-bearing adjustment unit is driven by a servo motor and adjusts the height of the load-bearing frame based on closed-loop control of displacement sensor feedback. The material conveying unit controls the start and stop of the conveyor belt and adjusts its speed via frequency conversion.

Images