A self-adapting foam box forming process parameter real-time matching system

Abstract

The present application relates to the technical field of foam box forming matching, in particular to a self-adaptive raw material characteristic foam box forming process parameter real-time matching system. It comprises a raw material characteristic sensing unit that collects the physical property parameters of the foaming raw material and pre-processes them into a raw material characteristic vector; a process parameter decision unit that dynamically calculates the optimal process parameter combination based on the vector through a pre-trained process parameter prediction model; a forming process regulation unit that converts the optimal parameters into control instructions to drive the execution mechanism; and a quality evaluation unit that calculates the deviation through real-time monitoring data and adjusts the optimal process parameter combination parameters online. The present application builds a process parameter prediction model through a support vector mechanism, dynamically matches the process parameter combination of the raw material characteristics, and solves the problem that the traditional fixed parameter mode cannot adapt to the raw material characteristics and batch fluctuation differences, leading to quality problems such as uneven foam holes and insufficient strength of the foam box, and improves the product quality stability.

Description

Technical Field

[0001] This invention relates to the field of foam box molding matching technology, and more specifically, to a real-time matching system for foam box molding process parameters that adapts to the characteristics of raw materials. Background Technology

[0002] Foam box molding systems are core packaging production equipment that integrates material processing technology and process control logic. They are primarily used to process polymer materials such as EPS, EPE, and EPP into packaging products with heat insulation and cushioning functions through pre-foaming and molding processes. These products are widely used in food transportation, electronic product protection, and pharmaceutical warehousing. However, existing foam box molding systems mostly employ fixed process parameters, making it difficult to adapt to the differences in characteristics of different raw materials (such as different types of foaming agents and modified matrix materials). Furthermore, they cannot respond in real-time to dynamic factors such as batch fluctuations in raw materials and changes in environmental temperature and humidity, easily leading to problems such as uneven cell structure, insufficient strength, and dimensional deviations in the foam boxes. They also suffer from high energy consumption and difficulty in controlling the scrap rate. Therefore, this paper proposes a real-time matching system for foam box molding process parameters that adapts to the characteristics of raw materials. Summary of the Invention

[0003] The purpose of this invention is to provide a real-time matching system for foam box molding process parameters that adapts to the characteristics of raw materials, in order to solve the problem mentioned in the background art that existing foam box molding systems mostly adopt fixed process parameter modes, which are difficult to adapt to the differences in characteristics of different raw materials (such as different types of foaming agents and modified matrix materials).

[0004] To achieve the above objectives, the present invention aims to provide a real-time matching system for foam box molding process parameters that adapts to raw material characteristics, including a raw material characteristic sensing unit. The raw material characteristic sensing unit collects the physical property parameters of the foaming raw material and performs normalization preprocessing to obtain a raw material characteristic vector. The process parameter decision unit dynamically calculates the optimal combination of process parameters, including steam pressure, molding temperature, holding time, and cooling time, based on the raw material characteristic vector and a pre-trained process parameter prediction model. The molding process control unit converts the optimal combination of process parameters into a sequence of control commands for the steam valve, heater and mold motion mechanism through a model predictive control algorithm, thereby driving the actuator to complete the molding process. The quality assessment unit collects real-time temperature and pressure inside the mold cavity, compares them with the corresponding set values ​​in the optimal process parameter combination, calculates the process deviation index, and adjusts the process parameter prediction model online when the process deviation index exceeds a preset threshold.

[0005] As a further improvement to this technical solution, the specific steps involved in the raw material characteristic sensing unit collecting the physical property parameters of the foaming raw material and performing normalization preprocessing are as follows: Real-time acquisition of physical property parameters of foaming raw materials; including chemical composition, particle size distribution and flow characteristics data; The collected physical property parameters were cleaned to remove obvious outliers, and missing data were filled in using the moving average of historical data from the same batch. The maximum and minimum normalization processes were applied to the various physical property parameters after cleaning. The normalized physical property parameters are combined in a fixed order to form a raw material characteristic vector.

[0006] As a further improvement to this technical solution, the process parameter decision unit includes a model building module and a parameter generation module; The model building module uses raw material characteristic vectors collected from historical production batches and the optimal combination of process parameters for successful production to form a training sample set. It uses a support vector machine regression algorithm to build and train the process parameter prediction model offline. The parameter generation module inputs the raw material characteristic vector into the trained process parameter prediction model, and outputs the optimal combination of process parameters after boundary verification based on preset physical constraints.

[0007] As a further improvement to this technical solution, the specific steps involved in constructing and offline training the process parameter prediction model using the support vector machine regression algorithm in the model building module are as follows: The raw material characteristic vectors collected from historical production batches are combined with the optimal process parameter combinations for successful production to form a historical sample set, which is then divided into a training set and a validation set. The process parameters include steam pressure. Molding temperature Holding time and cooldown time ; For each process parameter Each of the support vector machine regression mapping functions based on the radial basis kernel function is constructed, and the support vector machine regression mapping functions of all process parameters constitute the process parameter prediction model. We employ a grid search combined with cross-validation to optimize the hyperparameters of each support vector machine regression mapping function, and train the machine by minimizing the structured risk function. The process parameter prediction model is evaluated using a validation set. The root mean square error (RMSE) of the actual and predicted values ​​is calculated. For any process parameter... All satisfy the root mean square error Below the preset threshold Then, after verification, the trained process parameter prediction model is saved.

[0008] As a further improvement to this technical solution, the specific steps involved in the parameter generation module inputting the raw material characteristic vector into the trained process parameter prediction model are as follows: Vectorize the raw material characteristics of the current batch Input the trained process parameter prediction model, and for each process parameter to be predicted... By using its corresponding support vector machine regression mapping function, the preliminary predicted values ​​of each process parameter are calculated sequentially. The preliminary predicted values ​​of each process parameter are arranged in order to form a preliminary process parameter combination. .

[0009] As a further improvement to this technical solution, the parameter generation module involves the following specific steps in boundary verification based on preset physical constraints: Set physical constraints on steam pressure, molding temperature, holding time, and cooling time to limit each process parameter to between its safe lower limit and safe upper limit; Based on physical constraint boundaries, preliminary process parameter combinations Each process parameter is verified and corrected one by one. If the preliminary predicted value of the process parameter is lower than its lower safety limit, then its lower safety limit value is adopted; if the preliminary predicted value of the process parameter is higher than its upper safety limit, then its upper safety limit value is adopted; otherwise, the original predicted value is retained. The optimal combination of process parameters is obtained by arranging the predicted values ​​of all corrected process parameters in sequence. .

[0010] As a further improvement to this technical solution, the molding process control unit uses a model predictive control algorithm to convert the optimal combination of process parameters into a sequence of control commands for the steam valve, heater, and mold motion mechanism. The specific steps involved are as follows: The state is defined as temperature, pressure and mold position, and the control inputs are steam valve opening, heater power and mold target position. A discrete-time state-space model describing the relationship between the state and the control inputs is established to output measurable temperature and pressure. At the beginning of each control cycle, the prediction time domain of the model predictive control algorithm is set. and control time domain ; The objective function is set to make the predicted temperature and pressure as close as possible to the optimal combination of process parameters. The corresponding set value is used to control the input changes to be gradual. During the optimization process, state safety constraints are set based on physical constraint boundaries to limit temperature and pressure within a set range; Set control input constraints to limit valve opening, heater power, and mold position to ensure they do not exceed the equipment's allowable range; Set a control input change rate constraint to limit the change range of each control step from not exceeding a set threshold. In each control cycle Based on the current state The optimal process parameter combination, including the steam pressure setpoint and molding temperature setpoint, is used to solve an optimization problem through a discrete-time state-space model, under state safety constraints, control input constraints, and control input rate of change constraints, to obtain the future... The optimal control input sequence for each step; Use the first control input in the optimal control input sequence The control commands actually executed in the current cycle include steam valve opening commands, heater power commands, and mold target positions; The control command is converted into a specific execution signal, and after execution, a control cycle is waited for. Update status Then, perform the next round of rolling optimization until the optimal combination of process parameters is achieved, including the pressure holding time setting and the cooling time setting.

[0011] As a further improvement to this technical solution, the quality assessment unit includes a deviation calculation module and a model optimization module; The deviation calculation module collects real-time temperature and pressure, uses a moving average filtering algorithm for smoothing, and calculates the difference with the corresponding set value in the optimal process parameter combination to obtain process deviation indicators, including temperature deviation and pressure deviation. When the process deviation index exceeds the preset threshold, the model optimization module adjusts the process parameter prediction model online through an incremental learning update mechanism.

[0012] As a further improvement to this technical solution, the deviation calculation module uses a moving average filtering algorithm for smoothing and calculates the difference with the corresponding set value in the optimal process parameter combination. The specific steps involved are as follows: Real-time temperature is collected by temperature and pressure sensors installed inside the mold cavity at fixed sampling intervals. and real-time pressure ; Based on real-time temperature and real-time pressure The moving average filtering algorithm is used to calculate the length of... Smoothed temperature within the moving average window and smoothed pressure ; From the optimal combination of process parameters Extract molding temperature set value and steam pressure setpoint Calculate the smoothed temperature and smoothed pressure The absolute deviation between the set value and the temperature deviation is obtained. and pressure deviation , as a process deviation indicator.

[0013] As a further improvement to this technical solution, the specific steps involved in the online parameter adjustment of the process parameter prediction model by the model optimization module through the incremental learning update mechanism are as follows: Set temperature deviation threshold and pressure deviation threshold If the temperature deviation is satisfied Exceeding the temperature deviation threshold Pressure deviation Exceeding the pressure deviation threshold If any one of the conditions is met, the current process deviation index is determined to be outside the allowable range, triggering the following incremental learning and update mechanism; Extract raw material characteristic vectors from the current batch of production. And the combination of process parameters for stable production during the actual molding process. , constitute incremental samples ; The incremental support vector machine learning method is adopted to add incremental samples to the original process parameter prediction model. Perform local retraining and update the support vector machine regression mapping function for each process parameter. The adjustable parameters include a weight vector and a bias term. The updated process parameter prediction model is used to predict the current raw material characteristic vector. Prediction is performed to obtain new combinations of process parameters. ; Check new process parameter combinations If the physical constraint boundary is met, the original model is replaced with the updated process parameter prediction model; if not, the original model is retained and an anomaly log is recorded.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. In this real-time matching system for foam box molding process parameters that adapts to raw material characteristics, multi-dimensional physical property parameters such as chemical composition and particle size distribution are collected and preprocessed into raw material characteristic vectors. Combined with support vector machine regression algorithm, a process parameter prediction model is constructed. Compared with the traditional fixed process parameter mode, it fully adapts to the differences in raw material characteristics and batch fluctuations, and solves the quality problems such as uneven foam cell structure and insufficient strength caused by the inability of parameters to dynamically match raw material characteristics in traditional systems, thereby improving the stability of product quality.

[0015] 2. In this real-time matching system for foam box molding process parameters that adapts to the characteristics of raw materials, the temperature and pressure data inside the mold cavity are collected in real time. After processing with a moving average filtering algorithm, the deviation index is calculated. When the deviation exceeds the threshold, the model parameters are updated through incremental learning. Compared with the traditional method that relies on manual debugging or offline optimization, the online dynamic correction of process parameters can be achieved without manual intervention. This solves the problems of lag in parameter adjustment and poor adaptability in traditional systems, and can continuously optimize the molding effect and adapt to different production conditions. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the overall process of the present invention.

[0017] The meanings of the labels in the diagram are as follows: 1. Raw material characteristic sensing unit; 2. Process parameter decision-making unit; 21. Model building module; 22. Parameter generation module; 3. Molding process control unit; 4. Quality assessment unit; 41. Deviation calculation module; 42. Model optimization module. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example

[0019] Please see Figure 1 As shown, a real-time matching system for foam box molding process parameters that adapts to raw material characteristics is provided. The system includes a raw material characteristic sensing unit 1, which collects the physical property parameters of the foaming raw material and performs normalization preprocessing to obtain a raw material characteristic vector. In this embodiment, the physical property parameters of the foaming raw material are collected in real time through sensors and online detection devices; wherein, the physical property parameters include chemical composition, particle size distribution and flow characteristics data; Specifically, the chemical composition includes the pentane content and residual monomer ratio in expandable polystyrene; the particle size distribution includes the d10 characteristic particle size, d50 characteristic particle size, and d90 characteristic particle size, which are measured by a laser particle size analyzer; the flow characteristics include flow parameters such as bulk density and angle of repose, which are obtained by a flowability tester. The collected physical property parameters are cleaned to remove obvious outliers (such as values ​​that exceed the reasonable physical range), and missing data are filled in using the moving average of historical data from the same batch. The maximum and minimum normalization processes were applied to the various physical property parameters after cleaning. For example, regarding pentane content, assuming the minimum value in the training data is 5% (mass fraction) and the maximum value is 8%, then when the pentane content of a certain batch of raw materials is 6.5%, the normalized value is... ; The normalized physical property parameters are combined in a fixed order to form a raw material property vector; ; in, This is a vector of raw material properties, with each term corresponding to a normalized physical property parameter; This represents the normalized pentane content. The normalized 10% is the particle size (d10). The median particle size (d50) is the normalized value. The normalized 90% is the particle size (d90). This is the normalized packing density. The angle of rest after normalization.

[0020] The adaptive material characteristic foam box molding process parameter real-time matching system also includes a process parameter decision unit 2. The process parameter decision unit 2 dynamically calculates the optimal combination of process parameters, including steam pressure, molding temperature, holding time and cooling time, based on the material characteristic vector and through a pre-trained process parameter prediction model. In this embodiment, the process parameter decision unit 2 includes a model building module 21 and a parameter generation module 22; Among them, the model building module 21 uses the raw material characteristic vectors collected in historical production batches and the optimal process parameter combination of the corresponding successful production to form a training sample set, and uses the support vector machine regression algorithm to build and train the process parameter prediction model offline. Specifically, raw material characteristic vectors collected from historical production batches are combined with the optimal process parameter combinations for successful production to form a historical sample set, which is then divided into a training set and a validation set; among which, process parameters include steam pressure. (Unit: bar) Molding temperature (Unit: °C), Holding Time (Unit: s) and cooling time (Unit: s); ; in, For training set; For the first The raw material characteristic vector corresponding to each historical sample; For the first The optimal combination of process parameters corresponding to each historical sample; The total number of samples in the training set; For each process parameter Each of the support vector machine regression mapping functions based on the radial basis kernel function is constructed, and the support vector machine regression mapping functions of all process parameters constitute the process parameter prediction model. Support Vector Machine Regression Mapping Function: ; in, Corresponding process parameters The support vector machine regression mapping function; For process parameters The weight vector; For process parameters The bias term; This represents the historical raw material characteristic vector; For kernel function mapping, this system uses the radial basis kernel function: ; in, For the first The sample and the first The kernel function calculation value of the raw material characteristic vector of each sample; For the first The raw material characteristic vector of each historical sample (i.e. ), For the first The raw material characteristic vector of each historical sample (i.e. ); These are the kernel coefficients of the radial basis kernel function; This indicates the calculation of Euclidean distance; A grid search combined with cross-validation method is used to optimize the hyperparameters (penalty coefficients) of the regression mapping function for each support vector machine. kernel coefficient Training is performed by minimizing the structured risk function: ; in, For process parameters The penalty coefficient is used to balance the weights of the regularization term and the loss term; For process parameters The true value is taken from the optimal combination of process parameters in the training sample set. ; To determine the insensitive loss threshold, take... ;Grid search combined with cross-validation selects the optimal hyperparameter configuration by traversing preset hyperparameter combinations and evaluating model performance on different data partitions. The process parameter prediction model is evaluated using a validation set. The root mean square error (RMSE) of the actual and predicted values ​​is calculated. For any process parameter... All satisfy the root mean square error Below the preset threshold (For example If the process parameter prediction model is verified, the trained model can be saved. The verification standard is root mean square error: ; in, To determine the number of samples in the validation set, Process parameters output by the process parameter prediction model The predicted value; For process parameters The root mean square error.

[0021] The parameter generation module 22 inputs the raw material characteristic vector into the trained process parameter prediction model, and after boundary verification based on preset physical constraints (such as upper limit of steam pressure and safe temperature range), outputs the optimal combination of process parameters. Specifically, the raw material characteristic vector of the current batch Input the trained process parameter prediction model, and for each process parameter to be predicted... Through its corresponding support vector machine regression mapping function The preliminary predicted values ​​of each process parameter are calculated sequentially. ; ; ; ; in, The support vector machine regression mapping function is the steam pressure. These are the support vectors in the steam pressure model. This represents the number of support vectors in the steam pressure model. These are the Lagrange multipliers corresponding to the steam pressure model. This is the bias term for the steam pressure model; The support vector machine regression mapping function is the molding temperature. These are the support vectors in the molding temperature model. This represents the number of support vectors in the molding temperature model. The Lagrange multiplier corresponding to the molding temperature model. This is a bias term for the molding temperature model; The support vector machine regression mapping function is the holding time. These are the support vectors in the holding time model. This represents the number of support vectors in the holding time model. The Lagrange multipliers corresponding to the holding time model, This is the bias term for the holding time model; The support vector machine regression mapping function has a cooling time. These are the support vectors in the cooling time model. This represents the number of support vectors in the cooling time model. For the Lagrange multipliers corresponding to the cooling time model, This is the bias term for the cooling time model; This is a preliminary prediction of the steam pressure. This is a preliminary prediction of the molding temperature. This is a preliminary prediction of the pressure holding time. This is a preliminary prediction of the cooling time; This is the radial basis kernel function in the vapor pressure model. This refers to the radial basis kernel function in the molding temperature model. For the radial basis kernel function in the holding time model, This refers to the radial basis kernel function in the cooling time model; The preliminary predicted values ​​of each process parameter are arranged in order to form a preliminary process parameter combination. ; ; in, This is a preliminary combination of process parameters, reflecting the parameter configuration that the process parameter prediction model infers based on the current raw material characteristics and is the best match in historical optimal production experience.

[0022] To prevent the output of the process parameter prediction model from exceeding the equipment safety range or the process feasibility range, the preliminary process parameter combination is subjected to boundary verification and correction based on physical constraints. The constraints are set according to the actual production line equipment specifications, mold material tolerance, and process safety standards. The specific steps are as follows: Set physical constraints on steam pressure, molding temperature, holding time, and cooling time to limit each process parameter to between its safe lower limit and safe upper limit; Steam pressure : , The lower limit of safe steam pressure. For example, the safe upper limit of steam pressure The strength of the mold structure is determined by the boiler's supply capacity and the mold's structural strength. Molding temperature : , To the lower limit of the molding temperature safety, For example, the upper limit of the molding temperature safety limit It is necessary to ensure that the raw materials are fully softened and foamed while avoiding overheating and degradation; Holding time : , To ensure a safe lower limit for pressure holding time, For example, the upper limit of the pressure holding time. This ensures that the molding process is complete and does not affect the production cycle. Cooldown time : , To ensure a safe lower limit for cooldown time, For example, to ensure a safe upper limit for cooldown time This ensures that the product is fully shaped and can be easily demolded; Based on physical constraint boundaries, preliminary process parameter combinations Each process parameter is verified and corrected one by one. If the preliminary predicted value of the process parameter is lower than its lower safety limit, then its lower safety limit value is adopted; if the preliminary predicted value of the process parameter is higher than its upper safety limit, then its upper safety limit value is adopted; otherwise, the original predicted value is retained. The verification and correction process is as follows: ; in, The final value for each process parameter, For the initial combination of process parameters A certain process parameter value; This is the lower limit of the safety parameters for this process. The upper limit of the safety of this process parameter is determined by the physical constraint boundary; The optimal combination of process parameters is obtained by arranging the predicted values ​​of all corrected process parameters in sequence. ; ; in, For the optimal combination of process parameters, Set the steam pressure value. This is the set value for the molding temperature. Set the pressure holding time value. Set the cooldown time value.

[0023] The real-time matching system for foam box molding process parameters that adapts to the characteristics of raw materials also includes a molding process control unit 3. The molding process control unit 3 converts the optimal combination of process parameters into a sequence of control commands for steam valves, heaters and mold motion mechanisms through a model predictive control algorithm, thereby driving the actuators to complete the molding process. In this embodiment, based on the physical process of foam box forming, the state is defined as temperature, pressure and mold position, and the control inputs are steam valve opening, heater power and mold target position. A discrete-time state-space model describing the relationship between the state and the control inputs is established to output measurable temperature and pressure. The discrete-time state-space model is based on the principles of energy balance and mass balance, and uses a first-order lag form to approximate the actual dynamics: ; ; in, This indicates the current control cycle number, and the length of each cycle is [length missing]. (For example ); The state vector for the current control cycle includes the real-time temperature inside the mold cavity. (Unit: °C), Real-time Pressure (Unit: bar) Current position of the mold (Unit: mm); The control input vector for the current control cycle includes the steam valve opening. (Unit: %) Heater power (Unit: %) Target position of mold motion mechanism (Unit: mm); The output vector for the current control cycle, including the measurable temperature. With pressure ; The state matrix, For the input matrix, The output matrix is ​​obtained by identifying historical data or estimating based on physical parameters, and reflects the dynamic characteristics of the system. At the beginning of each control cycle, the prediction time domain of the model predictive control algorithm is set. (For example (Step) and control time domain (For example step); in, To predict the time-domain step size, To control the time-domain step size; The objective function is set to make the predicted temperature and pressure as close as possible to the optimal combination of process parameters. The corresponding set value is used to control the input change smoothly and save energy. Objective function: ; in, For the first The objective function value within each control cycle; For reference output, namely the molding temperature setpoint and steam pressure setpoint in the optimal combination of process parameters; For the first The change in control input over each control cycle; For the tracking error weight matrix, To control the weight matrix of input changes, typically Take the larger value of the diagonal elements (e.g.) ), Take the smaller value (e.g.) Prioritize ensuring the tracking of process parameters; During the optimization process, state safety constraints are set based on physical constraint boundaries to limit temperature and pressure within a set range; State safety constraints: ; Set control input constraints to limit valve opening, heater power, and mold position to ensure they do not exceed the equipment's allowable range; Control input constraints: ; in, For the steam valve opening, For heater power, The target position of the mold motion mechanism. This is the minimum value for the mold position. This represents the maximum value of the mold position. Set a control input change rate constraint to limit the change amplitude of each control step to no more than a set threshold, thus avoiding equipment impact. Control input rate of change constraint: ; in, This represents the change in the opening degree of the steam valve. This represents the change in heater power. This represents the change in the target position of the mold motion mechanism; In each control cycle Based on the current state The optimal process parameter combination, including the steam pressure setpoint and molding temperature setpoint, is used to solve an optimization problem through a discrete-time state-space model, under state safety constraints, control input constraints, and control input rate of change constraints, to obtain the future... The optimal control input sequence for each step; Optimization issues: ; ;

[0024] in, For future control input sequences; To solve the optimization problem and obtain the optimal control input sequence; Use the first control input in the optimal control input sequence The control commands actually executed in the current cycle include steam valve opening commands, heater power commands, and mold target positions; ; in, It is the first control input in the optimal control input sequence. This is a command to instruct the steam valve to open. For heater power command, The target position for the mold; Convert control commands into specific execution signals: Steam valve opening command After D / A conversion, the proportional valve is driven to regulate the steam flow rate; Heater power command The heater output power is adjusted using a solid-state relay; Mold target position Send the signal to the servo driver to control the mold motion mechanism to move to the specified position; After execution, wait for one control cycle. Update status Then proceed to the next round of rolling optimization until the pressure holding time setting value in the optimal combination of process parameters is reached. and cooling time setting value .

[0025] The real-time matching system for foam box molding process parameters that adapts to raw material characteristics also includes a quality assessment unit 4. The quality assessment unit 4 collects the real-time temperature and pressure inside the mold cavity, compares them with the corresponding set values ​​in the optimal process parameter combination, calculates the process deviation index, and adjusts the process parameter prediction model online when the process deviation index exceeds the preset threshold. In this embodiment, the quality assessment unit 4 includes a deviation calculation module 41 and a model optimization module 42; Among them, the deviation calculation module 41 collects real-time temperature and real-time pressure through temperature sensors and pressure sensors installed in the mold cavity, performs smoothing processing using a moving average filtering algorithm, and calculates the difference with the corresponding set value in the optimal process parameter combination to obtain process deviation indicators, including temperature deviation and pressure deviation. Specifically, real-time temperature is collected using temperature sensors (such as thermocouples) and pressure sensors installed inside the mold cavity at a fixed sampling period (such as 0.1 seconds). and real-time pressure ; Based on real-time temperature and real-time pressure The moving average filtering algorithm is used to calculate the length of... (like Smoothed temperature within a moving average window (number of sampling points) and smoothed pressure ; ; ; in, For the current moment Forward Real-time temperature values ​​at each sampling point For the current moment Forward Real-time pressure values ​​at each sampling point; The moving average window length is set based on the actual noise level and control response speed. The temperature after smoothing. The pressure after smoothing; From the optimal combination of process parameters Extract molding temperature set value and steam pressure setpoint Calculate the smoothed temperature and smoothed pressure The absolute deviation between the set value and the temperature deviation is obtained. and pressure deviation As a process deviation indicator: ; ; in, Temperature deviation, in °C; Pressure deviation, in bar.

[0026] When the process deviation index exceeds the preset threshold, the model optimization module 42 adjusts the process parameter prediction model online through an incremental learning update mechanism. Set temperature deviation threshold (like and pressure deviation threshold (like If the temperature deviation is satisfied Exceeding the temperature deviation threshold Pressure deviation Exceeding the pressure deviation threshold If any one of the conditions is met, the current process deviation index is determined to be outside the allowable range, triggering the following incremental learning and update mechanism; Extract raw material characteristic vectors from the current batch of production. And the combination of process parameters for stable production during the actual molding process. , constitute incremental samples : ; in, To achieve a stable combination of process parameters that yields qualified products in actual control, This is the actual value of the steam pressure. This is the actual value of the molding temperature. This is the actual value of the pressure holding time. This is the actual cooling time value; This is the raw material characteristic vector for the current batch; For incremental samples; The incremental support vector machine learning method is adopted to add incremental samples to the original process parameter prediction model. Perform local retraining and update the support vector machine regression mapping function for each process parameter. The adjustable parameters are used without retraining on all historical data; the adjustable parameters include the weight vector and the bias term. The updated support vector machine regression mapping function is: ; in, This is the updated support vector machine regression mapping function; For the original mapping function, These are the Lagrange multipliers in the original model. This is the bias term of the original model. This represents the original number of support vectors; The incremental adjustment term is derived from the incremental samples and applied to the original function. It is obtained by solving an optimization problem with the new samples, ensuring that the new samples are correctly fitted while preserving the original model structure as much as possible. The specific form is: ; in, To increase the number of support vectors; This represents the Lagrange multiplier increment corresponding to the newly added sample; This is the increment of the bias term; For radial basis kernel functions; Incremental parameters and This was obtained by solving the following optimization problem: ; ; ; ; in, Let Lagrange multiplier vectors be used. For bias terms, For incremental samples Actual values ​​of process parameters; The threshold for insensitive loss is set to 0.05. This is the penalty coefficient; This is the first slack variable, used to handle situations where the predicted value is lower than the actual value; This is the second slack variable, used to handle situations where the predicted value is higher than the actual value; By solving the above optimization problem, we obtain and This allows for the updating of model parameters and the completion of online learning. The updated process parameter prediction model is used to predict the current raw material characteristic vector. Prediction is performed to obtain new combinations of process parameters. ; Check new process parameter combinations If the physical constraint boundary is met, the original model is replaced with the updated process parameter prediction model; if not, the original model is retained and an anomaly log is recorded.

[0027] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A real-time adaptive foam box forming process parameter matching system for raw material characteristics, characterized in that, include: Raw material characteristic sensing unit (1) collects the physical property parameters of foaming raw materials and performs normalization preprocessing to obtain raw material characteristic vector; The process parameter decision unit (2) dynamically calculates the optimal combination of process parameters, including steam pressure, molding temperature, holding time and cooling time, based on the raw material characteristic vector and through a pre-trained process parameter prediction model. The molding process control unit (3) converts the optimal combination of process parameters into a sequence of control commands for the steam valve, heater and mold motion mechanism through a model predictive control algorithm, thereby driving the actuator to complete the molding process. The quality assessment unit (4) collects the real-time temperature and pressure inside the mold cavity, compares them with the corresponding set values ​​in the optimal process parameter combination, calculates the process deviation index, and adjusts the process parameter prediction model online when the process deviation index exceeds the preset threshold.

2. The real-time matching system for foam box molding process parameters that adapts to raw material characteristics according to claim 1, characterized in that: The specific steps involved in the raw material property sensing unit (1) collecting the physical property parameters of the foaming raw material and performing normalization preprocessing are as follows: Real-time acquisition of physical property parameters of foaming raw materials; Among them, physical property parameters include chemical composition, particle size distribution and flow characteristics data; The collected physical property parameters were cleaned to remove obvious outliers, and missing data were filled in using the moving average of historical data from the same batch. The maximum and minimum normalization processes were applied to the various physical property parameters after cleaning. The normalized physical property parameters are combined in a fixed order to form a raw material characteristic vector.

3. The real-time matching system for foam box molding process parameters that adapts to raw material characteristics according to claim 2, characterized in that: The process parameter decision unit (2) includes a model building module (21) and a parameter generation module (22). Among them, the model building module (21) uses the raw material characteristic vector collected in historical production batches and the optimal process parameter combination of the corresponding successful production to form a training sample set, and uses the support vector machine regression algorithm to build and train the process parameter prediction model offline. The parameter generation module (22) inputs the raw material characteristic vector into the trained process parameter prediction model, and outputs the optimal combination of process parameters after boundary verification based on preset physical constraints.

4. The real-time matching system for foam box molding process parameters that adapts to raw material characteristics according to claim 3, characterized in that: The specific steps involved in the model building module (21) using the support vector machine regression algorithm to build and train the process parameter prediction model offline are as follows: The raw material characteristic vectors collected from historical production batches are combined with the optimal process parameter combinations for successful production to form a historical sample set, which is then divided into a training set and a validation set. The process parameters include steam pressure. Molding temperature Holding time and cooldown time ; For each process parameter Each of the support vector machine regression mapping functions based on the radial basis kernel function is constructed, and the support vector machine regression mapping functions of all process parameters constitute the process parameter prediction model. We employ a grid search combined with cross-validation to optimize the hyperparameters of each support vector machine regression mapping function, and train the machine by minimizing the structured risk function. The process parameter prediction model is evaluated using a validation set. The root mean square error (RMSE) of the actual and predicted values ​​is calculated. For any process parameter... All satisfy the root mean square error Below the preset threshold Then, after verification, the trained process parameter prediction model is saved.

5. The real-time matching system for foam box molding process parameters with adaptive raw material characteristics according to claim 3, characterized in that: The specific steps involved in the parameter generation module (22) inputting the raw material characteristic vector into the trained process parameter prediction model are as follows: Vectorize the raw material characteristics of the current batch Input the trained process parameter prediction model, and for each process parameter to be predicted... By using its corresponding support vector machine regression mapping function, the preliminary predicted values ​​of each process parameter are calculated sequentially. The preliminary predicted values ​​of each process parameter are arranged in order to form a preliminary process parameter combination. .

6. The real-time matching system for foam box molding process parameters that adapts to raw material characteristics according to claim 5, characterized in that: The parameter generation module (22) involves the following specific steps in boundary verification based on preset physical constraints: Set physical constraints on steam pressure, molding temperature, holding time, and cooling time to limit each process parameter to between its safe lower limit and safe upper limit; Based on physical constraint boundaries, preliminary process parameter combinations Each process parameter is verified and corrected one by one. If the preliminary predicted value of the process parameter is lower than its lower safety limit, then its lower safety limit value is adopted; if the preliminary predicted value of the process parameter is higher than its upper safety limit, then its upper safety limit value is adopted. Otherwise, retain the original predicted value; The optimal combination of process parameters is obtained by arranging the predicted values ​​of all corrected process parameters in sequence. .

7. The real-time matching system for foam box molding process parameters with adaptive raw material characteristics according to claim 1, characterized in that: The molding process control unit (3) uses a model predictive control algorithm to convert the optimal combination of process parameters into a sequence of control commands for the steam valve, heater, and mold motion mechanism. The state is defined as temperature, pressure and mold position, and the control inputs are steam valve opening, heater power and mold target position. A discrete-time state-space model describing the relationship between the state and the control inputs is established to output measurable temperature and pressure. At the beginning of each control cycle, the prediction time domain of the model predictive control algorithm is set. and control time domain ; The objective function is set to make the predicted temperature and pressure as close as possible to the optimal combination of process parameters. The corresponding set value is used to control the input changes to be gradual. During the optimization process, state safety constraints are set based on physical constraint boundaries to limit temperature and pressure within a set range; Set control input constraints to limit valve opening, heater power, and mold position to ensure they do not exceed the equipment's allowable range; Set a control input change rate constraint to limit the change range of each control step from not exceeding a set threshold. In each control cycle Based on the current state The optimal process parameter combination, including the steam pressure setpoint and molding temperature setpoint, is used to solve an optimization problem through a discrete-time state-space model, under state safety constraints, control input constraints, and control input rate of change constraints, to obtain the future... The optimal control input sequence for each step; Use the first control input in the optimal control input sequence The control commands actually executed in the current cycle include steam valve opening commands, heater power commands, and mold target positions; The control command is converted into a specific execution signal, and after execution, a control cycle is waited for. Update status Then, perform the next round of rolling optimization until the optimal combination of process parameters is achieved, including the pressure holding time setting and the cooling time setting.

8. The real-time matching system for foam box molding process parameters with adaptive raw material characteristics according to claim 1, characterized in that: The quality assessment unit (4) includes a deviation calculation module (41) and a model optimization module (42). Among them, the deviation calculation module (41) collects real-time temperature and real-time pressure, uses the moving average filtering algorithm for smoothing, and calculates the difference with the corresponding set value in the optimal process parameter combination to obtain the process deviation index, including temperature deviation and pressure deviation. When the process deviation index exceeds the preset threshold, the model optimization module (42) adjusts the process parameter prediction model online through an incremental learning update mechanism.

9. The real-time matching system for foam box molding process parameters that adapts to raw material characteristics according to claim 8, characterized in that: The specific steps involved in the deviation calculation module (41) using a moving average filtering algorithm for smoothing and calculating the difference with the corresponding set value in the optimal process parameter combination are as follows: Real-time temperature is collected by temperature and pressure sensors installed inside the mold cavity at fixed sampling intervals. and real-time pressure ; Based on real-time temperature and real-time pressure The moving average filtering algorithm is used to calculate the length of... Smoothed temperature within the moving average window and smoothed pressure ; From the optimal combination of process parameters Extract the molding temperature set value and steam pressure setpoint Calculate the smoothed temperature and smoothed pressure The absolute deviation between the set value and the temperature deviation is obtained. and pressure deviation , as a process deviation indicator.

10. The real-time matching system for foam box molding process parameters with adaptive raw material characteristics according to claim 8, characterized in that: The specific steps involved in the online parameter adjustment of the process parameter prediction model by the model optimization module (42) through the incremental learning update mechanism are as follows: Set temperature deviation threshold and pressure deviation threshold If the temperature deviation is satisfied Exceeding the temperature deviation threshold Pressure deviation Exceeding the pressure deviation threshold If any one of the conditions is met, the current process deviation index is determined to be outside the allowable range, triggering the following incremental learning and update mechanism; Extract raw material characteristic vectors from the current batch of production. And the combination of process parameters for stable production during the actual molding process. , constitute incremental samples ; The incremental support vector machine learning method is adopted to add incremental samples to the original process parameter prediction model. Perform local retraining and update the support vector machine regression mapping function for each process parameter. The adjustable parameters include a weight vector and a bias term. The updated process parameter prediction model is used to predict the current raw material characteristic vector. Prediction is performed to obtain new combinations of process parameters. ; Check new process parameter combinations If the physical constraint boundary is met, the original model is replaced with the updated process parameter prediction model; if not, the original model is retained and an anomaly log is recorded.

Images