Dynamic adjustment control system for carton production and dynamic adjustment method for carton production

Abstract

The invention relates to the technical field of intelligent industrial manufacturing and automation, in particular to a carton production dynamic adjustment control system and a carton production dynamic adjustment method thereof. The problems of parameter lag, process coupling interference, quality tracing difficulty and the like in traditional carton production are solved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent industrial manufacturing and automation technology, and more specifically, to a carton production dynamic adjustment control system and a carton production dynamic adjustment method thereof. Background Art

[0002] Cardboard boxes play a key role in daily life and industrial production, for example, they are needed in express delivery and cargo transportation. In traditional carton production, raw cardboard is pressed and bent into shape to form cartons. During the manufacturing process, fluctuations in raw materials, such as changes in grammage, moisture content, and environmental factors, can lead to errors in carton production and result in defective products. Therefore, it is necessary to adjust production parameters in the production process to overcome these errors. However, traditional control processes mostly rely on manual experience or fixed parameter control, which cannot respond to real-time adjustments. At the same time, during adjustment, there is dynamic coupling between the loading, pressing, and forming processes. For example, changes in raw materials affect the pressing pressure requirements, and pressing temperature affects the deformation of the forming mold. Traditional single-process control cannot decouple, resulting in unsatisfactory adjustment results.

[0003] Therefore, a carton production dynamic adjustment control system and method based on multimodal data fusion and hybrid model prediction is proposed, which is suitable for the intelligent control of packaging equipment such as corrugated cardboard production lines and carton forming machines. Summary of the Invention

[0004] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a carton production dynamic adjustment control system and a carton production dynamic adjustment method thereof, so as to solve the problems existing in the above-mentioned background technology.

[0005] The above technical objectives of the present invention are achieved through the following technical solutions: A carton production dynamic adjustment control system, comprising:

[0006] The data acquisition module is installed in each process of the carton production process to collect and transmit multimodal production parameters in the loading process, pressing process, forming process, drying process and quality inspection process;

[0007] An edge computing module is used to receive the multimodal production parameters collected by the data acquisition module, perform data preprocessing, and then use a feature extraction algorithm to sequentially obtain and transmit the input of the feeding model, the pressing model, the forming model, the drying model, and the quality inspection model;

[0008] A hybrid model module is used to receive the model input transmitted by the edge computing module, input it into the model of the corresponding process, and then output and transmit the model prediction result;

[0009] An instruction calculation module is used to receive the multimodal production parameters collected by the data acquisition module and the model prediction results output by the hybrid model module, and dynamically generate and transmit adjustment instructions for corresponding processes after comprehensive judgment;

[0010] The cross-process execution module is used to receive the adjustment instruction transmitted by the instruction calculation module, and according to the adjustment instruction, dynamically adjust the production parameters of the corresponding process before executing the adjustment instruction to overcome the process coupling interference, and then execute the adjustment instruction for adjustment.

[0011] Optionally, the hybrid model module includes:

[0012] A random forest multi-objective optimization model is used to receive the input of the feeding model, predict the influence trend of raw materials and environmental humidity during the feeding process, and generate a feeding prediction result;

[0013] An LSTM-PINN hybrid model is used to receive the lamination model input, predict the cardboard deformation trend during the lamination process, and generate a lamination prediction result;

[0014] a physics-guided reinforcement learning model for receiving the molding model input, predicting the influence trend of mold spacing and optimized conveying speed during the molding process, and generating a molding prediction result;

[0015] A multivariable predictive control model is used to receive the drying model input, predict the drying efficiency and energy consumption trends during the drying process, and generate a drying prediction result;

[0016] A graph neural network knowledge graph model is used to receive the input of the quality inspection model, perform defect root cause analysis and quality traceability during the quality inspection process, and generate quality inspection prediction results;

[0017] The feeding prediction results, pressing prediction results, forming prediction results, drying prediction results and quality inspection prediction results are fitted to generate the model prediction results.

[0018] Optionally, the hybrid model module includes:

[0019] A random forest multi-objective optimization model is used to receive the input of the feeding model, predict the influence trend of raw materials and environmental humidity during the feeding process, and generate a feeding prediction result;

[0020] An LSTM-PINN hybrid model is used to receive the lamination model input, predict the cardboard deformation trend during the lamination process, and generate a lamination prediction result;

[0021] a physics-guided reinforcement learning model for receiving the molding model input, predicting the influence trend of mold spacing and optimized conveying speed during the molding process, and generating a molding prediction result;

[0022] A multivariable predictive control model is used to receive the drying model input, predict the drying efficiency and energy consumption trends during the drying process, and generate a drying prediction result;

[0023] A graph neural network knowledge graph model is used to receive the input of the quality inspection model, perform defect root cause analysis and quality traceability during the quality inspection process, and generate quality inspection prediction results;

[0024] The feeding prediction results, pressing prediction results, forming prediction results, drying prediction results and quality inspection prediction results are fitted to generate the model prediction results.

[0025] Optionally, the instruction calculation module includes:

[0026] an adjustment threshold calculation unit, which dynamically generates an adjustment threshold for each process based on the multimodal production parameters collected by the data collection module;

[0027] An adjustment determination calculation module compares the feeding prediction result, the pressing prediction result, the forming prediction result, the drying prediction result, and the quality inspection prediction result with the adjustment threshold of the corresponding process to determine whether adjustment is required;

[0028] The adjustment target value calculation module calculates the corresponding adjustment target value of the process that needs to be adjusted based on the determination result of the adjustment determination calculation module, and generates the adjustment instruction based on the generated adjustment target value.

[0029] Optionally, the cross-process execution module includes:

[0030] an instruction receiving unit, configured to receive the adjustment instruction transmitted by the instruction calculation module and analyze the adjustment target value of the corresponding process;

[0031] A multi-process coupling modeling unit is used to establish a process dynamic model of the dynamic relationship between multiple processes, and based on the adjustment target value and the output result of the process dynamic model, fine-tune the process parameters in advance to offset the adjustment interference;

[0032] a program execution module, configured to adjust corresponding process equipment based on the adjustment target value analyzed by the instruction receiving unit;

[0033] The fine-tuning execution module is used to detect the production parameters of the adjusted process in real time, and perform fine-tuning based on the comparison between the production parameters of the process and the trigger target threshold.

[0034] Optionally, the trigger target threshold includes: a prediction result deviation value, a process coupling interference trigger value, a process safety limit value, and a quality feedback trigger value.

[0035] Optionally, the data acquisition module includes:

[0036] Data sensing units, installed in each process, are used to collect multimodal data;

[0037] A data alignment unit, configured to bind and mark the multimodal data collected by the data sensing unit with a time series to perform initial time synchronization;

[0038] A data transmission unit is used to transmit the multimodal data processed by the data alignment unit.

[0039] Optionally, the multimodal production parameters include: mechanical parameters, optical parameters, environmental parameters and material parameters.

[0040] Optionally, the edge computing module includes:

[0041] A data receiving unit, configured to receive the multimodal data collected by the data collection module;

[0042] A data preprocessing unit, configured to perform data cleaning preprocessing on the multimodal data to remove noise and invalid data, and perform time synchronization alignment to achieve secondary time synchronization;

[0043] a feature extraction unit, configured to extract the feeding model input, the pressing model input, the forming model input, the drying model input, and the quality inspection model input from the multimodal data using a feature extraction algorithm;

[0044] The feature input unit is used to input the model input quantity extracted by the feature extraction unit into the hybrid model module.

[0045] Optionally, it also includes a self-healing security module for integrating a blockchain instruction storage unit and a three-level fault degradation control mechanism to respond to emergencies.

[0046] A carton production dynamic adjustment method based on the above-mentioned carton production dynamic adjustment control system includes:

[0047] Step A, multimodal data collection: multimodal production parameters in normal production are collected based on data collection modules set in the loading process, pressing process, forming process, drying process and quality inspection process;

[0048] Step B, model input extraction: After preprocessing the multimodal production parameters through the edge computing module, the input of the feeding model, the pressing model, the forming model, the drying model, and the quality inspection model are extracted based on the feature extraction algorithm;

[0049] Step C, obtaining process prediction results: Based on the input of the feeding model, the pressing model, the forming model, the drying model, and the quality inspection model, the corresponding inputs are input into the hybrid model module to obtain the model prediction results;

[0050] Step D, obtaining the adjustment target value: The instruction calculation module compares the model prediction results corresponding to each process outputted in Step C with the dynamically generated adjustment thresholds for each process to determine whether adjustment is required. The module then calculates and obtains the adjustment target value for the process requiring adjustment, and generates an adjustment instruction based on the corresponding generated adjustment target value.

[0051] Step E, adjustment instruction execution: The cross-process execution module adjusts the production parameters of the corresponding process equipment based on the output results of the process dynamic model to overcome the process coupling interference, and then executes the adjustment instructions generated by the instruction calculation module to adjust the production equipment of each process;

[0052] Step F, target improvement and fine-tuning: The cross-process execution module monitors the adjusted production equipment, and after reaching the trigger target threshold, performs a second fine-tuning on the production equipment to improve the adjustment effect.

[0053] In summary, the present invention has the following beneficial effects:

[0054] 1. By setting up the data acquisition module in each process, the production parameters of each process in the carton production process can be effectively collected, involving mechanical parameters, optical parameters, environmental parameters and material parameters. When they are input into the hybrid model module, the model output results corresponding to different processes are output respectively to represent the carton production trend in the future period, and transmitted to the instruction calculation module to calculate the adjustment target value of the process to be adjusted and generate adjustment instructions to adjust the corresponding process equipment to ensure the yield rate of carton production.

[0055] 2. Before the cross-process execution module adjusts each process, in order to ensure that the processes avoid mutual interference during the adjustment process, online modeling and real-time decoupling control of multi-process dynamic coupling interference are adopted to further improve the adjustment process and achieve better adjustment effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic diagram of the system execution flow of the present invention;

[0057] Figure 2 It is a schematic flow chart of the steps of the system execution method of the present invention. DETAILED DESCRIPTION

[0058] To make the objectives, features, and advantages of the present invention more readily apparent, the following detailed description of the present invention is provided with reference to the accompanying drawings. The accompanying drawings illustrate several embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein.

[0059] In the present invention, unless otherwise expressly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be a communication between the two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances. The terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of such features.

[0060] In the present invention, unless otherwise expressly specified and limited, a first feature being "above" or "below" a second feature may include the first and second features being in direct contact, or may include the first and second features being in contact not directly but through another feature between them. Moreover, a first feature being "above," "above," and "above" a second feature includes the first feature being directly above and obliquely above the second feature, or simply indicates that the first feature is higher in level than the second feature. A first feature being "below," "below," and "below" a second feature includes the first feature being directly below and obliquely below the second feature, or simply indicates that the first feature is lower in level than the second feature. The terms "vertical," "horizontal," "left," "right," "above," "below," and similar expressions are for illustrative purposes only and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore should not be understood as limiting the present invention.

[0061] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0062] The present invention provides a carton production dynamic adjustment control system, such as Figure 1 As shown, including:

[0063] The data acquisition module is installed in each process of the carton production process to collect and transmit multimodal production parameters in the loading process, pressing process, forming process, drying process and quality inspection process;

[0064] An edge computing module is used to receive the multimodal production parameters collected by the data acquisition module, perform data preprocessing, and then use a feature extraction algorithm to sequentially obtain and transmit the input of the feeding model, the pressing model, the forming model, the drying model, and the quality inspection model;

[0065] A hybrid model module is used to receive the model input transmitted by the edge computing module, input it into the model of the corresponding process, and then output and transmit the model prediction result;

[0066] An instruction calculation module is used to receive the multimodal production parameters collected by the data acquisition module and the model prediction results output by the hybrid model module, and dynamically generate and transmit adjustment instructions for corresponding processes after comprehensive judgment;

[0067] The cross-process execution module is used to receive the adjustment instruction transmitted by the instruction calculation module, and according to the adjustment instruction, dynamically adjust the production parameters of the corresponding process before executing the adjustment instruction to overcome the process coupling interference, and then execute the adjustment instruction for adjustment.

[0068] Furthermore, the hybrid model module includes:

[0069] A random forest multi-objective optimization model is used to receive the input of the feeding model, predict the influence trend of raw materials and environmental humidity during the feeding process, and generate a feeding prediction result;

[0070] An LSTM-PINN hybrid model is used to receive the lamination model input, predict the cardboard deformation trend during the lamination process, and generate a lamination prediction result;

[0071] a physics-guided reinforcement learning model for receiving the molding model input, predicting the influence trend of mold spacing and optimized conveying speed during the molding process, and generating a molding prediction result;

[0072] A multivariable predictive control model is used to receive the drying model input, predict the drying efficiency and energy consumption trends during the drying process, and generate a drying prediction result;

[0073] A graph neural network knowledge graph model is used to receive the input of the quality inspection model, perform defect root cause analysis and quality traceability during the quality inspection process, and generate quality inspection prediction results;

[0074] The feeding prediction results, pressing prediction results, forming prediction results, drying prediction results and quality inspection prediction results are fitted to generate the model prediction results.

[0075] In Example 1, the random forest multi-objective optimization model uses single-tree prediction and multi-objective output. The model inputs are represented by: raw material characteristics (raw material weight, raw material moisture content, fiber orientation angle); environmental parameters (ambient humidity, ambient temperature); historical operating conditions (historical average energy consumption, raw material switching frequency). The above model inputs are all aligned with data of different sampling frequencies through dynamic time warping (DTW);

[0076] The output is represented as: Single tree prediction model Expressed as for the t-th tree, input sample x i The predicted value of R m Represented as leaf node area, c m Expressed as the node output value, the model can simultaneously predict the values ​​of three target variables, namely raw material cost C, energy consumption E and equipment wear W, which are expressed as At the same time, a triple is stored at each leaf of each tree.

[0077] For the multi-objective variable values ​​generated by calculation, the multi-objective optimization formula is used for optimization, which is expressed as Among them, u = [u1, u2, u3] is the control variable, where u1 represents the proportion of raw material A, u2 represents the humidity compensation coefficient, and u3 represents the feeding speed value. At the same time, conditional constraints are used to achieve the best optimization effect, which is expressed as: equipment safety constraint, 0.9P 额定 ≤P(u1,u3)≤1.1P 额定 ; Quality constraint, predicted edge pressure strength ≥ 7kN / m3; process constraint, process constraint: |u2-1| ≤ 0.2·(H 实时 -50%);

[0078] Based on the above target values, the raw material cost trend, energy consumption trend and equipment wear trend are output respectively. The raw material cost trend is expressed as ΔC = 0.15u1-0.08u3, and the energy consumption trend is expressed as ΔE = 1.2u3+0.05H. 2 , the equipment wear trend is expressed as Then output the feeding prediction results;

[0079] The LSTM-PINN hybrid model uses an LSTM network to predict cardboard deformation within the future production cycle. The model input is multidimensional time series data, including real-time pressure mean, pressure fluctuation variance, vibration signal 1-5 kHz energy ratio, temperature gradient, and historical deformation sequence.

[0080] The output is: through the LSTM network structure, represented as h t =LSTM(x t ,h t-1 θLSTM ), where the input gate / forget gate / output gate are parameterized, the time window length = 60s, the hidden layer dimension = 128, and the physical information constraint is added, which is expressed as Where E represents the elastic modulus, μ represents the damping coefficient, and a mixed loss function is used to optimize the prediction results, which is expressed as in Expressed as a data fitting term, It is represented as the physical residual term, and F is represented as the residual operator of the mechanical equation, where the elastic modulus E is updated by online maximum likelihood estimation and is expressed as Then the curvature radius change ΔR and thickness deviation prediction values ​​in the future cycle are obtained Where t1 represents the future cycle time, and then the pressing prediction result is output;

[0081] A physics-guided reinforcement learning model uses a state-space-action-space-reward function to predict the impact of die spacing and optimized conveyor speed. The input is a multi-dimensional state vector, including the pressing die spacing, real-time conveyor speed, cardboard stiffness coefficient, ambient humidity, and historical loss.

[0082] Output: Input the multidimensional state vector into the model, and the state space is represented by s t =d 模具 ,v 传送 ,S 挺度 ,H 湿度 ,E 能耗 ], the action space is represented as a t =[Δd 模具 ∈-0.1,0,+0.1}mm,Δv 传送 ∈{-0.5,0,+0.5}m / min], the reward function is expressed as R t = 0.6·QualityScore+0.3·Throughput-0.1·EnergyCost, where the quality score is based on dimensional deviation and crease angle. The optimal die spacing and optimal conveying speed are then calculated, and the lamination prediction result is output.

[0083] The multivariable predictive control model uses a state-space model to predict drying efficiency and energy consumption trends during the drying process. The input is multidimensional process variables, including hot air temperature, dehumidification fan speed, cardboard moisture content, instantaneous energy consumption, and ambient dew point temperature.

[0084] The output is: Input the multidimensional process variable into the state space model, represented as x k+1 =Ax k +Bu k +w k,y k =Cx k +v k , where the state variable x=[T 纸板 ,H 含水 ,P 能耗 ] T , control variable u=[ΔT 热风 ,ΔRPM 风机 ] T , and the optimization formula is used, which is expressed as rolling time domain optimization: The prediction time domain N p =10, control time domain N c =5, weight matrix Q = diag(0.7,0.3), R = diag(0.2,0.8); then calculate the optimal hot air temperature and optimal dehumidification fan speed, and output the drying prediction result;

[0085] The graph neural network knowledge graph model uses a graph convolutional network to perform defect root cause analysis and quality traceability. The input is multi-dimensional defect features, including: defect type, location coordinates, dimensional deviation, edge pressure strength, and associated process parameter snapshots;

[0086] The output is: Input the multi-dimensional defect features into the graph convolutional network, expressed as Adjacency Matrix Including the relationship between equipment, process, defects, node characteristics H (0) Contains historical case data and adopts multi-task learning objectives, expressed as This includes: classification tasks: defect type identification (cross entropy loss), regression tasks: root cause probability prediction (mean square error), and graph reconstruction tasks: knowledge graph completion (contrast loss). It then outputs defect phenomena and defect root cause probability distribution analysis, conducts quality traceability, links similar defect cases from the past 30 days, generates SPC control charts, identifies abnormal process parameters, and outputs quality inspection prediction results.

[0087] Furthermore, the instruction calculation module includes:

[0088] an adjustment threshold calculation unit, which dynamically generates an adjustment threshold for each process based on the multimodal production parameters collected by the data collection module;

[0089] An adjustment determination calculation module compares the feeding prediction result, the pressing prediction result, the forming prediction result, the drying prediction result, and the quality inspection prediction result with the adjustment threshold of the corresponding process to determine whether adjustment is required;

[0090] The adjustment target value calculation module calculates the corresponding adjustment target value of the process that needs to be adjusted based on the determination result of the adjustment determination calculation module, and generates the adjustment instruction based on the generated adjustment target value.

[0091] In the second embodiment, the adjustment threshold calculation unit is based on the multimodal production parameters collected by the data collection module during the current carton production. In the loading process, the adjustment threshold is calculated as the grammage fluctuation threshold: δ = μ ± 3σ. When the adjustment determination module determines that the real-time grammage deviation is greater than the grammage fluctuation threshold or the energy consumption exceeds the limit, the loading process is triggered to adjust.

[0092] In the lamination process, the adjustment threshold is the standard curvature radius change rate and the upper limit of the allowable thickness deviation. When the adjustment judgment module determines that the curvature radius change rate or the thickness deviation value is greater than the adjustment threshold, the standard curvature radius change rate and the upper limit of the allowable thickness deviation can be set manually or automatically generated by the system based on historical work records, triggering the adjustment of the lamination process;

[0093] In the molding process, the adjustment threshold is the size deviation threshold, which is expressed as the precision of the order product * the yield rate. When the adjustment judgment module determines that the real-time size deviation is greater than the size deviation threshold, or when the model confidence calculated by the hybrid model module is less than 85%, the adjustment of the molding process is triggered;

[0094] In the drying process, the adjustment threshold is the energy consumption safety threshold, which is expressed as 0.9P 额定 (1-0.1 vibration level): When the adjustment determination module determines that the instantaneous energy consumption in the drying process exceeds the energy consumption safety threshold, or the moisture content deviation is greater than 2% for five minutes, the adjustment of the drying process is triggered;

[0095] In the quality inspection process, the SPC control line formula for the inter-process is expressed as UCL = μ + 3σ, LCL = μ - 3σ, and then the defect probability and substandard product probability are obtained. When the adjustment judgment module determines that there are two consecutive substandard products or the defect probability is greater than 90%, the adjustment of the quality inspection process is triggered;

[0096] In the adjustment calculation of the loading process, the loading prediction results output by the hybrid model module are used to generate a Pareto solution set based on the NSGA-III algorithm. Three different adjustment mode schemes are generated: efficiency, economy, and comprehensive. Adjustments are made based on the results of system presets or manual control, and then the adjustment target value of the loading process is generated.

[0097] In the calculation of the adjustment of the pressing process, the calculated and predicted curvature deviation is expressed as ΔR 偏差 =ΔR-R set , thickness deviation is expressed as Δh 偏差 =Δh t+10 -h max, based on the PID adjustment amount, the pressure compensation base value is expressed as ΔP base =K p ·Δh 偏差 , the integral term (accumulated historical deviation) is expressed as The differential term (trend suppression) is expressed as The total pressure compensation ΔP is obtained based on the comprehensive calculation of the above items. ID , and then generate the adjustment target value of the pressing process;

[0098] In the adjustment calculation of the molding process, the optimal mold spacing and optimal conveying speed are obtained through the physical guidance reinforcement learning model, and then the adjustment target value of the molding process is generated;

[0099] In the regulation calculation of the drying process, the optimal hot air temperature and optimal dehumidification fan speed are obtained through the multivariable predictive control model, and then the regulation target value of the drying process is generated;

[0100] In the adjustment calculation of the quality inspection process, the graph neural network knowledge graph model is used to output the probability distribution analysis of defect phenomena and defect root causes, and conduct quality traceability, link similar defect cases in the past 30 days, generate SPC control charts, identify abnormal process parameters, and adjust the target value type as strategy distribution. For example: the defect phenomenon is insufficient edge pressure strength, and the probability distribution of the root causes is low pressing pressure (72%), low pressing pressure (72%), and raw material weight deviation (10%). The corresponding adjustment strategies are pressure compensation +0.2MPa, drying temperature increased by 8°C, and raw material ratio adjustment notification. Management personnel make corresponding adjustments through the adjustment strategy distribution to achieve perfect finished products.

[0101] Furthermore, the cross-process execution module includes:

[0102] an instruction receiving unit, configured to receive the adjustment instruction transmitted by the instruction calculation module and analyze the adjustment target value of the corresponding process;

[0103] A multi-process coupling modeling unit is used to establish a process dynamic model of the dynamic relationship between multiple processes, and based on the adjustment target value and the output result of the process dynamic model, fine-tune the process parameters in advance to offset the adjustment interference;

[0104] a program execution module, configured to adjust corresponding process equipment based on the adjustment target value analyzed by the instruction receiving unit;

[0105] The fine-tuning execution module is used to detect the production parameters of the adjusted process in real time, and perform fine-tuning based on the comparison between the production parameters of the process and the trigger target threshold.

[0106] In the third embodiment, during the execution of each process, the multi-process coupling modeling unit will first establish a process dynamic model of the dynamic relationship between multiple processes using a transfer function matrix to express it as Among them G ij (s) is the transfer function from the jth process to the ith process, i, j = 1, 2, ΔQ i Expressed as the output deviation of the i-th process (such as pressing thickness, forming accuracy), ΔP j It is expressed as the input change of the jth process (such as feeding pressure, conveying speed). In the specific implementation process, when the interference source process is the feeding process and the interfered source process is the pressing process, the interference is expressed as the change in pressing pressure demand caused by the fluctuation of raw material weight, and the transfer function is When the interference source process is the pressing process and the interfered process is the molding process, the interference is expressed as the change in pressing temperature affecting the thermal deformation of the molding die, and the transfer function is When the interference source process is the forming process and the interfered process is the quality inspection process, the interference is represented by the crease angle deviation caused by the excessive forming speed, and the transfer function is

[0107] The decoupling controller is designed using inverse model feedforward compensation, which is expressed as Where T f =0.5s, which is the filter time constant. The parameters are adjusted to ensure stability through frequency domain response matching. The gain margin is >6dB and the phase margin is >45°. At the same time, a feedback correction loop is used and a Smith predictor is used to compensate for the large time delay. The system is expressed as τ = 0.2s, expressed as the time delay, T = 1.8s, expressed as the time constant, the decoupling command is obtained through the above steps;

[0108] In a specific embodiment, the above material and pressing are taken as an example, and the interference prediction is expressed as ΔQ pred =G ij @Δu j , which means that the feeding speed is increased by 0.5m / min → the predicted pressing pressure needs to be increased by 0.075MPa, the decoupling instruction is generated, and the feedforward compensation amount is calculated: Δu 补偿 =C ff (s)·ΔQ pred The calculation results generate decoupling instructions to be executed in the corresponding process, and then the process parameters are fine-tuned in advance to offset the adjustment interference.

[0109] Furthermore, the trigger target threshold value includes: a prediction result deviation value, a process coupling interference trigger value, a process safety limit value and a quality feedback trigger value.

[0110] In other embodiments, after the adjustment is completed, a judgment is made, and the judgment result is compared with the prediction result deviation value, the process coupling interference trigger value, the process safety limit value, and the quality feedback trigger value, with the process safety limit value, the quality feedback trigger value, the prediction result deviation value, and the process coupling interference trigger value being prioritized in descending order; when the process safety limit value is triggered, a shutdown is triggered immediately; when the quality feedback trigger value is triggered, production is suspended and an alarm is issued; when the prediction result deviation value is triggered, parameter compensation is started for fine-tuning; when the process coupling interference trigger value is triggered, feedforward decoupling is performed; when multiple thresholds are triggered simultaneously, the highest-level action is executed according to the priority;

[0111] Obtaining the deviation value of the prediction result: By analyzing historical data and collecting production data from the past six months, we calculate the 3σ range of the model prediction error (confidence level 99.7%). Based on the confidence interval of the prediction result (such as the probability value output by the LSTM), we dynamically set the threshold. If the model confidence level is ≥90%, the threshold is relaxed by 10%; if the confidence level is ≤70%, the threshold is tightened by 20%, thus achieving calibration of the model confidence level.

[0112] Dynamic adjustment strategies and parameter fine-tuning are used to address threshold errors. Error statistics are recalculated every 24 hours (window size = the most recent 1000 batches of data). When the prediction error exceeds the threshold for 10 consecutive times, the model is quickly fine-tuned and the threshold is updated simultaneously.

[0113] The method for obtaining the process coupling interference trigger threshold is to identify the transfer function, measure the coupling gain between processes through step response experiments, and calculate the dynamic coupling degree, which is expressed as When C ij ≥30% was considered as significant interference;

[0114] The threshold is optimized and adjusted through dynamic adjustment strategy, including online transfer function update: Recursive least squares (RLS) re-identification of G every ten minutes ij , and update the coupling threshold, expressed as At the same time, during the changeover phase, the threshold is temporarily reduced to 20% to increase sensitivity and achieve production mode adaptation to adjust the normal operation between processes affected by coupling interference;

[0115] The method for obtaining the process safety limit threshold is to set a hard safety value through the equipment rated parameters and determine the extreme temperature threshold that the material can withstand through the cardboard material convention. Through equipment health compensation and environmental adaptability adjustment, the threshold is made more in line with the actual requirements of the actual work.

[0116] The method for obtaining the quality feedback trigger threshold is to set the dimensional deviation threshold and edge pressure strength threshold through the process standards of actual production specifications, and at the same time calculate the control line (UCL / LCL) through statistical process control, UCL = μ + 3σ; when optimizing the threshold, the threshold of high-precision orders (such as pharmaceutical packaging) with order severity matching is tightened by 50%, and defect pattern recognition is used. If a specific defect (such as delamination) is detected, the relevant threshold is temporarily lowered by 30% to achieve fine-tuning matching of the threshold.

[0117] Furthermore, the data acquisition module includes:

[0118] Data sensing units, installed in each process, are used to collect multimodal data;

[0119] A data alignment unit, configured to bind and mark the multimodal data collected by the data sensing unit with a time series to perform initial time synchronization;

[0120] A data transmission unit is used to transmit the multimodal data processed by the data alignment unit.

[0121] Furthermore, the multimodal production parameters include: mechanical parameters, optical parameters, environmental parameters and material parameters.

[0122] Furthermore, the edge computing module includes:

[0123] A data receiving unit, configured to receive the multimodal data collected by the data collection module;

[0124] A data preprocessing unit, configured to perform data cleaning preprocessing on the multimodal data to remove noise and invalid data, and perform time synchronization alignment to achieve secondary time synchronization;

[0125] a feature extraction unit, configured to extract the feeding model input, the pressing model input, the forming model input, the drying model input, and the quality inspection model input from the multimodal data using a feature extraction algorithm;

[0126] The feature input unit is used to input the model input quantity extracted by the feature extraction unit into the hybrid model module.

[0127] In Example 4, the model input of the loading process is the raw material weight (g / ㎡), moisture content (%), fiber orientation angle (°), ambient temperature and humidity, and historical energy consumption (kWh) extracted by the data acquisition module. After data cleaning and preprocessing, feature extraction and time series synchronization are used to extract feature vectors as the model input of the loading process;

[0128] The model input of the pressing process is obtained from the data acquisition module: pressure time series, vibration spectrum, temperature gradient. The sliding window segmentation method is used, with a 60-second window and a step size of 1 second to generate a 60×15 matrix (15 dimensions / second). In feature extraction, time domain statistics are performed to calculate the mean, variance, and kurtosis. Through frequency domain energy analysis and temperature gradient calculation, the above data are fitted to obtain the feature vector as the input of the LSTM-PINN model, expressed as [μ_P,σ_P 2 ,K_P,E_1-5kHz,VT];

[0129] The model input of the molding process is obtained from the data acquisition module, which includes the mold spacing, conveying speed, and cardboard stiffness. In feature extraction, the original data is smoothed by Kalman filtering, and the spacing change rate is calculated as Through the speed-stiffness coupling characteristics, the dynamic stiffness coefficient and the product of speed-stiffness are calculated and expressed as ; Fit the above data to obtain the feature vector [d, Δd, v, S_dynamic, H_compensation], which serves as the input of the physics-guided reinforcement learning model;

[0130] The model input of the drying process is the hot air temperature, dehumidification fan speed, and moisture content obtained from the data acquisition module. In feature extraction, the hot air temperature fluctuation is analyzed and the sliding average within a certain time period is calculated. In this embodiment, ten minutes is selected and expressed as And the temperature gradient is calculated synchronously, which is expressed as Calculate the energy efficiency index, expressed as The main frequency amplitude of the fan speed frequency is extracted through frequency domain characteristics, A main = max(|FFT(RPM)|); and then the characteristic vector [T a vg,VT,n,A m ain], as the input of the multivariable predictive control model;

[0131] The model input of the quality inspection process obtains defect images, edge pressure strength and process parameters from the data acquisition module; in feature extraction, HOG features and LBP texture features are extracted through image feature extraction; in process parameter association, the defect-parameter association matrix is ​​constructed to extract the common defect phenomena in the process parameters; the edge pressure strength slope of the last 10 boxes is calculated to obtain the edge pressure strength trend, which is expressed as The above data are fitted to obtain the feature vector [HOG feature (64 dimensions), LBP feature (256 dimensions), M_ij, k], which is then used as the input of the graph neural network.

[0132] Furthermore, it also includes a self-healing safety module for integrating a blockchain instruction storage unit and a three-level fault degradation control mechanism to respond to sudden accidents.

[0133] In Example 5, the trigger condition for a level 1 fault is that the deviation of the prediction model exceeds the limit continuously (>15% for 10 seconds). The solution is to switch to the backup model and send an early warning to the MES system.

[0134] The triggering condition for a secondary fault is when a key parameter exceeds a safety threshold (e.g., pressure > 2.5 MPa) or the equipment vibration level is ≥ 3. The solution is to reduce the load (limit the power to 70%) and start online diagnosis.

[0135] Level 3 fault trigger conditions are hardware failure (such as motor overcurrent) and three consecutive regulation failures. The solution is emergency shutdown, locking the equipment, and generating a blockchain maintenance work order.

[0136] The self-healing safety module includes a degradation actuator, including: safety relay (cut off power supply, response time <50ms); redundant control channel (main and standby PLC switching time <200ms); emergency ventilation system (forced start when temperature rise >10℃ / min);

[0137] In other embodiments, the self-healing security module may operate in conjunction with other modules to increase protection for the device;

[0138] In one embodiment, the self-healing safety module interacts with the data acquisition module. When the data acquisition module detects a sudden increase in vibration level (from level 2 to level 4), it sends an alert to the self-healing safety module. The self-healing module verifies the historical blockchain evidence and triggers a level 2 downgrade after confirming it is not a false alarm. The downgrade instruction is written to the PLC via OPC UA, limiting the pressing pressure to 1.8 MPa.

[0139] In one embodiment, the self-healing security module works in conjunction with the edge computing module. When the edge module detects five consecutive prediction deviations greater than 20%, it flags a level 1 fault. The self-healing module calls a smart contract, records the fault context to the blockchain, switches to a backup random forest model, and updates the edge computing feature weights.

[0140] In one embodiment, the self-healing safety module is linked to the cross-process execution module. The execution module reports a mold jam (level 3 fault). The self-healing module triggers an emergency shutdown and stores operational records from the 10 minutes before the fault. Maintenance tasks are dispatched through the blockchain work order system, and device permissions are locked simultaneously.

[0141] In one embodiment, the self-healing safety module and the hybrid model module perform data verification. After the physical guidance reinforcement learning model generates a control instruction, the SM3 hash is calculated; the instruction content and the hash value are packaged and uploaded to the chain; and before execution, it is verified through the digital twin. If the success rate is <90%, the instruction is frozen and an alarm is issued.

[0142] A carton production dynamic adjustment method based on the above-mentioned carton production dynamic adjustment control system includes:

[0143] Step A, multimodal data collection: multimodal production parameters in normal production are collected based on data collection modules set in the loading process, pressing process, forming process, drying process and quality inspection process;

[0144] Step B, model input extraction: After preprocessing the multimodal production parameters through the edge computing module, the input of the feeding model, the pressing model, the forming model, the drying model, and the quality inspection model are extracted based on the feature extraction algorithm;

[0145] Step C, obtaining process prediction results: Based on the input of the feeding model, the pressing model, the forming model, the drying model, and the quality inspection model, the corresponding inputs are input into the hybrid model module to obtain the model prediction results;

[0146] Step D, obtaining the adjustment target value: The instruction calculation module compares the model prediction results corresponding to each process outputted in Step C with the dynamically generated adjustment thresholds for each process to determine whether adjustment is required. The module then calculates and obtains the adjustment target value for the process requiring adjustment, and generates an adjustment instruction based on the corresponding generated adjustment target value.

[0147] Step E, adjustment instruction execution: The cross-process execution module adjusts the production parameters of the corresponding process equipment based on the output results of the process dynamic model to overcome the process coupling interference, and then executes the adjustment instructions generated by the instruction calculation module to adjust the production equipment of each process;

[0148] Step F: Target refinement and fine-tuning: The cross-process execution module monitors the adjusted production equipment and, after reaching the trigger target threshold, performs a second fine-tuning on the production equipment to improve the adjustment effect. The present invention provides a carton production dynamic adjustment and control system.

[0149] The present invention can effectively collect production parameters in each process of the carton production process through the data acquisition module set in each process, involving mechanical parameters, optical parameters, environmental parameters and material parameters. When it is input into the hybrid model module, the model output results corresponding to different processes are output respectively to represent the carton production trend in the future period, and are transmitted to the instruction calculation module to calculate the adjustment target value of the process to be adjusted and generate adjustment instructions to adjust the corresponding process equipment to ensure the yield rate of carton production; before the cross-process execution module adjusts the process of each process, in order to ensure that the processes avoid mutual interference during the adjustment process, online modeling and real-time decoupling control of multi-process dynamic coupling interference are adopted to further improve the adjustment process and achieve better adjustment effect.

[0150] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A carton production dynamic adjustment and control system, characterized by comprising: The data acquisition module is installed in each process of the carton production process to collect and transmit multimodal production parameters in the loading process, pressing process, forming process, drying process and quality inspection process; An edge computing module is used to receive the multimodal production parameters collected by the data acquisition module, perform data preprocessing, and then use a feature extraction algorithm to sequentially obtain and transmit the input of the feeding model, the pressing model, the forming model, the drying model, and the quality inspection model; A hybrid model module is used to receive the model input transmitted by the edge computing module, input it into the model of the corresponding process, and then output and transmit the model prediction result; An instruction calculation module is used to receive the multimodal production parameters collected by the data acquisition module and the model prediction results output by the hybrid model module, and dynamically generate and transmit adjustment instructions for corresponding processes after comprehensive judgment; The cross-process execution module is used to receive the adjustment instruction transmitted by the instruction calculation module, and according to the adjustment instruction, dynamically adjust the production parameters of the corresponding process before executing the adjustment instruction to overcome the process coupling interference, and then execute the adjustment instruction for adjustment.

2. A carton production dynamic adjustment and control system according to claim 1, characterized in that the hybrid model module comprises: A random forest multi-objective optimization model is used to receive the input of the feeding model, predict the influence trend of raw materials and environmental humidity during the feeding process, and generate a feeding prediction result; An LSTM-PINN hybrid model is used to receive the lamination model input, predict the cardboard deformation trend during the lamination process, and generate a lamination prediction result; a physics-guided reinforcement learning model for receiving the molding model input, predicting the influence trend of mold spacing and optimized conveying speed during the molding process, and generating a molding prediction result; A multivariable predictive control model is used to receive the drying model input, predict the drying efficiency and energy consumption trends during the drying process, and generate a drying prediction result; A graph neural network knowledge graph model is used to receive the input of the quality inspection model, perform defect root cause analysis and quality traceability during the quality inspection process, and generate quality inspection prediction results; The feeding prediction results, pressing prediction results, forming prediction results, drying prediction results and quality inspection prediction results are fitted to generate the model prediction results.

3. A carton production dynamic adjustment and control system according to claim 2, characterized in that the instruction calculation module comprises: an adjustment threshold calculation unit, which dynamically generates an adjustment threshold for each process based on the multimodal production parameters collected by the data collection module; An adjustment determination calculation module compares the feeding prediction result, the pressing prediction result, the forming prediction result, the drying prediction result, and the quality inspection prediction result with the adjustment threshold of the corresponding process to determine whether adjustment is required; The adjustment target value calculation module calculates the corresponding adjustment target value of the process that needs to be adjusted based on the determination result of the adjustment determination calculation module, and generates the adjustment instruction based on the generated adjustment target value.

4. A carton production dynamic adjustment and control system according to claim 3, characterized in that the cross-process execution module includes: an instruction receiving unit, configured to receive the adjustment instruction transmitted by the instruction calculation module and analyze the adjustment target value of the corresponding process; A multi-process coupling modeling unit is used to establish a process dynamic model of the dynamic relationship between multiple processes, and based on the adjustment target value and the output result of the process dynamic model, fine-tune the process parameters in advance to offset the adjustment interference; a program execution module, configured to adjust corresponding process equipment based on the adjustment target value analyzed by the instruction receiving unit; The fine-tuning execution module is used to detect the production parameters of the adjusted process in real time, and perform fine-tuning based on the comparison between the production parameters of the process and the trigger target threshold.

5. A carton production dynamic adjustment and control system according to claim 4, characterized in that the trigger target threshold includes: prediction result deviation value, process coupling interference trigger value, process safety limit value and quality feedback trigger value.

6. A carton production dynamic adjustment and control system according to claim 1, characterized in that the data acquisition module comprises: Data sensing units, installed in each process, are used to collect multimodal data; A data alignment unit, configured to bind and mark the multimodal data collected by the data sensing unit with a time series to perform initial time synchronization; A data transmission unit is used to transmit the multimodal data processed by the data alignment unit.

7. A carton production dynamic adjustment and control system according to claim 6, characterized in that the multimodal production parameters include: mechanical parameters, optical parameters, environmental parameters and material parameters.

8. The carton production dynamic adjustment and control system according to claim 1, wherein the edge computing module comprises: A data receiving unit, configured to receive the multimodal data collected by the data collection module; A data preprocessing unit, configured to perform data cleaning preprocessing on the multimodal data to remove noise and invalid data, and perform time synchronization alignment to achieve secondary time synchronization; a feature extraction unit, configured to extract the feeding model input, the pressing model input, the forming model input, the drying model input, and the quality inspection model input from the multimodal data using a feature extraction algorithm; The feature input unit is used to input the model input quantity extracted by the feature extraction unit into the hybrid model module.

9. A carton production dynamic adjustment and control system according to claim 1, characterized in that it also includes a self-healing safety module for integrating a blockchain instruction evidence unit and a three-level fault degradation control mechanism to respond to emergencies.

10. A carton production dynamic adjustment method executed by a carton production dynamic adjustment control system according to any one of claims 1 to 9, characterized by comprising: Step A, multimodal data collection: multimodal production parameters in normal production are collected based on data collection modules set in the loading process, pressing process, forming process, drying process and quality inspection process; Step B, model input extraction: After preprocessing the multimodal production parameters through the edge computing module, the input of the feeding model, the pressing model, the forming model, the drying model, and the quality inspection model are extracted based on the feature extraction algorithm; Step C, obtaining process prediction results: Based on the input of the feeding model, the pressing model, the forming model, the drying model, and the quality inspection model, the corresponding inputs are input into the hybrid model module to obtain the model prediction results; Step D, obtaining the adjustment target value: The instruction calculation module compares the model prediction results corresponding to each process outputted in Step C with the dynamically generated adjustment thresholds for each process to determine whether adjustment is required. The module then calculates and obtains the adjustment target value for the process requiring adjustment, and generates an adjustment instruction based on the corresponding generated adjustment target value. Step E, adjustment instruction execution: The cross-process execution module adjusts the production parameters of the corresponding process equipment based on the output results of the process dynamic model to overcome the process coupling interference, and then executes the adjustment instructions generated by the instruction calculation module to adjust the production equipment of each process; Step F, target improvement and fine-tuning: The cross-process execution module monitors the adjusted production equipment, and after reaching the trigger target threshold, performs a second fine-tuning on the production equipment to improve the adjustment effect.

Images