A Predictive Optimization Control Method and System for Tire Vulcanization Process Based on Process Constraints

Abstract

This disclosure provides a predictive optimization control method and system for tire vulcanization process based on process constraints, relating to the field of intelligent tire manufacturing control technology. It includes employing a hierarchical predictive collaborative optimization control framework to achieve multivariate collaborative rolling control of vulcanization kinetics and external temperature disturbances. In the upper offline stage, a neural network integrating vulcanization kinetics and process constraints is constructed to generate a baseline temperature trajectory that combines optimality and physical feasibility, providing a feasible initial solution for the lower tracking layer. Subsequently, in the lower online control stage, an event-triggered mechanism drives a sparse Bayesian learner to update the coupled model online. The feedforward compensator then quickly parses the control command based on this model and directly adjusts the steam valve, thereby offsetting coupling interference in advance. This disclosure can effectively suppress temperature disturbances and improve system robustness.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of intelligent manufacturing control of tires, in particular to a tire vulcanization process prediction and optimization control method and system based on process constraints. BACKGROUND

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute the prior art.

[0003] In the tire manufacturing process, the vulcanization step is the last step and also the most critical step, which directly affects the performance and service life of the tire. The quality of tire vulcanization is directly related to the safety and durability of the tire, therefore, it is crucial to optimize the control method of the tire vulcanization process.

[0004] There are several major problems in the existing tire vulcanization process. First, the raw materials used for production may vary, and the molecular structure and reactivity of different rubbers are different, so the required vulcanization temperature is also different, therefore, it is impossible to use a fixed temperature curve to vulcanize all types of tires; second, high-temperature steam is usually used as a heat source in the vulcanization process, and the steam flows in a complex pipe network, the fluctuations in pressure, flow and dryness of the steam will affect the heat exchange efficiency of the entire system, therefore, precise control of steam supply is crucial to ensure tire quality; in addition, in the vulcanization reaction, the control of vulcanization temperature is crucial, too high temperature will cause over-vulcanization, and too low temperature will cause under-vulcanization, therefore, to ensure the uniformity of the overall vulcanization of the tire, the temperature must be precisely controlled to avoid the problem of local over-vulcanization or under-vulcanization.

[0005] The existing control method of tire vulcanization process generally uses PID control or traditional model predictive control (MPC) based on fixed temperature curve. These methods have inherent defects:

[0006] (1): unable to adapt to changes in rubber and specifications, fixed parameter control is easy to cause over-vulcanization or under-vulcanization of different batches of products, poor quality consistency;

[0007] (2): difficult to overcome the complex thermal coupling of multiple temperature zones, resulting in large temperature fluctuations and low control precision;

[0008] (3): weak resistance to steam disturbance, feedback control has a lag;

[0009] (4): lack of economic optimization mechanism, conservative process is adopted to ensure safety, resulting in energy waste and efficiency loss. SUMMARY

[0010] The present disclosure proposes a process constraint-based tire vulcanization process prediction optimization control method and system to solve the above problems, which includes an offline stage and an online stage. In the offline stage, a physical information temperature model integrating vulcanization dynamics and process constraints is constructed to generate a benchmark temperature trajectory with optimal performance and physical feasibility, providing a feasible initial solution for the lower tracking layer. In the online control stage, the sparse Bayesian learner is driven by an event-triggered mechanism to update the coupled model online, and then the control command is quickly analyzed by the feedforward compensator to directly adjust the steam valve, thereby offsetting the coupling disturbance in advance, effectively suppressing temperature disturbance, and improving system robustness.

[0011] According to some embodiments, the present disclosure adopts the following technical solutions:

[0012] The process constraint-based tire vulcanization process prediction optimization control method includes that the method realizes multivariable collaborative rolling control of vulcanization dynamics and external temperature disturbance by using a hierarchical prediction collaborative optimization control framework.

[0013] The hierarchical prediction collaborative optimization control framework includes an upper-layer offline stage and a lower-layer online control stage. In the upper-layer offline stage, a multi-objective process constraint optimization loss function is established according to historical working condition parameters, and a temperature trajectory prediction model is trained. In the lower-layer online control stage, a temperature set point trajectory is generated based on the temperature trajectory prediction model trained in the upper-layer offline stage according to real-time working condition parameters, a thermal coupling model is tracked by using model predictive control and feedforward compensation to track the temperature set point trajectory, and the thermal coupling model is updated online by a sparse Bayesian learner driven by an event-triggered mechanism to ensure that the optimal temperature trajectory is accurately tracked while solving the steam valve opening control amount.

[0014] According to the steam valve opening control amount, a control command is issued to realize the multivariable collaborative rolling control process.

[0015] According to some embodiments, the present disclosure adopts the following technical solutions:

[0016] The process constraint-based tire vulcanization process prediction optimization control system includes:

[0017] The collaborative optimization module is configured to realize multivariable collaborative rolling control of vulcanization dynamics and external temperature disturbance by using a hierarchical prediction collaborative optimization control framework.

[0018] The hierarchical predictive collaborative optimization control framework includes an upper-level offline stage and a lower-level online control stage. In the upper-level offline stage, a multi-objective process constraint optimization loss function is established based on historical operating parameters, and a temperature trajectory prediction model is trained. In the lower-level online control stage, a temperature setpoint trajectory is generated based on the temperature trajectory prediction model trained in the upper-level offline stage according to real-time operating parameters. A thermally coupled model of model predictive control and feedforward compensation is used to track the temperature setpoint trajectory. An event-triggered mechanism drives a sparse Bayesian learner to update the thermally coupled model online, ensuring that the optimal temperature trajectory is accurately tracked while solving the steam valve opening control quantity.

[0019] The control module is used to issue control commands based on the steam valve opening control quantity to realize a multi-variable collaborative rolling control process.

[0020] According to some embodiments, the present disclosure adopts the following technical solutions:

[0021] A computer program product includes a computer program that, when executed by a processor, implements the aforementioned process-constrained predictive optimization control method for tire vulcanization process.

[0022] According to some embodiments, the present disclosure adopts the following technical solutions:

[0023] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned predictive optimization control method for tire vulcanization process based on process constraints.

[0024] According to some embodiments, the present disclosure adopts the following technical solutions:

[0025] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the aforementioned process-constrained tire vulcanization process prediction and optimization control method.

[0026] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0027] This disclosed method for predictive optimization control of tire vulcanization process based on process constraints mainly includes an upper offline stage and a lower online control stage. In the offline stage, a physical information neural network (temperature trajectory prediction model) generates a temperature setpoint trajectory based on real-time operating conditions, customizing the most suitable vulcanization temperature for each batch of tires, thus solving the problem that a fixed temperature curve cannot adapt to changes in raw materials that affect vulcanization quality. In the online control stage, a strategy combining model predictive control (MPC) and feedforward compensation is adopted. Through an event-triggered mechanism, a sparse Bayesian learning algorithm is intelligently driven to identify and update the weight parameters representing the thermo-coupling relationship online. MPC uses high-precision temperature sensor measurements as the system state and performs multivariate optimization calculations based on the real-time coupling model. At the same time, the feedforward compensator generates control commands in advance based on the coupling relationship, forming a composite control with the feedback output of MPC. This control structure effectively suppresses thermo-coupling interference between multiple temperature zones and achieves precise and coordinated control of the temperature of each temperature zone.

[0028] The disclosed method for predictive optimization control of tire vulcanization process based on process constraints uses a physical information neural network (PINN) as a temperature trajectory prediction model in the offline stage. The PINN dynamically generates a temperature setting trajectory based on real-time operating parameters (such as rubber activation energy and tire thickness), which solves the problem that fixed parameter control cannot adapt to changes in raw materials and ensures the consistency of vulcanization quality of different batches of products from the source.

[0029] The disclosed method for predictive optimization control of tire vulcanization process based on process constraints, in the online control stage, drives a sparse Bayesian learner to update the coupled model online through an event triggering mechanism. Then, the feedforward compensator quickly parses the control command according to the model and directly adjusts the steam valve, thereby offsetting the coupling interference in advance, effectively suppressing temperature disturbances, and improving the robustness of the system.

[0030] The disclosed method for predictive optimization control of tire vulcanization process based on process constraints utilizes the lower-level MPC in the online control stage to perform multivariate collaborative optimization using its internal model (explicitly including thermal coupling relationships). Through a mechanism combining feedback correction and feedforward compensation based on direct measurement, disturbances are suppressed in real time, overcoming the thermal coupling effect in the temperature range and improving the accuracy of temperature control and system stability. In the offline stage, PINN embeds physical constraints in trajectory generation, while MPC also handles safety constraints in rolling optimization. While pursuing control performance (such as tracking accuracy), it ensures that the system operation is always within the process safety boundary, avoiding the risk of over-temperature. Attached Figure Description

[0031] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0032] Figure 1 This is a flowchart of a tire vulcanization process prediction and optimization control method based on process constraints according to an embodiment of this disclosure;

[0033] Figure 2 This is a schematic diagram of the hierarchical predictive collaborative optimization control framework architecture of the tire vulcanization process prediction and optimization control method based on process constraints according to an embodiment of this disclosure. Detailed Implementation

[0034] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0035] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0036] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0037] Example 1

[0038] One embodiment of this disclosure provides a predictive optimization control method for tire vulcanization process based on process constraints, the method comprising:

[0039] A hierarchical predictive collaborative optimization control framework is adopted to achieve multivariable collaborative rolling control of sulfurization kinetics and external temperature disturbances;

[0040] The hierarchical predictive collaborative optimization control framework includes an upper-level offline stage and a lower-level online control stage. In the upper-level offline stage, a multi-objective process constraint optimization loss function is established based on historical operating parameters, and a temperature trajectory prediction model is trained. In the lower-level online control stage, a temperature setpoint trajectory is generated based on the temperature trajectory prediction model trained in the upper-level offline stage according to real-time operating parameters. A thermally coupled model of model predictive control and feedforward compensation is used to track the temperature setpoint trajectory. An event-triggered mechanism drives a sparse Bayesian learner to update the thermally coupled model online, ensuring that the optimal temperature trajectory is accurately tracked while solving the steam valve opening control quantity.

[0041] Control commands are issued based on the steam valve opening control quantity to realize a multi-variable collaborative rolling control process.

[0042] As one embodiment, the tire vulcanization process prediction and optimization control method disclosed herein based on process constraints firstly constructs a physical information neural network integrating vulcanization kinetics and process constraints in the upper offline stage to generate a temperature setpoint trajectory that combines optimality and physical feasibility. This provides a feasible initial solution for the tracking layer in the lower online control stage. In the lower online control stage, an event-triggered mechanism drives a sparse Bayesian learner to update the coupled model online. Subsequently, a feedforward compensator quickly parses the control command based on the model and directly adjusts the steam valve, thereby offsetting coupling interference in advance, effectively suppressing temperature disturbances, and improving system robustness. The specific implementation process is as follows:

[0043] A hierarchical predictive collaborative optimization control framework is adopted to achieve multivariable collaborative rolling control of sulfurization kinetics and external temperature disturbances;

[0044] Step 1: In the upper-level offline stage, a temperature trajectory prediction model is trained based on historical operating condition parameters;

[0045] Step 11: Data preparation and construction of operating condition parameter vectors;

[0046] (1) Acquisition and preprocessing of multi-source historical operating condition parameters;

[0047] Specifically, operating parameter data of historical high-quality vulcanization processes, including rubber compound activation energy, are extracted from MES (Manufacturing Execution System), SCADA (Supervisory Control and Data Acquisition System), and sensor databases. Tire thickness Ambient temperature Target sulfidation .

[0048] Furthermore, the preprocessing process includes:

[0049] ① Data cleaning: Remove outliers caused by sensor malfunctions and batch data that fail to meet final product quality standards to ensure data validity.

[0050] ② Alignment and normalization: Unify data from different sources and with different sampling rates onto the same time base, and standardize all numerical features to accelerate network training and improve numerical stability.

[0051] (2) Extract the operating parameters affecting the vulcanization process from the preprocessed operating parameters, and define the input vector of the subsequent temperature trajectory prediction model, including: collecting all operating parameters affecting the vulcanization process to form the input vector. ;

[0052]

[0053] in, The activation energy of the rubber compound, For tire thickness, For ambient temperature, For the target degree of sulfidation, T represents the temperature field. 。

[0054] (3) Define the control output sequence that needs to be optimized, which constitutes the output vector of the subsequent temperature trajectory prediction model, i.e., the baseline control trajectory. :

[0055]

[0056] in, To predict the time domain, Set point for mold temperature. This is the steam pressure setpoint.

[0057] Based on the defined input and output vectors, the preprocessed key parameters are constructed as follows: , Labeled sample pairs, which are the mapping relationships between key input parameters and corresponding optimal control trajectories in the output, are randomly divided into training and test sets in batches.

[0058] Step 12: Define the optimization problem, establish a multi-objective process constraint optimization loss function, and train the temperature trajectory prediction model;

[0059] Specifically, (1) define the optimization problem as minimizing energy consumption and time cost:

[0060]

[0061]

[0062] in, This is the energy cost coefficient. This is the time cost coefficient. The average steam pressure, This is the vulcanization cycle; The baseline control trajectory is the output of the neural network. Input vector for key parameters; It is the economic objective function, representing the total cost within a vulcanization cycle; It is an equality constraint that ensures the final degree of sulfidation reaches the target value; It is an inequality constraint that ensures the mold temperature does not exceed the safe upper limit; It is an inequality constraint that ensures the steam pressure does not fall below the safety lower limit; To predict the degree of sulfide at the time-domain terminal point; The target degree of vulcanization required by the process; For the first The mold temperature setpoint at discrete moments; This is the safe upper limit for mold temperature; For the first Steam pressure setpoints at discrete moments; This is the safe lower limit for steam pressure.

[0063] (2) Based on the physical laws and kinetic characteristics of the sulfidation process, heat transfer equations and sulfidation reaction kinetic equations are constructed, including:

[0064] Models are built based on physical laws to ensure that they accurately reflect the dynamic behavior of the system.

[0065] Heat transfer equation:

[0066]

[0067] in, The density of rubber, This refers to the specific heat capacity of rubber. For temperature field, It is a spatial coordinate vector. For continuous time, For the thermal conductivity of rubber, The enthalpy of the sulfidation reaction. For the degree of sulfidation, For gradient operators, It is a divergence operator.

[0068] The kinetic equation for the sulfidation reaction (Kamal-Sourour model):

[0069]

[0070]

[0071] in, The reaction rate constant is... Pre-exponential factor, For activation energy, Let be the ideal gas constant. denoted as the reaction order.

[0072] (3) Design the neural network architecture for the temperature trajectory prediction model;

[0073] Construct a deep neural network whose output trajectory not only conforms to physical laws but also achieves economic optimization:

[0074]

[0075] in, The temperature trajectory and vapor pressure trajectory are predicted by the temperature trajectory prediction model. These are key operating parameters. These are network parameters.

[0076] (4) Establish a multi-objective process constraint optimization loss function;

[0077] Using the minimization of data fitting loss, dynamic residual loss, constraint violation loss, and KKT conditional residual loss as optimization objectives, a multi-objective process constraint optimization loss function is constructed. Through multi-objective optimization, the trajectory output by the network is ensured to simultaneously satisfy physical laws, data fitting, constraint conditions, and economic optimality.

[0078] 1) Data fitting loss:

[0079]

[0080] in, For the first i training samples, The total number of training samples, Neural network for the first i Temperature-vapor pressure trajectory output for each sample; No. i The expected process trajectory of each sample.

[0081] 2) Dynamic residual loss: Ensures that the output trajectory conforms to physical laws;

[0082]

[0083] in, Indicates the first k The trajectory is output by the neural network at each discrete time step. Predicted degree of sulfidation obtained through numerical integration; For network output The first obtained by numerical integration k Instantaneous rate of vulcanization; The "theoretically correct instantaneous rate" given by the Kamal-Sourour dynamic equations.

[0084] 3) Constraint violation losses: Ensure that the output meets process safety constraints;

[0085]

[0086] in, Returns x if x > 0, otherwise returns 0; For the first kThe mold temperature predicted by the neural network; The maximum mold temperature allowed by the process; For the first k Steam pressure trajectory predicted by a neural network; This is the lowest steam pressure allowed by the process.

[0087] 4) KKT conditional residual loss: guides the output towards economic optimality:

[0088]

[0089] in, , , , These are weighting coefficients, which respectively control the penalty intensity for stationarity, original feasibility, dual feasibility, and complementary relaxation. Let ζ be the gradient of the Lagrange function with respect to the control trajectory ζ of the output; The residuals are constrained by the equation for achieving the required degree of sulfidation. and This represents the negative part of the temperature / pressure inequality constraint. and These are the Lagrange multipliers corresponding to temperature and pressure constraints; and It is a complementary relaxation product.

[0090] Furthermore, by using KKT conditional loss, we ensure that the trajectory output by the network not only satisfies physical laws and constraints, but also achieves economic optimization.

[0091] Wherein, the Lagrange function is:

[0092]

[0093] in, Is it with equality constraints? The corresponding Lagrange multipliers, and Is it an inequality constraint? and The corresponding Lagrange multipliers.

[0094] Furthermore, calculate the multipliers and solve the linear equation:

[0095] in, Indicates the trajectory The gradient operator for a (temperature / pressure sequence) represents the "rate of change".

[0096] Furthermore, the KKT conditions include:

[0097] Stationarity condition: The sum of the gradients of the objective function and the constraints is zero;

[0098]

[0099] Original feasibility conditions: Constraints must be met;

[0100]

[0101] Duality feasibility condition: Lagrange multipliers must be non-negative;

[0102]

[0103] Complementary relaxation condition: The product of the Lagrange multiplier and the constraint is zero;

[0104]

[0105] (5) The neural network architecture for training the temperature trajectory prediction model;

[0106] Based on the above multiple optimization objectives, a neural network is trained that can comprehensively balance data fitting, physical laws, constraint safety, and economic optimality, including:

[0107] Multi-objective composite loss function:

[0108]

[0109] in, , , and These are weighting coefficients used to balance the importance of each loss term.

[0110] Furthermore, the optimization process for network parameters is as follows:

[0111]

[0112] Optimization is performed using gradient descent algorithms (such as the Adam optimizer), with gradients calculated and network parameters updated via backpropagation.

[0113] As one embodiment, the specific training process includes: setting key operating condition parameters Mapping to trajectory in one step Then use trajectory Simultaneously, the physical residuals, constraint residuals, and economic objectives are calculated. Then, the 3×3 linear equation system is solved again to obtain the Lagrange multipliers. Finally, all residuals are weighted, summed, and backpropagated to update the network parameters. Finally, a well-trained temperature trajectory prediction model is obtained.

[0114] Further, network verification and adjustments will be made;

[0115] Verify the performance of the trained neural network and adjust it as needed.

[0116] S1: Validation. Using an independent validation dataset, check whether the trajectory output by the network meets the optimization objective and constraints.

[0117] S2: Adjustment: Adjust the weighting coefficients based on the verification results. , , and The network is then retrained until satisfactory performance is achieved.

[0118] Step 2: In the lower-level online control stage, based on the real-time operating parameters, the temperature setpoint trajectory is generated based on the temperature trajectory prediction model trained in the upper-level offline stage. The thermally coupled model of model predictive control and feedforward compensation is used to track the temperature setpoint trajectory. The sparse Bayesian learner is driven to update the thermally coupled model online through the event triggering mechanism to ensure that the optimal temperature trajectory is accurately tracked while solving the steam valve opening control quantity.

[0119] Step 21: Real-time predictive sensing, outputting a baseline control trajectory, including:

[0120] For each batch of new tires vulcanization process, real-time operating parameters are obtained. The data is then input into a pre-trained temperature trajectory prediction model to quickly infer and generate a baseline control trajectory. :

[0121]

[0122] in, The baseline control trajectory is generated by the trained temperature trajectory prediction model. N It predicts the length of the time domain; The optimal temperature trajectory prediction model has been trained. These are real-time operating parameters.

[0123] Step 22: Hot start mapping, split the reference control trajectory into two setpoint trajectory sequences to obtain the temperature setpoint trajectory and the steam pressure setpoint trajectory respectively;

[0124] In the lower-level online control stage, only the temperature setpoint trajectory is tracked; that is, the lower-level tracking layer only tracks temperature. Split into two setpoint sequences:

[0125]

[0126] in,N To predict the length of the time domain; The temperature setpoint trajectory sequence is sent to the lower-level temperature tracking MPC as a setpoint. The steam pressure setpoint trajectory sequence is directly used as the steam pressure feedforward value and no longer appears in the optimization variables of the lower-level MPC; the MPC only calculates the valve opening or power increment to ensure that T accurately follows. That's all.

[0127] Step 23: Use a thermally coupled model of model predictive control and feedforward compensation to track the temperature setpoint trajectory. The lower-level tracking control receives the temperature setpoint trajectory output by the optimal temperature trajectory prediction model. Real-time suppression of thermal coupling interference ensures accurate tracking of the optimal temperature trajectory. The measured temperature of the mold in each temperature zone is directly used as the state feedback for model predictive control, enabling online identification. The specific implementation process includes:

[0128] (1) System state assumptions

[0129] Assume that the measured temperature values ​​of the mold in each temperature zone are sufficient to reflect the complete thermal state of that zone. In other words, the measured temperature of the mold in each temperature zone... Directly used as status feedback for MPC ;

[0130] (2) Unify input / output interfaces and define parameter quantities;

[0131] Unify all input / output interfaces to ensure seamless integration with subsequent modules.

[0132]

[0133] in, This is the controller sampling period.

[0134] Optimize the opening degree of variable steam valves:

[0135]

[0136] in, For the first k The control vector for each step, with elements representing either the steam valve opening or the heating power. This indicates the total number of temperature zones in the vulcanizing machine system.

[0137] Set trajectory:

[0138]

[0139] in, The temperature setpoint trajectory is issued.

[0140] The actual temperature measured by the sensor:

[0141]

[0142] in, For field sensors at the first k The actual temperature vector obtained by step measurement.

[0143] (3) Construct a thermally coupled model;

[0144] S1: State vector

[0145]

[0146] in, It is a state vector; This is the temperature subvector.

[0147] S2: Discrete state equations:

[0148]

[0149] in, For the online-updated coupling matrix, the off-diagonal elements are (i≠j) directly characterizes the temperature range. j Temperature range i The heat transfer effect; The input matrix; The output matrix only outputs the temperature portion.

[0150]

[0151] in, The first in the coupling matrix i Line 1 j Column elements; The heat transfer coefficient is identified in real time; p The density of rubber; Cp Specific heat capacity; V For the temperature range volume; It's a long walk away.

[0152] (4) Construct an online identification coupling matrix;

[0153] S1: Function approximation theory - Laguerre orthogonal polynomial basis;

[0154] Since any function can be approximated by a complete set of orthogonal basis functions, Laguerre polynomials are relevant. yes Interval with respect to the weight function Orthogonal polynomial sequences. This differs from traditional methods that directly identify parameters. Unlike other disclosures, this disclosure will use coupling coefficients. It can be viewed as a time function and represented by a linear combination of a set of Laguerre functions:

[0155]

[0156] in, For the first is the Discrete moments; The one currently in use is the first one. The first-order basis function or the second-order basis function Each weight; For the first Laguerre polynomials of order one are a set of polynomials defined on On the interval, with respect to the weight function Orthogonal basis functions; The system sampling period; These are weighting coefficients. It is the order of the basis functions.

[0157] S2: Coupling relationship modeling;

[0158] Temperature zone Temperature range The thermal coupling effect is modeled as its effect of temperature difference on temperature. Convolutional model of the effect of temperature in the region:

[0159]

[0160] in, Indicates temperature zone At the present moment The mold temperature; Indicates temperature zone At the previous sampling time The mold temperature; This indicates that at the previous moment, the adjacent temperature zones The mold temperature; This indicates the depiction of the "temperature zone" Temperature range "When considering the dynamic characteristics of this specific coupling channel, the first..." The weighting coefficients of the Laguerre basis function; For the first The value of the first-order Laguerre basis function at time zero; This indicates the steam valve opening degree at the previous moment relative to the current temperature zone. The direct heating effect produced by temperature; B It is the input matrix, and u is the control input.

[0161] Furthermore, the time-varying coupling coefficients Parameterized as:

[0162]

[0163] The objective of this online identification is to estimate time-varying parameters. Transformed into estimating fixed parameter weighting coefficients .

[0164] S3: Sparse Bayesian learning;

[0165] 1) Constructing the regression vector:

[0166] For any pair of temperature regions with thermal coupling relationship Construct its discrete time step regression vector As input for subsequent sparse Bayesian learning algorithms:

[0167]

[0168] in, This indicates the adjacent temperature zones at the previous time. Compared with the current target temperature zone The temperature difference of the mold; This indicates the adjacent temperature zone that acted as a heat source or thermal disturbance at the previous moment. The measured mold temperature; Indicates the target temperature zone at the previous moment. The actual measured temperature of the mold itself; This represents the highest order of the selected Laguerre basis function. Temperature difference. Defined as: Laguerre polynomials of all orders in The value at that location is always 1: .

[0169] Regression vector It can be simplified to a system based on temperature difference. Constant vectors in scalar multiplication:

[0170]

[0171] in, It is A column vector of all 1s.

[0172] 2) The sparse Bayesian learner is updated online via an event-triggered mechanism, where the event trigger is as follows:

[0173] To reduce unnecessary computation and improve system efficiency and robustness, the updates of the coupled model are controlled by an event-triggered mechanism, rather than the traditional timed periodic updates.

[0174] a. First, define the prediction error:

[0175] At each discrete time step Calculate the temperature range Temperature prediction error :

[0176]

[0177] in, It is a temperature zone At any moment The actual measured temperature value; The coupled model at the previous time step For the current moment The predicted temperature is calculated as follows:

[0178] The system measurement output from the previous moment With the control input already executed Substitute these values ​​into the discrete state-space model of the system, and use the coupling matrix obtained from the previous cycle. Perform forward prediction:

[0179]

[0180] in, This is the output matrix; Control input matrix.

[0181] b. Define trigger conditions:

[0182] Define a trigger function based on prediction error. and a dynamic threshold :

[0183]

[0184] The coupled model is only updated when the value of the trigger function is greater than zero.

[0185]

[0186]

[0187] c. Design dynamic thresholds:

[0188] This disclosure employs a hybrid threshold that combines a fixed tolerance band and an error integral term, and its discrete recursive implementation is as follows:

[0189]

[0190] in, It is a fixed tolerance band, used to ignore minor noise and disturbances; It is the integral gain coefficient ( ), used to adjust the sensitivity of the threshold to accumulated error; This is the system sampling time.

[0191] 3) Sparse Bayesian Update:

[0192] In the sparse Bayesian learning framework, a prior probability distribution is defined for the weight coefficients to be estimated:

[0193]

[0194] in, This indicates the temperature range before any data is observed. Temperature range The weight vector of the coupling channel The initial assumptions; It follows a normal distribution; Let be the covariance matrix of the prior distribution. The larger the covariance matrix, the greater the probability that the weights are allowed to deviate from 0.

[0195] The covariance matrix (inverse of the precision matrix) of the prior distribution is determined by the hyperparameters:

[0196]

[0197] in, For hyperparameters, each Each weighting coefficient is controlled individually. The prior distribution of .

[0198] Covariance matrix update:

[0199]

[0200] in, The updated posterior covariance matrix represents the weight vector after incorporating the new observation data. The degree of uncertainty in the estimated value; The precision of the observed noise (the reciprocal of the variance); This is the regression vector.

[0201] Mean vector (weight) update:

[0202]

[0203] in, The updated posterior mean vector represents the fusion of the current time step. After the new data, the coupling channel Weight coefficient vector Optimal point estimation; For the current moment Temperature zone Temperature sensor measured value ; To use the old weight estimation For the current system output (temperature range) Predicted values ​​for temperature.

[0204] Sparsity hyperparameter update:

[0205]

[0206] in, It is the first Weight coefficients The corresponding new precision value; For the first The posterior mean of each weight, derived from the latest updated mean vector. ; The posterior covariance matrix is ​​the first... One diagonal element.

[0207] Noise accuracy update:

[0208]

[0209] in, This is the normalization constant.

[0210] S4: Output coupling matrix;

[0211] Through the Sparse Bayesian Learning (SBL) algorithm, a set of fixed and optimal weight coefficients was learned:

[0212]

[0213] In subsequent control, whenever the lower-level MPC requires a coupling matrix... At that time, using the fixed weights learned above Using the known Laguerre function, the time-varying coupling coefficient at the current moment is calculated in real time:

[0214]

[0215] Time-varying coupling coefficients of all channels Fill in the coupling matrix The corresponding location is provided for use by the MPC predictive control model. (5) State feedback;

[0216] The lower-level MPC controller directly receives measurement values ​​from temperature sensors in each temperature zone. and take it as the current state of the system. Direct observations are used for rolling optimization of the prediction model, i.e.:

[0217] = ,

[0218] This design omits complex state observers, reduces online computational complexity, and makes the system easier to implement and maintain.

[0219] (6) Feedforward compensation

[0220] Although MPC introduces feedforward compensation, it does not directly handle the compensation of external disturbances. Instead, it generates a better control strategy as part of the model, which introduces a certain computational delay. In order to immediately counteract the effects of disturbances, the required control quantity is calculated using an "analytical formula" before the feedback controller (MPC) acts, thus canceling out known coupled disturbances in advance, thereby reducing the burden on the feedback loop and improving tracking accuracy.

[0221] Feedforward compensation (linear coupling) based on analytical model:

[0222]

[0223] in, To analyze the feedforward control quantity; This is the feedforward gain matrix; This is the set of temperature zone indices adjacent to temperature zone i; The coupling heat transfer coefficient is identified in real time; Set the temperature for adjacent temperature zones; Set the temperature for this temperature range. Calculate the sign and magnitude of future interference and prepare for reverse compensation in advance.

[0224] Employing a lightweight feedforward network responsible for "fine-tuning correction" to specifically counteract nonlinear coupling errors that occur online, this disclosure designs a simple single-hidden-layer feedforward neural network (FNN) specifically for predicting and counteracting these nonlinear disturbances.

[0225] Feedforward compensation based on neural networks (nonlinear coupling):

[0226]

[0227] in, This is the feedforward control input for the neural network; , For output layer weights and biases; , For hidden layer weights and biases; For activation functions; The input vector for the FNN network is defined as:

[0228]

[0229] in, The network input vector; The temperature of the adjacent temperature zone at the previous moment; This is the temperature of this temperature range at the previous moment. Capture the "actual ≠ plan" deviation caused by nonlinear dynamics (such as valve hysteresis, heat transfer coefficient changes with temperature); This is the set of temperature zone indices adjacent to temperature zone i.

[0230] Total feedforward control quantity:

[0231]

[0232] in, This is the final feedforward control value; The fusion weights of the parser and the neural network are defined, with values ​​ranging from 0 to 1.

[0233] (7) MPC prediction model

[0234] In the prediction model of MPC, a feedforward matrix is ​​introduced to handle the feedforward path inside the system and to take into account known external inputs, thereby improving the accuracy of prediction.

[0235] Generate future Np-step output prediction:

[0236]

[0237]

[0238]

[0239]

[0240]

[0241]

[0242] in, The temperature vectors of each temperature zone directly measured by the sensor at the current moment are directly used as the system state; The output sequence represents the predicted temperatures for all temperature zones in the future period. For valve opening sequence; This is the initial state transition matrix; This is the dynamic matrix from valve opening to output. This is the influence matrix of the control quantity.

[0243] (8) Solving the quadratic programming problem (QP);

[0244] Find the optimal control increment that satisfies the constraints:

[0245]

[0246] in, Given the current state x(k) and the valve opening sequence u(k),…,u(k+Nc-1), it is the predicted output at the (k+j)th step. It is the trajectory set at step k+j; For prediction in the time domain; For control in the time domain; Np is for output prediction; i l represents the vulcanizing machine number; Q and R are weight matrices used to balance tracking error and control actions; S This is a coupling suppression weight matrix, used to reduce the effects of thermal coupling:

[0247]

[0248] in, The objective function value; To predict the output sequence (obtained from the previous step); To set the trajectory sequence; The tracking error weight block diagonal matrix; This is the diagonal matrix of the valve opening weight blocks; The diagonal matrix of the coupling suppression weight block;

[0249] Will Substituting and combining, we get:

[0250]

[0251]

[0252]

[0253] in, It is a Hessian matrix; This is the vector of linear term coefficients; This is a constant term.

[0254] Control limits:

[0255]

[0256] in, , This refers to the upper and lower limit vectors of the control quantity.

[0257] Output soft constraints:

[0258]

[0259] in, , To output the upper and lower bound vectors; This is a vector of slack variables.

[0260] QP standard format:

[0261]

[0262] in, The penalty coefficient is the slack variable. It is a vector consisting entirely of 1s.

[0263]

[0264]

[0265] in, , , Let be a linear inequality matrix and vector formed by stacking upper and lower bound constraints.

[0266] Solving QP yields the optimal feedback control sequence. Take the first item .

[0267] Feedforward and feedback superposition:

[0268]

[0269] in, This is for the final issuance of control quantities; This is the feedback control quantity (calculated by MPC).

[0270] Step 3: Execute the loop in real time to realize the multi-variable collaborative rolling control process;

[0271] The entire closed loop is completed every second, and the specific implementation process is as follows:

[0272] S1: Read sensor data With the set trajectory ;

[0273] S2: If the event is triggered, update the coupling matrix. ;

[0274] S3: Will Set as current state ;

[0275] S4: Based on Calculate feedforward amount with set value ;

[0276] S5: Solve QP to obtain the optimal control quantity Take the first element as Synthetic control quantity obtained And issued;

[0277] Finally, store the data and prepare for the next cycle.

[0278] Example 2

[0279] One embodiment of this disclosure provides a predictive optimization control system for tire vulcanization process based on process constraints, including:

[0280] The collaborative optimization module is used to achieve multivariable collaborative rolling control of sulfurization kinetics and external temperature disturbances using a hierarchical predictive collaborative optimization control framework.

[0281] The hierarchical predictive collaborative optimization control framework includes an upper-level offline stage and a lower-level online control stage. In the upper-level offline stage, a multi-objective process constraint optimization loss function is established based on historical operating parameters, and a temperature trajectory prediction model is trained. In the lower-level online control stage, a temperature setpoint trajectory is generated based on the temperature trajectory prediction model trained in the upper-level offline stage according to real-time operating parameters. A thermally coupled model of model predictive control and feedforward compensation is used to track the temperature setpoint trajectory. An event-triggered mechanism drives a sparse Bayesian learner to update the thermally coupled model online, ensuring that the optimal temperature trajectory is accurately tracked while solving the steam valve opening control quantity.

[0282] The control module is used to issue control commands based on the steam valve opening control quantity to realize a multi-variable collaborative rolling control process.

[0283] Example 3

[0284] One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned process-constrained tire vulcanization process prediction and optimization control method.

[0285] Example 4

[0286] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions. When these computer instructions are executed by a processor, they implement the aforementioned method for predictive optimization control of tire vulcanization process based on process constraints.

[0287] Example 5

[0288] One embodiment of this disclosure provides an electronic device, including a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the aforementioned process-constrained tire vulcanization process prediction and optimization control method.

[0289] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0290] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0291] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A process constraint based prediction optimal control method for tire curing process, characterized in that, The application relates to a method for realizing multivariable collaborative rolling control of vulcanization kinetics characteristics and external temperature disturbance. The method adopts a hierarchical predictive collaborative optimization control framework to realize the multivariable collaborative rolling control of the vulcanization kinetics characteristics and the external temperature disturbance. The hierarchical predictive collaborative optimization control framework comprises an upper-layer offline stage and a lower-layer online control stage. In the upper-layer offline stage, a multi-objective process constraint optimization loss function is established according to historical working condition parameters, and a temperature trajectory prediction model is trained. In the lower-layer online control stage, a temperature set point trajectory is generated based on the temperature trajectory prediction model trained in the upper-layer offline stage according to real-time working condition parameters. The temperature set point trajectory is tracked by using a model predictive control and a feedforward compensation thermal coupling model, and a sparse Bayesian learner is driven by an event triggering mechanism to update the thermal coupling model online, so that the optimal temperature trajectory is accurately tracked and the steam valve opening control amount is solved. In the upper-layer offline stage, historical working condition parameters are first acquired, and a training set and a test set are constructed.

2. The process constraint based tire curing process predictive optimization control method of claim 1 wherein, The working condition parameters of a high-quality vulcanization process are extracted from a manufacturing execution system, a data acquisition and monitoring control system and a sensor database, and are pretreated. Key parameters affecting the vulcanization process are extracted from the pretreated working condition parameters, input vectors and output vectors are defined, and a label sample pair, that is, a mapping relationship between the input key parameters and the corresponding optimal temperature set point trajectory, is constructed. The training set and the test set are randomly divided according to batches. A temperature trajectory prediction model is constructed by defining an optimization problem, constructing a heat transfer equation and a vulcanization reaction kinetics equation based on physical laws and kinetics characteristics of the vulcanization process, and taking the minimum of a data fitting loss, a kinetics residual loss, a constraint violation loss and a KKT condition residual loss as an optimization target. A multi-objective process constraint optimization loss function is constructed, and multi-objective optimization iteration is carried out by using a gradient descent algorithm. Optimal network parameters are generated to train a temperature trajectory prediction model which comprehensively balances data fitting, physical laws, constraint safety and economic optimization. Control instructions are issued according to the steam valve opening control amount to realize the multivariable collaborative rolling control process. In the lower-layer online control stage, the temperature set point trajectory is generated based on the temperature trajectory prediction model trained in the upper-layer offline stage according to real-time working condition parameters. For each batch of new tire vulcanization process, real-time working condition parameters are acquired, a reference control trajectory is generated by using the temperature trajectory prediction model trained in the upper-layer offline stage, and the reference control trajectory is split into two set point trajectory sequences to obtain a temperature set point trajectory and a steam pressure set point trajectory.

3. The process constraint based tire curing process predictive optimization control method of claim 1 wherein, The model predictive control and feedforward compensation thermal coupling model tracks the temperature set point trajectory, including: constructing a model predictive control and feedforward compensation thermal coupling model, modeling the thermal coupling effect between temperature zones as a convolution model of the temperature difference between the temperature zones on the temperature of the temperature zone, only tracking the temperature set point trajectory in the lower layer online control stage, the lower layer online control stage receives the temperature set point trajectory output by the temperature trajectory prediction model for tracking control, and the measured temperature of each temperature zone mold is directly used as the state feedback of the model predictive control to realize online identification.

4. The process constraint based tire curing process predictive optimization control method of claim 1 wherein, The event-triggered mechanism drives the sparse Bayesian learner to update the thermal coupling model online, including: the update of the thermal coupling model is controlled by an event-triggered mechanism, the prediction error is defined, and the trigger function and a dynamic threshold based on the prediction error are defined, when the value of the trigger function is greater than zero, the update of the thermal coupling model is performed, a set of fixed and optimal weight coefficients are learned by the sparse Bayesian learning algorithm, and the time-varying coupling coefficient at the current time is calculated in real time by using the learned weight sparsity and the known Laguerre basis function whenever the lower layer model predictive control needs the coupling matrix, and the time-varying coupling coefficient is fed back to the model predictive control to obtain the optimal control amount and is issued.

5. A process constraint based prediction optimal control system for a tire curing process characterized in that, It comprises: The collaborative optimization module is used to realize the multivariable collaborative rolling control of vulcanization dynamics and external temperature disturbance by using a hierarchical prediction collaborative optimization control framework; The hierarchical prediction collaborative optimization control framework comprises an upper layer offline stage and a lower layer online control stage, in the upper layer offline stage, a multi-objective process constraint optimization loss function is established according to historical working condition parameters, and a temperature trajectory prediction model is trained; In the lower layer online control stage, a temperature set point trajectory is generated based on the temperature trajectory prediction model trained in the upper layer offline stage according to real-time working condition parameters, a model predictive control and feedforward compensation thermal coupling model is used to track the temperature set point trajectory, and a sparse Bayesian learner is driven by an event-triggered mechanism to update the thermal coupling model online, so that the optimal temperature trajectory is accurately tracked and the steam valve opening control amount is solved; In the upper layer offline stage, first, historical working condition parameters are acquired, and a training set and a test set are constructed, including: extracting historical working condition parameters of high-quality vulcanization processes from a manufacturing execution system, a data acquisition and monitoring control system and a sensor database and preprocessing the working condition parameters, extracting key parameters affecting the vulcanization process from the preprocessed working condition parameters, defining input vectors and output vectors, and constructing label sample pairs, that is, the mapping relationship between the input key parameters and the corresponding optimal temperature set point trajectory output, and randomly dividing the label sample pairs into a training set and a test set according to batches. The building of the multi-target process constraint optimization loss function, the training of the temperature trajectory prediction model comprises: defining an optimization problem, and building a heat transfer equation and a vulcanization reaction kinetics equation based on the physical law and the kinetic characteristics of the vulcanization process, building a temperature trajectory prediction model, taking the minimization of a data fitting loss, a kinetic residual loss, a constraint violation loss and a KKT condition residual loss as an optimization objective, building a multi-target process constraint optimization loss function, performing multi-target optimization iteration through a gradient descent algorithm, and generating optimal network parameters, so as to train a temperature trajectory prediction model which comprehensively balances data fitting, physical law, constraint safety and economic optimization; The control module is configured to issue a control instruction according to the steam valve opening degree control quantity, and realize multi-variable collaborative rolling control.

6. A computer program product comprising a computer program, characterized in that, The computer program, when executed by a processor, implements the process constraint based tire vulcanization process prediction and optimization control method of any one of claims 1-4.

7. A non-transitory computer-readable storage medium, comprising: The non-transitory computer readable storage medium is configured to store computer instructions, and the computer instructions, when executed by a processor, implement the process constraint based tire vulcanization process prediction and optimization control method of any one of claims 1-4.

8. An electronic device, comprising: Comprise: A processor, a memory and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the process constraint based tire vulcanization process prediction and optimization control method of any one of claims 1-4.

Images