Intelligent control system and control method for full-process production of maackia floribunda shaving board

Abstract

The application relates to the field of whole-process intelligent control technology, and discloses a whole-process intelligent control system and method for producing lamina board of maackia fauriei, which comprises the following modules: a material attribute sensing module, which is used for collecting initial material attribute vectors; a material attribute transfer token management module, which is used for updating, transferring and archiving material attribute transfer tokens; a global instruction planning and correction module, which is used for planning or correcting recommended control instruction sets of each process unit; a distributed instruction execution and negotiation module, which is used for locally predicting the feasibility of the recommended control instructions; and a process model self-optimization module, which is used for periodically optimizing a global process model. The material is subjected to multidimensional real-time detection by the material attribute sensing module, accurate initial and process attribute data are obtained, and the whole-process seamless transfer of material attributes, instructions and constraints is realized with the aid of the material attribute transfer token management module, so that the product defect rate is reduced.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for the entire production process, specifically to an intelligent control system and control method for the entire production process of goldenrain tree particleboard. Background Technology

[0002] The industrial manufacturing of goldenrain tree particleboard is a complex, multi-stage process. It encompasses a series of precise operations, from wood chipping, drying, and glue mixing to laying, molding, and hot-pressing curing. The parameter settings at each stage directly affect the physicochemical transformation of the materials, thereby determining the performance indicators of the final product, such as strength, density, moisture resistance, and environmental characteristics.

[0003] Currently, the control practices of goldenrain tree particleboard production lines mainly rely on empirical formulas and partial automation methods. Each process unit is often independently equipped with basic sensors and controllers. For example, the drying section may use temperature and humidity sensors and PID loops to maintain a preset drying curve; the adhesive mixing section adds adhesive according to the formula ratio; and the hot press operates according to a fixed pressure-temperature-time program. Although these local controls can ensure the basic operation of their respective links, the status information of materials between different sections is usually not effectively integrated.

[0004] However, existing control technologies lack a full-process, real-time updated material attribute data carrier, and information cannot be seamlessly transmitted between different production stages. This makes it difficult for downstream stages to make proactive and adaptive control adjustments based on the actual output material status of upstream stages. Commands are mostly issued in a rigid, one-way manner, failing to fully consider the real-time operating status and physical capacity of local equipment. Once recommended commands exceed the limits of local equipment, the system struggles to provide timely feedback, leading to equipment wear, malfunctions, or production interruptions. Furthermore, it is difficult to autonomously learn and update based on dynamic changes in the production environment (such as long-term equipment wear, raw material characteristic drift, or the introduction of new formulas). Therefore, this invention provides an intelligent control system and method for the entire production process of goldenrain tree particleboard to address the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent control system and method for the entire production process of goldenrain tree particleboard, which solves the problems of large fluctuations in material properties, difficulty in real-time and accurate adjustment of process parameters, production efficiency limited by human experience, and poor product quality in the existing particleboard production process.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of this invention provides an intelligent control system for the entire production process of goldenrain tree particleboard, comprising: The material attribute sensing module is used to collect the initial material attribute vectors of batches of materials entering the production line at different process unit nodes; The material property transfer token management module is used to receive the collected initial material property vector, generate and initialize a unique material property transfer token based on the initial material property vector, the material property transfer token carries the full process information of this material batch, and updates, transfers and archives the material property transfer token according to the system running status. The global instruction planning and correction module is used to receive the material attribute transfer token containing the material attribute vector, and use the global process model that digitally represents each process unit in the production process of goldenrain wood particleboard, including drying, glue mixing and hot pressing stages, to plan or correct the recommended control instruction set of each process unit, and write the recommended control instruction set into the instruction set field of the material attribute transfer token. The distributed instruction execution and negotiation module is used to receive the material property transfer token containing the recommended control instruction and to perform a local feasibility prediction on the recommended control instruction. The process model self-optimization module is used to call the historical material property transfer token data archived by the material property transfer token management module to periodically optimize the global process model in the global instruction planning and correction module.

[0007] Preferably, the material property sensing module includes one or more online sensing units deployed at key process unit nodes of the production line. The online sensing unit integrates various sensor hardware to collect multi-dimensional material properties. The sensor hardware includes: Near-infrared spectrometer used for collecting near-infrared spectral data of materials; Machine vision systems used to acquire morphological features of materials; Microwave sensors used to measure the bulk properties of materials.

[0008] Preferably, the material property sensing module further includes an internally integrated data processing function, which includes a data preprocessing unit and a feature calculation unit; The data preprocessing unit is used to reduce noise and correct the raw signals output by various sensor hardware. The feature calculation unit is used to extract quantified material properties from the noise-reduced and corrected data. The extraction process includes calling a pre-established chemometric model to calculate the near-infrared spectral data collected by the near-infrared spectrometer into material chemical properties.

[0009] Preferably, the material property transfer token management module includes: The material property transfer token data structure definition unit is used to define the data fields contained in the material property transfer token and to provide a structural template for the creation and management of material property transfer token instances. The instance creation and initialization unit is used to create and initialize a unique material property transfer token instance based on the data fields determined by the material property transfer token data structure definition unit and the received initial material property vector, so as to carry the full process information of this material batch. The status update and routing unit is used to receive the unique material property transfer token instance and dynamically update the material property vector and control instruction set in the material property transfer token instance based on the feedback from each process unit in the production process. The archiving management unit is used to receive the material property transfer token instance that has been transmitted by the status update and routing unit after the completion of the entire production process, and to store the material property transfer token instance in the historical database.

[0010] Preferably, the global instruction planning and correction module includes: The global process model unit has a built-in global process model that is a digital simulator of the entire production process of goldenrain tree particleboard. Based on material properties and control instructions, it predicts the processing results of each process unit. The initial instruction planning unit is used to receive the initial material attribute vector contained in the material attribute transfer token, and based on the iterative optimization method of the global process model, with the goal of optimizing the final product attributes, reversely derive and construct the initial recommended control instruction set. The dynamic feedforward correction unit is used to receive the material property transfer token containing the material property vector and call the global process model to dynamically adjust the initial recommended control instruction set to generate the corrected recommended control instruction set. The constraint-based replanning unit receives a material property transfer token containing an execution constraint vector and calls the global process model to replan the modified recommended control instruction set. Under the premise of satisfying the execution constraints, it recalculates the suboptimal recommended control instructions.

[0011] Preferably, the initial instruction planning unit receives the initial material attribute vector from the material attribute transfer token and uses the initial material attribute vector as input to the global process model. It then calculates the initial recommended control instruction set through forward calculation, wherein the forward calculation is expressed as: ; in, A set containing recommended control instructions for each process unit. ... , Represents the global process model. This is the initial material attribute vector. This is the current parameter set of the global process model. This represents the total number of process units in the production process.

[0012] Preferably, the constraint-based replanning unit receives the execution constraint vector from the material property transfer token. and the execution constraint vector As a solution process unit The hard boundary conditions for control commands are invoked by calling the global process model, while satisfying the execution constraint vector. Under the premise of [the above conditions], the suboptimal recommended control instructions are recalculated for the process unit.

[0013] Preferably, the distributed instruction execution and negotiation module is located in each process unit of the production process, and the distributed instruction execution and negotiation module includes: The state self-sensing unit is used to continuously monitor the real-time operating status of the controlled process unit and form a multi-dimensional equipment state vector. ; The feasibility prediction unit has a built-in physical capability constraint model for determining whether recommended control commands can be safely executed. It uses a multi-dimensional equipment state vector and the recommended control command vector of the process unit as inputs for verification calculations. The verification process can be represented as follows: ; in, The output is a boolean value representing the verification result. This represents a model representing the physical capacity constraints of the equipment. Indicates the process unit Recommended control command vector, Indicates process unit Real-time device state vector; The constraint feedback generation unit is used in the... When the result is false, an execution constraint vector containing specific physical boundaries is generated, and the execution constraint vector is written into the constraint set field of the material property transfer token; Instruction execution and monitoring unit, used in the above When true, the recommended control command is translated into an operation on the physical device and monitored.

[0014] Preferably, the process model self-optimization module includes: A historical data batch processing unit is used to construct a dataset from a historical database for training the global process model; The model performance evaluation unit is used to define a loss function, which quantifies the deviation between the global process model's predicted values ​​of the processing results for each batch of materials in the dataset and the actual production results. The loss function can be defined as follows: ; in, Based on the model parameter set The loss function value for the variable. Represents the training dataset The total number of samples in the sample, This represents the total number of process units in the production process. This is the first Weighting coefficients for the prediction accuracy allocation of each process unit. It is the first The sample left the first The measured material property vector after each process unit The representation model is based on the first The sample enters the... Measured properties before each process unit and the control commands ultimately executed by this process unit Predicted values ​​for the output attributes of this process unit. The square of the L2 norm of a vector; The parameter optimization and solution unit is used to solve for the model parameter set that minimizes the loss function by employing a numerical optimization algorithm with the loss function as the optimization objective. The model update management unit is used to deploy the solved model parameter set to the global instruction planning and correction module to update the global process model.

[0015] A second aspect of this invention also provides an intelligent control method for the entire production process of Koelreuteria paniculata particleboard, comprising the following steps: S1 collects the initial attributes of the batches of materials entering the production line and generates an initial recommended control instruction set covering the entire process based on these initial attributes. S2, the distributed instruction execution and negotiation module receives the initial recommended control instruction set, performs local feasibility prediction and closed-loop negotiation, and executes the finally confirmed instructions; S3, to measure the properties of the material after processing, and to make feedforward corrections to the recommended control commands of all process units based on the measurement results; S4 transmits the updated material property transfer tokens level by level on the production line, and repeats the feasibility prediction, closed-loop negotiation, and feedforward correction process for subsequent process units. S5. Archive the material property transfer tokens that have completed the entire process, use a numerical optimization algorithm to solve for the optimized model parameter set, and update the global process model using the optimized model parameter set.

[0016] This invention provides an intelligent control system and method for the entire production process of Koelreuteria paniculata particleboard. It has the following beneficial effects: 1. This invention utilizes a material attribute sensing module to perform multi-dimensional real-time detection of materials, acquiring accurate initial and process attribute data. Furthermore, it leverages a material attribute transfer token management module to achieve seamless transmission of material attributes, instructions, and constraints throughout the entire process. The global instruction planning and correction module, based on a global process model, performs forward-looking planning and dynamic feedforward correction with the goal of optimizing the final product, effectively addressing material fluctuations. This closed-loop intelligent control ensures that different batches of materials are processed according to the optimal strategy, thereby reducing the product defect rate.

[0017] 2. This invention utilizes a global process model to optimize and plan the entire process of instructions, avoiding the overall suboptimal problem that may result from local optimization. The distributed instruction execution and negotiation module performs local feasibility prediction and negotiation for instructions in each process unit, ensuring that instructions can be executed safely and efficiently, reducing downtime or adjustment time caused by unsuitable instructions, and reducing unnecessary energy consumption and raw material waste through precise control of the material processing process, effectively improving the yield of particleboard and the comprehensive utilization efficiency of various production resources.

[0018] 3. This invention introduces a process model self-optimization module, which can periodically utilize massive amounts of historical production data archived using material attribute transfer tokens. Through loss function optimization and iterative algorithms, the global process model is used to learn and update parameters. As production data accumulates, the model continuously evolves, autonomously identifying and adapting to factors such as changes in the production environment, equipment aging, or the introduction of new materials. This continuously improves the model's prediction accuracy and the effectiveness of control strategies, thereby ensuring that the intelligent control system maintains optimal operating conditions over the long term. Attached Figure Description

[0019] Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is a schematic diagram of the instruction execution bidirectional negotiation mechanism of the present invention; Figure 3 This is a flowchart of the method of the present invention. Detailed Implementation

[0020] The technical solutions in 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.

[0021] See attached document Figure 1 and attached Figure 2 This invention provides an intelligent control system for the entire production process of goldenrain tree particleboard. The system may include: a material property perception module, a material property transfer token management module, a global instruction planning and correction module, a distributed instruction execution and negotiation module, and a process model self-optimization module.

[0022] The material attribute awareness module is deployed at key process nodes on the production line to collect real-time attribute data of materials flowing through it. The material attribute transfer token management module is responsible for creating, updating, transferring, and ultimately archiving a unique material attribute transfer token for each batch of materials. The global instruction planning and correction module runs in the process coprocessor and dynamically generates or corrects recommended control instructions for all downstream process units based on the latest material attributes carried in the material attribute transfer tokens. The distributed instruction execution and negotiation module is deployed in the distributed control nodes of each process unit to receive and predict the feasibility of recommended control instructions and execute the finally confirmed control instructions. The process model self-optimization module is responsible for offline iterative optimization of the global process model in the global instruction planning and correction module using material attribute transfer tokens archived in the historical database.

[0023] In one embodiment of the present invention, the overall flow of the control method executed by the intelligent control system is as follows: First, the material attribute sensing module detects a new batch of materials at the entrance of the production line and collects its initial attributes. The material attribute transfer token management module generates a unique material attribute transfer token for the batch of materials and writes the initial attributes.

[0024] The global instruction planning and correction module receives the initialized material property transfer token and plans a set of initial recommended control instructions covering the entire process. These instructions, along with the material property transfer token, are pre-sent to the distributed instruction execution and negotiation module of the first process unit.

[0025] At this process unit, the distributed instruction execution and negotiation module performs feasibility prediction and two-way negotiation on the instructions to determine the final execution instructions and complete the processing. After processing, the material attribute sensing module collects the post-processing attributes of the material again, and the material attribute transfer token management module updates this new attribute into the material attribute transfer token.

[0026] Based on this latest measured attribute, the global instruction planning and correction module feeds forward to correct the recommended control instructions for all subsequent process units. The above-mentioned steps of execution, sensing, updating, and correction are carried out step by step and in a cascading manner on the production line.

[0027] When a batch of materials completes all processes, its material property transfer tokens, which are full of historical data, are archived by the material property transfer token management module. The process model self-optimization module will periodically call these archived data to train and optimize the global process model.

[0028] In one specific implementation, the material property sensing module consists of one or more online sensing units deployed at key process nodes of the production line (e.g., the inlet and outlet of the drying unit, the outlet of the mixing unit, etc.). The function of this module is to perform real-time, non-contact property data acquisition and calculation of the Koelreuteria paniculata wood shavings flowing in the physical conveying channel.

[0029] Each online sensing unit can integrate multiple sensor hardware to achieve comprehensive characterization of the multi-dimensional properties of materials. In one embodiment, the unit may include a near-infrared spectrometer for analyzing the chemical composition of the material. The near-infrared spectrometer illuminates the surface of the material by emitting near-infrared light in a specific wavelength range and receives the spectral signals reflected or transmitted from it, which are used to calculate the chemical properties of the material, such as moisture content, sizing amount, or wood extract content.

[0030] In another embodiment, the unit may include a machine vision system for acquiring morphological features of the material. The machine vision system consists of an industrial camera, a light source with a specific angle and wavelength, and an image acquisition card. By capturing continuous images of the material, it analyzes the physical morphological properties of the wood shavings, such as their dimensional distribution, aspect ratio, color, and surface impurities.

[0031] In another embodiment, the unit may further include a microwave sensor for measuring the bulk properties of the material. This microwave sensor calculates the macroscopic average density and overall internal moisture content of the material batch by measuring the attenuation and phase shift of the microwave signal to the material. The material property sensing module may include a combination of any one or more of the above-mentioned sensors, depending on specific measurement requirements.

[0032] The material property sensing module integrates data processing capabilities, which transform the raw signals collected by the sensors into structured property data usable by downstream modules. This data processing capability may include a data preprocessing unit and a feature calculation unit.

[0033] The data preprocessing unit is responsible for denoising and correcting the raw electrical or image signals output by the sensors. For example, it applies Savitzky-Golay smoothing filters to near-infrared spectral data to eliminate random noise; and it performs illumination non-uniformity correction and distortion correction on images acquired by the machine vision system.

[0034] The feature extraction unit is responsible for extracting quantified material properties from the preprocessed data. For near-infrared spectral data, this unit calls a pre-established chemometric model, such as a partial least squares regression model, to extract specific chemical property values ​​from the spectral matrix. The mathematical representation of this model is as follows: ; in, The vector of chemical properties of the material to be predicted (e.g., including moisture content); The input is a matrix of near-infrared spectral data; This is the regression coefficient matrix determined through calibration experiments; This is the residual matrix of the model.

[0035] For machine vision data, this feature solving unit applies a series of image processing algorithms, such as Canny edge detection and contour discovery algorithms to identify individual wood shavings, and then calculates the pixel area and minimum bounding rectangle of each contour to obtain the statistical distribution of morphological parameters such as wood shaving size and aspect ratio.

[0036] Finally, the material property perception module integrates all the calculated attribute values ​​into a standardized material property vector. It is then associated with timestamps and material batch information and output to the material property transfer token management module or other system modules that require this data.

[0037] In one specific implementation, the Material Property Transfer Token Management Module is responsible for the full lifecycle management of Material Property Transfer Tokens (MPPTs). These MPPTs are structured digital information packages used in this invention to bind a single material batch and carry its entire process information. The MPPT Management Module includes a Material Property Transfer Token data structure definition unit, an instance creation and initialization unit, a status update and routing unit, and an archive management unit.

[0038] The material property transfer token data structure definition unit specifies the data fields contained in each material property transfer token. In one embodiment, the data structure includes at least: A unique identifier field used to uniquely identify material batches; The timestamp field records the timestamps of each key event; Attribute set fields used to store the measured attribute vectors of materials before and after processing in each process unit in sequence; This field is used to store the instruction set fields that are recommended control instructions for each process unit, generated by the global instruction planning and correction module. This field is used to store constraint set information about instruction execution constraints fed back by the distributed instruction execution and negotiation module.

[0039] The function of the instance creation and initialization unit is to trigger a new material property transfer token instance when the material property sensing module at the production line inlet detects a new material batch and outputs its initial attribute vector. This unit generates a globally unique identifier (e.g., a universally unique identifier UUID or a string composed of a timestamp and batch number) and instantiates an MPPT object conforming to the aforementioned data structure. Subsequently, the unit fills the generated unique identifier and the acquired initial attribute vector into the corresponding fields of the MPPT instance, completing the initialization.

[0040] The State Update and Routing Unit is responsible for dynamically maintaining and transmitting MPPT instances throughout the production process. As a data exchange hub, this unit receives update requests from other modules in the system. When it receives updated data carrying a unique identifier for a specific MPPT, such as the measured attribute vector after processing by a process unit, or a revised recommended control instruction set, the State Update and Routing Unit performs an atomic write operation, appending or overwriting the new data to the corresponding fields of the MPPT instance to ensure data consistency and integrity. After completing the data update, the State Update and Routing Unit routes the updated MPPT instance to the next target module according to the preset control flow logic. For example, upon receiving a measured attribute update, the MPPT is routed to the global instruction planning and correction module, or upon receiving an instruction update, the MPPT is routed to the distributed instruction execution and negotiation module of the next process unit.

[0041] The archiving management unit is responsible for processing MPPT instances that have completed the entire production process. When a material batch is taken offline, the status of its corresponding MPPT instance is marked as terminated. This archiving management unit then removes this complete MPPT instance from the live database or memory and stores it in a historical database for long-term storage. To support data analysis by the process model self-optimization module, the archiving management unit can index key fields in the MPPT, such as unique identifiers, production completion time, key process parameters, and final product attributes, during storage to enable rapid retrieval and batch access to historical production data.

[0042] In one specific implementation, the global instruction planning and correction module, serving as the system's real-time decision-making unit, is typically deployed on the central process coprocessor (PCP). The core function of this module is to generate, correct, and replan control instructions applied to each process unit based on the real-time material properties carried by the Material Property Transfer Token (MPPT). This global instruction planning and correction module includes a global process model unit, an initial instruction planning unit, a dynamic feedforward correction unit, and a constraint-based replanning unit.

[0043] The global process model is embedded within the global process model unit. The global process model is a digital representation of the complete production process of Koelreuteria paniculata particleboard, describing the quantitative causal relationships between material properties, control commands, and process results. In one embodiment, It can be composed of multiple cascaded sub-models representing various process units (such as drying, mixing, hot pressing, etc.). These sub-models can be constructed using mechanistic models based on physicochemical principles, or empirical models or artificial intelligence models trained on historical production data, such as multiple regression models, neural network models, or expert systems. The model consists of a set of optimizable internal parameters. Characterization.

[0044] The function of the initial instruction planning unit is to generate initial recommended control instructions for each batch of materials entering the production line. When the global instruction planning and correction module receives an MPPT instance initialized by the material property transfer token management module, the initial instruction planning unit is triggered by extracting the initial material attribute vector from the MPPT. and use it as a global process model The input is used to calculate a recommended set of control instructions covering all downstream process units that optimizes the final product attributes. The calculation process can be expressed as follows: ; in, A set containing recommended control instructions for each process unit. ... ; Represents the global process model; This is the initial material attribute vector; This is the current parameter set of the global process model. The calculated parameters are... It is written into the instruction set field of the MPPT instance. This represents the total number of process units in the production process.

[0045] The dynamic feedforward correction unit realizes the adaptive control across process units in this invention. In the production process, when a process unit... Once processing is complete, the global instruction planning and correction module will receive updated measured attributes of the material exiting the process unit. The MPPT. Once this dynamic feedforward correction unit is triggered, it will use all known measured property sequences up to date. The global process model is invoked again as a new, more precise initial condition. For all downstream process units that have not yet been executed (from arrive A revised set of recommended control instructions was recalculated. It can compensate for any deviations generated by upstream process units by proactively adjusting the processing strategies of downstream process units.

[0046] The constraint-based replanning unit is used to handle the conflict between ideal instructions and physical device capabilities. When the global instruction planning and correction module receives an MPPT, its constraint set field contains the execution constraint vector fed back by a distributed instruction execution and negotiation module. At that time, the constraint-based replanning unit is triggered. It will then apply the constraint vector... (For example, maximum available power, maximum allowable temperature, etc.) are used to solve for this process unit. Hard boundary conditions for control commands. This constraint-based replanning unit calls the global process model. Under the constraints Under the premise of this, recalculate the suboptimal recommended control instructions for this process unit. The calculation process can be expressed as follows: ; in, These are the measured properties of the materials before they enter this process unit; The execution constraints fed back by this process unit; This is the current parameter set for the global process model. This replanning instruction... The update will be sent to MPPT and then sent back to the corresponding distributed instruction execution and negotiation module.

[0047] In one specific implementation, the distributed instruction execution and negotiation module, serving as the system's local execution and feedback unit, is deployed in the distributed control nodes (DCNs) of each process unit on the production line. This module aims to transform the recommended control instructions generated by the global instruction planning and correction module into precise operations on physical equipment, and to establish a closed-loop negotiation mechanism to ensure the physical feasibility of the instructions. The distributed instruction execution and negotiation module includes a state self-awareness unit, a feasibility prediction unit, a constraint feedback generation unit, and an instruction execution and monitoring unit.

[0048] The state-sensing unit is responsible for continuously monitoring the real-time operating status of the process equipment it controls. This unit collects and integrates data from various sensors connected to the equipment to form a multi-dimensional equipment state vector. The state vector It may include parameters describing the current performance and constraints of the equipment, such as the real-time temperature of the dryer's combustion chamber, the current pressure of the hot press's hydraulic system, the current load of the mixer's motor, and the cumulative working time or wear assessment value of key consumable parts (such as blades and filters).

[0049] The feasibility prediction unit is the core of the command negotiation function. This feasibility prediction unit internally contains a model of equipment physical capability constraints. The physical capability constraint model of this equipment is a set of rules or functions relating to the equipment's safe operating boundaries, maximum dynamic response rate, and performance limits under the current state. When this module receives the recommended control command vector for this process unit from the Material Property Transfer Token (MPPT). Then, the unit immediately invokes the constraint model and sets the current device state vector. With recommended instruction vector As input, it undergoes validation. The validation process can be formally described as follows: ; in, The output is a boolean value, representing the verification result; This represents a model representing the physical capacity constraints of the equipment. Indicates the process unit Recommended control command vector; Indicates process unit The real-time device state vector. If the output is true, it indicates that the instruction can be executed safely and effectively in the current device state, and the instruction is acknowledged; if the output is false, the constraint feedback generation unit is triggered.

[0050] The constraint feedback generation unit is activated when the feasibility prediction unit determines the result to be false. This unit's function is to generate structured execution constraint vectors. This vector not only indicates that the instruction is infeasible, but also specifically specifies the parameters that cause the infeasibility and their physical boundaries. For example, if the recommended instruction requires a heating rate that exceeds the maximum value currently achievable by the device, the unit generates a constraint vector containing the conflicting parameter names, the required instruction values, and the current achievable limits of the device. This execution constraint vector... It is then written into the constraint set field of the MPPT instance corresponding to the batch of materials, and sent back to the global instruction planning and correction module by the material attribute transfer token management module to request instruction replanning.

[0051] The instruction execution and monitoring unit is responsible for the final execution of the instructions. Once the feasibility assessment unit confirms the feasibility of the recommended control instruction (whether an initial instruction or one that has been negotiated and modified), this unit converts it from a logical instruction into an electrical signal or communication message that can be parsed by the underlying controller (such as a programmable logic controller (PLC) or a dedicated motion controller). During instruction execution, this unit continuously monitors the actual values ​​of key process variables using local sensors and compares them with the instruction setpoints. It then adjusts the output in real time using a local closed-loop control algorithm (such as proportional-integral-derivative PID control) to ensure that the physical execution process accurately tracks the instruction requirements.

[0052] In one specific implementation, the process model self-optimization module serves as the system's long-term learning and evolution unit. Its function is to utilize historical production data to optimize the global process model used by the global instruction planning and correction module. Offline optimization is performed to continuously improve the accuracy of system decisions. This process model self-optimization module includes a historical data batch processing unit, a model performance evaluation unit, a parameter optimization solution unit, and a model update management unit.

[0053] The historical data batch processing unit is responsible for constructing a dataset for model training from the historical database maintained by the Material Property Transfer Token (MPPT) management module. This unit is activated under preset triggering conditions (e.g., a fixed time period or when the number of archived MPPT samples reaches a specified threshold). After activation, the unit batch-extracts complete MPPT instances representing completed production processes from the database and uses the full-process data recorded in these instances—including initial material properties, intermediate process properties, final execution instructions, and process constraint feedback—to construct a structured training dataset. .

[0054] Model performance evaluation unit definition is used to quantify the global process model. Loss function of the deviation between prediction accuracy and actual production results In one embodiment, the loss function can be defined as the weighted mean square error between the predicted and measured attributes of all historical data samples and all process units. The specific mathematical expression for this function is:

[0055] in, Based on the model parameter set The loss function value for the variable; Represents the training dataset The total number of samples in the sample; This represents the total number of process units in the production process. This is the first Weighting coefficients for the prediction accuracy allocation of each process unit; It is the first The sample left the first Measured material attribute vectors after each process unit; The representation model is based on the first The sample enters the... Measured properties before each process unit and the control commands ultimately executed by this process unit Predicted values ​​for the output attributes of this process unit; It represents the square of the L2 norm of a vector.

[0056] The parameter optimization solution unit employs a numerical optimization algorithm, using the aforementioned loss function. To optimize the objective, solve for... Minimize the set of model parameters This parameter optimization unit can implement one or more iterative optimization algorithms, such as stochastic gradient descent and Adam (Adaptive Moment Estimation). The goal of this solution process is to find a set of optimal parameters. , so that: ; in, This indicates the search for a loss function that makes the loss function... Model parameters corresponding to the minimum value .

[0057] The model update management unit is responsible for safely deploying the optimized new model parameters to the online system. The parameter optimization and solution unit calculates the new parameter set. Subsequently, the model update management unit does not directly replace the currently used model parameters. In one embodiment, the model update management unit performs performance evaluation on an offline validation dataset using the model equipped with the new parameters to ensure that its prediction accuracy is better than the old model. After successful validation, the unit updates the model parameters called by the global instruction planning and correction module in a transactional manner. In addition, the model update management unit also archives and manages the model parameters of historical versions, so as to perform retrospective analysis of system performance or version review when needed.

[0058] See attached document Figure 3 , Figure 3 This is a flowchart of an intelligent control method for the entire production process of Koelreuteria paniculata particleboard according to an embodiment of the present invention. The present invention provides an intelligent control method for the entire production process of Koelreuteria paniculata particleboard, comprising the following steps: S1. Initial attribute collection is performed on the batch of materials entering the production line, and an initial recommended control instruction set covering the entire process is generated based on these initial attributes. Specifically, the material attribute sensing module collects the initial attribute vector of the newly entering batch of materials; the material attribute transfer token management module generates a unique material attribute transfer token (MPPT) for the batch of materials and writes the initial attribute vector into the MPPT; the global instruction planning and correction module receives the initialized MPPT, calls the global process model to plan a set of initial recommended control instructions for it, and writes the instruction set into the MPPT.

[0059] S2 performs local feasibility prediction and closed-loop negotiation on the received recommended control instructions, and executes the finally confirmed instructions. Specifically, the distributed instruction execution and negotiation module of the corresponding process unit receives the recommended control instructions in MPPT and performs feasibility prediction based on its own real-time operating status. If the prediction is feasible, the instruction is executed; if the prediction is infeasible, an execution constraint vector containing specific physical boundaries is generated and sent back to the global instruction planning and correction module through MPPT to request instruction replanning until a control instruction that has been locally confirmed as executable is generated and executed.

[0060] S3 measures the post-processing properties of the material and uses these properties to make feedforward corrections to the recommended control commands for all downstream process units. Specifically, the material property sensing module collects the measured property vector of the batch of material after it leaves the current process unit; the material property transfer token management module updates the measured property vector to the MPPT; the global command planning and correction module receives the updated MPPT and, based on this latest measured property vector, recalculates and corrects the recommended control commands for all downstream process units.

[0061] S4 involves passing the updated material property transfer token level by level on the production line, and for each subsequent process unit, repeating the negotiated execution of step S2 and the measured correction process of step S3. This process is cascaded until all production processes for that material batch are completed.

[0062] S5 archives the material property transfer tokens (MPPTs) that complete the entire process and performs periodic self-optimization of the global process model based on accumulated historical data. Specifically, the material property transfer token management module archives the MPPTs that complete all production processes to build a historical production dataset; the process model self-optimization module periodically calls this historical production dataset, solves a loss function aimed at minimizing model prediction bias, iteratively optimizes the internal parameters of the global process model, and deploys the optimized model parameters to the system.

[0063] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A fully intelligent control system for the production process of goldenrain tree particleboard, characterized in that: include: The material attribute sensing module is used to collect the initial material attribute vectors of batches of materials entering the production line at different process unit nodes; The material property transfer token management module is used to receive the collected initial material property vector, generate and initialize a unique material property transfer token based on the initial material property vector, the material property transfer token carries the full process information of this material batch, and updates, transfers and archives the material property transfer token according to the system running status. The global instruction planning and correction module is used to receive the material attribute transfer token containing the material attribute vector, and use the global process model that digitally represents each process unit in the production process of goldenrain wood particleboard, including drying, glue mixing and hot pressing stages, to plan or correct the recommended control instruction set of each process unit, and write the recommended control instruction set into the instruction set field of the material attribute transfer token. The distributed instruction execution and negotiation module is used to receive the material property transfer token containing the recommended control instruction and to perform a local feasibility prediction on the recommended control instruction. The process model self-optimization module is used to call the historical material property transfer token data archived by the material property transfer token management module to periodically optimize the global process model in the global instruction planning and correction module. The global instruction planning and correction module includes: The global process model unit has a built-in global process model that is a digital simulator of the entire production process of goldenrain tree particleboard. Based on material properties and control instructions, it predicts the processing results of each process unit. The initial instruction planning unit is used to receive the initial material attribute vector contained in the material attribute transfer token, and based on the iterative optimization method of the global process model, with the goal of optimizing the final product attributes, reversely derive and construct the initial recommended control instruction set. The dynamic feedforward correction unit is used to receive the material property transfer token containing the material property vector and call the global process model to dynamically adjust the initial recommended control instruction set to generate the corrected recommended control instruction set. The constraint-based replanning unit receives a material property transfer token containing an execution constraint vector and calls the global process model to replan the modified recommended control instruction set. Under the premise of satisfying the execution constraints, it recalculates the suboptimal recommended control instructions.

2. The intelligent control system for the entire production process of Koelreuteria paniculata particleboard according to claim 1, characterized in that, The material property sensing module includes one or more online sensing units deployed at key process unit nodes of the production line. Each online sensing unit integrates various sensor hardware to collect multi-dimensional material properties. The sensor hardware includes: Near-infrared spectrometer used for collecting near-infrared spectral data of materials; Machine vision systems used to acquire morphological features of materials; Microwave sensors used to measure the bulk properties of materials.

3. The intelligent control system for the entire production process of Koelreuteria paniculata particleboard according to claim 2, characterized in that, The material property sensing module also includes an internally integrated data processing function, which includes a data preprocessing unit and a feature calculation unit. The data preprocessing unit is used to reduce noise and correct the raw signals output by various sensor hardware. The feature calculation unit is used to extract quantified material properties from the noise-reduced and corrected data. The extraction process includes calling a pre-established chemometric model to calculate the near-infrared spectral data collected by the near-infrared spectrometer into material chemical properties.

4. The intelligent control system for the entire production process of Koelreuteria paniculata particleboard according to claim 1, characterized in that, The material property transfer token management module includes: The material property transfer token data structure definition unit is used to define the data fields contained in the material property transfer token and to provide a structural template for the creation and management of material property transfer token instances. The instance creation and initialization unit is used to create and initialize a unique material property transfer token instance based on the data fields determined by the material property transfer token data structure definition unit and the received initial material property vector, so as to carry the full process information of this material batch. The status update and routing unit is used to receive the unique material property transfer token instance and dynamically update the material property vector and control instruction set in the material property transfer token instance based on the feedback from each process unit in the production process. The archiving management unit is used to receive the material property transfer token instance that has been transmitted by the status update and routing unit after the completion of the entire production process, and to store the material property transfer token instance in the historical database.

5. The intelligent control system for the entire production process of Koelreuteria paniculata particleboard according to claim 1, characterized in that, The initial instruction planning unit receives the initial material attribute vector from the material attribute transfer token and uses the initial material attribute vector as input to the global process model. It then calculates the initial recommended control instruction set through forward calculation, which is expressed as follows: ； in, A set containing recommended control instructions for each process unit. ... , Represents the global process model. This is the initial material attribute vector. This is the current parameter set of the global process model. This represents the total number of process units in the production process.

6. The intelligent control system for the entire production process of Koelreuteria paniculata particleboard according to claim 1, characterized in that, The constraint-based replanning unit receives the execution constraint vector from the material property transfer token. and the execution constraint vector As a solution process unit The hard boundary conditions for control commands are invoked by calling the global process model, while satisfying the execution constraint vector. Under the premise of [the above conditions], the suboptimal recommended control instructions are recalculated for the process unit.

7. The intelligent control system for the entire production process of Koelreuteria paniculata particleboard according to claim 1, characterized in that, The distributed instruction execution and negotiation module is installed in each process unit of the production process, and the distributed instruction execution and negotiation module includes: The state self-sensing unit is used to continuously monitor the real-time operating status of the controlled process unit and form a multi-dimensional equipment state vector. ; The feasibility prediction unit has a built-in physical capability constraint model for determining whether recommended control commands can be safely executed. It uses a multi-dimensional equipment state vector and the recommended control command vector of the process unit as inputs for verification calculations. The verification process can be represented as follows: ； in, The output is a boolean value representing the verification result. This represents a model representing the physical capacity constraints of the equipment. Indicates the process unit Recommended control command vector, Indicates process unit Real-time device state vector; The constraint feedback generation unit is used in the... When the result is false, an execution constraint vector containing specific physical boundaries is generated, and the execution constraint vector is written into the constraint set field of the material property transfer token; Instruction execution and monitoring unit, used in the above When true, the recommended control command is translated into an operation on the physical device and monitored.

8. The intelligent control system for the entire production process of Koelreuteria paniculata particleboard according to claim 1, characterized in that, The process model self-optimization module includes: A historical data batch processing unit is used to construct a dataset from a historical database for training the global process model; The model performance evaluation unit is used to define a loss function, which quantifies the deviation between the global process model's predicted values ​​of the processing results for each batch of materials in the dataset and the actual production results. The loss function can be defined as follows: ； in, Based on the model parameter set The loss function value for the variable. Represents the training dataset The total number of samples in the sample, This represents the total number of process units in the production process. This is the first Weighting coefficients for the prediction accuracy allocation of each process unit. It is the first The sample left the first The measured material property vector after each process unit The representation model is based on the first The sample enters the... Measured properties before each process unit and the control commands ultimately executed by this process unit Predicted values ​​for the output attributes of this process unit. The square of the L2 norm of a vector; The parameter optimization and solution unit is used to solve for the model parameter set that minimizes the loss function by employing a numerical optimization algorithm with the loss function as the optimization objective. The model update management unit is used to deploy the solved model parameter set to the global instruction planning and correction module to update the global process model.

9. A method for intelligent control of the entire production process of Koelreuteria paniculata particleboard, applied to the intelligent control system for the entire production process of Koelreuteria paniculata particleboard as described in any one of claims 1-8, characterized in that, Includes the following steps: S1 collects the initial attributes of the batches of materials entering the production line and generates an initial recommended control instruction set covering the entire process based on these initial attributes. S2, the distributed instruction execution and negotiation module receives the initial recommended control instruction set, performs local feasibility prediction and closed-loop negotiation, and executes the finally confirmed instructions; S3, to measure the properties of the material after processing, and to make feedforward corrections to the recommended control commands of all process units based on the measurement results; S4 transmits the updated material property transfer tokens level by level on the production line, and repeats the feasibility prediction, closed-loop negotiation, and feedforward correction process for subsequent process units. S5. Archive the material property transfer tokens that have completed the entire process, use a numerical optimization algorithm to solve for the optimized model parameter set, and update the global process model using the optimized model parameter set.

Images