Wood deep processing process parameter intelligent optimization system based on industrial internet

Abstract

The application provides a kind of wood deep processing process parameter intelligent optimization system based on industrial internet, it is related to industrial internet process parameter optimization technical field, system includes dielectric thermal multi-parameter online perception etc. Module, embedded in multi-source in-situ sensing network, capture slab dielectric response and thermodynamic state upload to cloud, build the digital twin engine embedded in moisture diffusion and adhesive curing kinetics constraints, inversion slab global reaction uniformity and curing process;System uses multi-branch collaborative decision mechanism, dynamically solves the multi-dimensional joint modulation strategy covering high-frequency power supply operating frequency, electrode driving voltage, vacuum degree and hydraulic current, and penetrates the completion of telecommunication signal reconstruction with the help of industrial internet hierarchical architecture, combined with stress wave nondestructive testing feedback calculation comprehensive reward value;The application solves the adjustment conflict under the condition of multi-physical field strong coupling, reduces the core layer under-dry, surface coking and bubble stratification defects, improves the consistency of finished product and energy utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of industrial internet process parameter optimization technology, and in particular to an intelligent optimization system for wood deep processing process parameters based on industrial internet. Background Technology

[0002] As a natural biomass material, wood is seeing a surge in demand for deep-processed products such as plywood, blockboard, and engineered wood products made from fast-growing poplar, eucalyptus, and imported pine, driven by my country's vigorous promotion of green upgrading in the home furnishing industry. However, in actual production, even from the same log and the same batch of sawn boards, significant differences in moisture content, oven-dry density, and fluid permeability at different cut points exist due to uneven release of growth stress, differences in earlywood and latewood density, and micro-cracks remaining in the early drying stage. This inherent volatility in raw materials presents a fundamental and irreconcilable contradiction with the continuous, large-scale, and uniform production model pursued by modern factories. For a long time, the mainstream practice in the industry has been to set hot-pressing temperatures, pressing times, and adhesive coating amounts based on so-called optimal process packages obtained through offline experiments. These process packages are essentially static parameter sets built on the assumption of homogenization, attempting to use a fixed curve to cope with the highly discrete physical properties of the processed object. The inevitable result is that quality risks are amplified at each stage. Taking high-frequency vacuum hot pressing, a core process in deep wood processing, as an example, after the board is placed between the electrodes, it needs to be heated from the inside out under the influence of an electromagnetic field. This causes the moisture to vaporize rapidly and be removed by the vacuum system, while the adhesive completes cross-linking and curing under thermal activation. Theoretically, this process involves multiple complex physicochemical processes, such as the attenuation and penetration of electromagnetic waves in a lossy medium, the phase transition migration of free water and bound water, the softening and reorganization of hemicellulose and lignin, and the formation of chemical bonds in the adhesive layer. However, in real workshops, faced with non-standard raw materials with moisture content that can vary by more than ten percentage points, on-site operators often have to rely on experience to visually assess the color of the board surface or feel the humidity, and roughly adjust the power knob and timer on the equipment panel. Due to the lack of transparent means of perceiving the instantaneous reaction state inside the board, it is often only in the middle and late stages of hot pressing that it is discovered that the core layer moisture has not been completely removed, forcing an extension of the holding time. This not only results in wasted energy but also easily leads to pyrolysis and carbonization of the surface wood due to overheating, while at this time the core adhesive layer may not have reached the accumulated heat required for complete curing.

[0003] In recent years, the Industrial Internet has swept through the wood processing industry, with many manufacturing companies investing in equipment IoT systems, manufacturing execution systems (MES), and online visual inspection devices. By installing vibration and temperature sensors on the hot press spindle, hydraulic station, and steam pipeline, companies have indeed achieved transparency in equipment operation and maintenance early warnings. Using machine vision to identify surface defects such as blistering and sand leakage has also improved the pass rate of finished products to some extent. However, these information technology methods only solve the problem of making data visible and management processes smooth. The data they collect mostly stays at the peripheral level, such as equipment start-up and shutdown, current and voltage, and finished product appearance quality, without truly touching the core logic of process control. Existing information systems generally adopt a layered and isolated architecture. The execution of electrical signals at the equipment end is managed by the PLC, and the setting of process parameters is managed by the MES. What chemical reactions actually occur inside the wood relies entirely on the human experience of process engineers to guess. This disconnect between information and action means that even if companies accumulate massive amounts of production data, they cannot proactively intervene in equipment operation midway through the hot pressing process based on the evolution of the internal moisture and curing fields. Some solutions improve heating efficiency by combining steam and high-frequency dual heat sources; others have developed a breathing-type hot-pressing regime based on changes in the internal steam pressure of the wood; still others have specified stepped heating and cooling ranges for different tree species. A few solutions introduce sensor feedback to determine the current temperature or pressure, simply triggering an alarm or shutdown if the pressure exceeds the limit, completely lacking the ability to reverse-engineer and optimize electrical signal output strategies based on differences in the internal reaction processes of the wood. Process engineers write operating procedures based on paper or screen reaction mechanisms, electricians adjust high-frequency power supplies and hydraulic proportional valves at the control console, and information software schedules and reports production on office computers. Due to the lack of a unified architecture that can convert reaction states such as dielectric properties, temperature field, and moisture content into plate drive voltage and servo current modulation in real time, the existing control loops are extremely fragile. For example, during the hot pressing of thick boards, if heating is stopped simply once the surface temperature reaches the set value, the electromagnetic waves often haven't penetrated to the core layer. Internal moisture remains trapped within the board, forming high-pressure steam, which easily causes large-scale blistering and delamination after pressing. Conversely, if heating time is excessively extended to ensure the core layer is completely dry, the surface layer will char due to excessive energy accumulation. Similarly, if the residual internal stress from the hot pressing process cannot be quantified and transmitted to the subsequent curing chamber in a timely manner, the curing process can only rely on extending the storage time and hoping for the best, significantly consuming the company's cash flow and warehouse space. Therefore, there is an urgent need in the market for a new intelligent optimization system for wood deep processing parameters based on the Industrial Internet. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent optimization system for wood deep processing parameters based on the Industrial Internet, in order to solve the technical bottlenecks in existing technologies, such as the extreme opacity of the internal moisture migration trajectory and adhesive curing process caused by wood as a typical heterogeneous porous medium, the inability to detect the strong fluctuations of fast-growing and high-moisture raw materials in real time by traditional surface temperature monitoring methods, leading to mismatch in hot pressing processes, and the fact that existing solutions' Industrial Internet platforms only stop at the level of data acquisition and visualization without converting dielectric response into dynamic execution commands for underlying equipment, and the lack of coordinated feedforward based on physical state between upstream and downstream processes, resulting in both insufficient core drying and surface charring, and persistently high residual stress in the finished product. The specific technical solution is as follows: This invention provides an intelligent optimization system for wood deep processing parameters based on the Industrial Internet. The system includes an online sensing module for multiple dielectric and thermal parameters, a reaction principle-driven digital twin module, a multi-branch collaborative decision-making module, a hierarchical electrical signal modulation module, and a cross-process closed-loop self-evolution module. The dielectric thermal multi-parameter online sensing module is used to perform multi-dimensional in-situ measurements of slabs under high-frequency vacuum hot pressing environment with sub-second time resolution, and convert the physical response into a standardized wood state feature vector. The reaction principle-driven digital twin module is used in a hybrid modeling architecture based on physical priors to receive the feature vector and output a set of reaction process indicators for the entire slab domain in real time. The multi-branch collaborative decision-making module is used to generate a joint modulation instruction matrix based on the reaction process index set through a hierarchical multi-agent deep reinforcement learning framework. The hierarchical electrical signal modulation module is used to decompose the joint modulation instruction matrix into execution instructions of different levels and issue them based on the hierarchical architecture of the Industrial Internet, so as to complete the dynamic modulation and process execution of electrical signals. The cross-process closed-loop self-evolution module is used to collect the actual state feedback value of the slab, update the weights of the coupled reaction model and decision algorithm online, and output collaborative control parameters to upstream and downstream processes.

[0005] Furthermore, the online sensing module for multiple dielectric and thermal parameters includes a flexible implantable sensing unit, an edge signal conditioning unit, and a high-precision synchronous acquisition card; The flexible implantable sensing unit is arranged in a matrix along the slab surface, specifically including a coaxial array dielectric probe, a thin-film platinum resistance thermometer, and a miniature pressure-sensitive thin film; the coaxial array dielectric probe operates at a 1 MHz excitation frequency and is sensitive to the polarization relaxation of free water and adsorbed water in the wood; the thin-film sensor is attached to the gap between the veneers in a non-invasive manner. The edge signal conditioning unit uses lock-in amplification technology to extract the imaginary part of the dielectric signal and suppress common-mode interference caused by high-frequency electromagnetic fields. The synchronous acquisition card supports hardware triggering, ensuring that the sampling points of all channels are strictly synchronized with the power frequency cycle of the high-frequency power supply, thus guaranteeing strict alignment of multiple physical quantities in the spatiotemporal dimension.

[0006] Furthermore, the online sensing module for dielectric and thermal multi-parameters specifically scans the slab in sections using a coaxial array probe to obtain the relative permittivity and dielectric loss tangent of each sub-region. The initial moisture content distribution field is obtained by inverting the dielectric hybrid model. Based on the multi-point temperature and pressure sequence, the instantaneous temperature gradient field and pressure distribution in the thickness direction of the slab are fitted using the moving least squares method. After the collected physical data is processed by moving average filtering and normalization, it is aggregated to generate a standardized wood state feature vector containing dielectric constant, dielectric loss tangent, thickness temperature gradient, core layer measured pressure, high-frequency power supply operating frequency setting, and electrode operating voltage setting.

[0007] Furthermore, the reaction principle-driven digital twin module adopts a physical information neural network architecture, incorporating Fick's second law describing water migration and Arrhenius's equation describing adhesive crosslinking as soft constraints into the network loss function. The physical information neural network architecture includes a shared coding network, a water migration physical branch, a solidification dynamics physical branch, and a physical constraint decoding layer; The shared coding network maps the feature vectors to a high-dimensional latent space; The physical branch of water migration outputs the water content decay rate of each zone, and uses the residual of the unsteady water diffusion equation as the physical loss term by automatic differentiation. The curing kinetics branch predicts activation energy and reaction rate constant based on latent space characteristics, and calculates real-time degree of curing by combining Arrhenius relation; The physical constraint decoding layer integrates the results of the two branches and outputs a set of reaction process indicators including the predicted core layer moisture content, the adhesive layer curing degree index, and the residual stress risk factor, while simultaneously calculating the uniformity thermal map of the entire slab.

[0008] Furthermore, the multi-branch collaborative decision-making module includes a state assessment branch, a thermodynamic modulation branch, a mechanical modulation branch, and a collaborative arbitration network; The state assessment branch receives the reaction process index set in real time, quantifies the deviation vector of the slab from the ideal curing state, and outputs the quality potential gradient. The thermodynamic modulation branch input includes a deviation vector and dielectric loss feedback, and integrates an attention-enhanced recurrent neural network to dynamically output the operating frequency of the high-frequency power supply and the plate driving voltage. The network end is connected to a micro thermal balance solver to verify the theoretical temperature rise rate. The mechanical modulation branch determines the vacuum degree and hydraulic servo current based on the progress of curing degree and quality potential gradient, and uses a conditional residual network to take wood type and veneer thickness as dynamic parameters. The collaborative arbitration network calculates the coupling matching degree between the thermodynamic branch and the mechanical branch outputs, and outputs a joint modulation command matrix that includes the target operating frequency of the high-frequency power supply, the target driving voltage of the electrode plate, the target vacuum degree, and the target control current of the hydraulic servo system.

[0009] Furthermore, the multi-branch collaborative decision-making module has built-in operating condition discrimination logic. When it is determined that the core layer inversion water content is more than 15 percent higher than the surface layer, the penetration enhancement modulation strategy is automatically triggered to lower the high-frequency power supply operating frequency and simultaneously increase the electrode driving voltage, thereby realizing abnormal handling of multiple parameters. In the multi-branch collaborative decision-making module, if the adhesive curing degree index is still lower than the preset threshold after the pressure holding time is preset, a temperature-increasing and pressure-boosting modulation strategy is adopted to switch the high-frequency excitation to the pulse intermittent mode, increase the hydraulic pressure, and adjust the overall target temperature of the slab according to the temperature sensitivity to activate the curing agent activity. If the dielectric loss tangent value exhibits abnormal amplitude oscillation, an intermittent homogeneous modulation strategy is adopted. The high-frequency power output is paused and the vacuum pump is kept running at full speed to expel the internally accumulated free water vapor. Then, the power supply is restored in a low-energy mode and the power is increased stepwise until the dielectric signal is stable. The unique dielectric feature code is recorded and stored in the edge-side feature fingerprint database.

[0010] Furthermore, the hierarchical electrical signal modulation module implements instruction penetration based on the five-layer industrial internet architecture. The joint modulation instruction matrix is ​​parsed into device semantic instructions at the operation layer, and then sent to the programmable logic controller via time-sensitive network and OPCUA protocol. For high-frequency power supplies, the operating frequency is converted into a frequency control word, and the adjustment resolution is accurate to 0.1 MHz. For plate control, the pulse width modulation duty cycle of the insulated gate bipolar transistor is dynamically rewritten, and the plate spacing is adjusted in 1 mm increments via a lead screw actuator. For hydraulic and vacuum systems, the servo current and vacuum level are mapped to the real-time current or frequency setpoint of the proportional valve and frequency converter. During execution, the deviation between the actual feedback value and the instruction setting value is collected every 500 milliseconds. If the deviation is greater than the allowable tolerance, local PID fine-tuning is triggered, and the fine-tuning record is reported as a working condition disturbance event.

[0011] Furthermore, during the execution of instructions, if the temperature rise rate of the core layer of the slab exceeds the safety threshold per unit time, the hierarchical electrical signal modulation module adopts a gentle slope current limiting execution mode to limit the maximum output current of the high-frequency inverter unit, increase the electrode spacing to dilute the electromagnetic energy density, and activate the circulating cooling water valve built into the hot press plate for cooling protection. If the deviation between the vacuum feedback value and the set value continues to exceed the preset deviation value, the fault-tolerant redundant execution mode will be adopted to automatically identify and issue an early warning, compensate the high-frequency power supply output power proportionally, mark the batch of slabs as objects to be strengthened and cured, and automatically generate a maintenance work order after hot pressing is completed. If the pressure standard deviation of each zone of the multi-cylinder hot press is greater than the preset value, the pressure difference equalization execution mode is adopted to generate a zone pressure regulation matrix. Current bias commands are issued to the high-pressure zone and the low-pressure zone respectively to ensure that the total pressure is constant until the pressure standard deviation of the entire zone converges to the allowable range.

[0012] Furthermore, the cross-process closed-loop self-evolution module includes a circular port detection module and a parameter closed-loop adjustment module; The circular opening detection module extracts the actual mass vector containing elastic modulus, bonding strength and internal defect distribution through stress wave tomography and hyperspectral imaging. The parameter closed-loop adjustment module calculates the reinforcement learning comprehensive reward value by using the difference between the actual quality vector and the expected quality. The reward value is composed of a weighted average of the finished product quality item, energy consumption item, and defect penalty item. Based on this reward value, the system uses a strategy gradient algorithm in the background to periodically fine-tune the network weights of the multi-branch collaborative decision-making module. At the same time, it pushes the environmental parameter requirements of the health care process as a feedforward signal to the downstream system to guide the dynamic setting of the temperature and humidity slow-release curve of the health care room. If the residual stress detected by the stress wave is higher than the preset threshold, the cross-process closed-loop self-evolution module adopts a deep self-reflection update mode, freezes the moisture diffusion coefficient in the digital twin model, extracts the full-cycle sequence of abnormal slabs and labels it as high-value negative samples, expands the training set with an adversarial generative network and fine-tunes the strategy network, and is re-launched after verification.

[0013] Furthermore, the system is configured to execute an intelligent optimization method for wood deep processing parameters, the method comprising: S10 synchronously acquires dielectric response signals, thermodynamic signals, and equipment-end electrical signals of wood blanks during hot pressing through multi-source heterogeneous sensing units, and performs spatiotemporal alignment and preprocessing on the multi-source data. S20 is a process twin engine that uses time-sensitive networking to upload pre-processed multi-dimensional data streams to the cloud industrial internet platform via an edge industrial gateway. S30 utilizes a process twin engine to construct a coupled reaction model of moisture migration and adhesive curing within the slab, and inverts the global reaction uniformity and curing degree indices in real time. S40, based on real-time inversion indicators, solves the joint modulation strategy of high-frequency power supply frequency, plate spacing, vacuum degree and hydraulic pressure through built-in multi-objective reinforcement learning algorithm; S50 breaks down the joint modulation strategy into execution instructions at different levels, and sends them down to the field equipment layer to complete the dynamic modulation of electrical signals and process execution. S60 collects the actual status feedback value of the slab after execution, performs online incremental updates to the cloud-based coupling model and algorithm weights, and outputs collaborative control parameters to upstream and downstream processes.

[0014] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, performs the steps of the method described herein.

[0015] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the method.

[0016] The beneficial effects of this invention are as follows: By in-situ deploying dielectric and thermal sensing networks within a high-frequency vacuum hot-pressing mold, the system achieves full-time transparent perception of the internal moisture field and curing process of heterogeneous slabs, ending the blind speculation of internal states based solely on surface temperature or human experience. Furthermore, by employing a digital twin engine embedded with constraints on moisture diffusion and cross-linking kinetics, the previously fragmented physicochemical reaction mechanisms are transformed into calculable digital indicators, ensuring the scientific validity and generalizability of decision-making logic when dealing with different tree species and non-standard raw materials. Simultaneously, the system utilizes a multi-branch collaborative decision-making architecture combined with quality potential gradient dynamics to solve for the high-frequency power supply operating frequency, electrode driving voltage, and hydraulic vacuum... The joint modulation matrix of multi-dimensional parameters such as air density, and the transmission of electrical signals to the underlying execution end through the industrial internet hierarchical architecture, effectively bridges the mismatch between reaction demand and energy input. It properly resolves the long-standing multi-physics coupling conflicts such as heat transfer lag in thick core layers, under-curing of adhesive layers, and steam accumulation and bubbling. In addition, the cross-process closed-loop self-evolution mechanism seamlessly connects post-compression stress feedback with curing environment parameters, and has the ability to optimize models online and enable offline autonomy at the edge. As a result, the overall consistency of finished adhesive strength and energy utilization efficiency are greatly improved, providing a practical technical support for the wood processing industry to move from experience-driven to a stage of deep integration of principle-driven and data-driven intelligent processing.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the steps of an intelligent optimization system for wood deep processing parameters based on the Industrial Internet, according to the present invention. Figure 2 This is a schematic diagram of the structure of an intelligent optimization method for wood deep processing parameters based on the Industrial Internet, according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0021] It should also be noted that when a component is described as "fixed to" or "set on" another component, it can be directly on the other component or there may be an intervening component present. When a component is described as "connected to" another component, it can be directly connected to the other component or there may be an intervening component present.

[0022] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0023] In an embodiment of the present invention, an intelligent optimization system for wood deep processing parameters based on the Industrial Internet is provided. Please refer to [link / reference]. Figure 1The system includes: an online sensing module for multiple dielectric and thermal parameters, a reaction principle-driven digital twin module, a multi-branch collaborative decision-making module, a hierarchical electrical signal modulation module, and a cross-process closed-loop self-evolution module.

[0024] S1, the online sensing module for dielectric and thermal multi-parameters, is the starting point for sensing the connection between physical wood and digital space. Its goal is to perform multi-dimensional, anti-interference, and in-situ measurements of slabs under high-frequency vacuum hot pressing environment with sub-second time resolution, and to convert the physical response into a standardized digital feature vector that can be processed by the intelligent engine.

[0025] To achieve simultaneous capture of the internal dielectric properties, thermodynamic state, and equipment execution signals of wood, this invention constructs a multi-sensor collaborative in-situ detection array, which is embedded between a high-frequency electrode plate and a hot-pressing plate.

[0026] The platform consists of a flexible implantable sensing unit, an edge signal conditioning unit, and a high-precision synchronous acquisition card.

[0027] The flexible implantable sensing unit is arranged in a matrix along the surface of the slab and consists of a coaxial array dielectric probe, a thin-film platinum resistance thermometer, and a miniature pressure-sensitive film. The coaxial array dielectric probe operates at an excitation frequency of 1.0 MHz and is highly sensitive to the polarization relaxation of free and adsorbed water in the wood. The thin-film sensor is non-invasively attached to the gap between the veneers, without affecting normal gluing and heat conduction.

[0028] The edge signal conditioning unit uses lock-in amplification technology to extract weak dielectric imaginary part signals and suppress common-mode interference caused by high-frequency electromagnetic fields.

[0029] The synchronous acquisition card supports hardware triggering, ensuring that the sampling points of all channels are strictly synchronized with the power frequency cycle of the high-frequency power supply, thus guaranteeing strict alignment of multiple physical quantities in the spatiotemporal dimension.

[0030] Specifically, the measurement and feature extraction methods are as follows: First, the coaxial line array probe scans the slab in sections to obtain the relative permittivity and dielectric loss tangent of each sub-region. Since the permittivity has a piecewise linear relationship with the oven-dry wood matrix and moisture content, the system uses a built-in dielectric mixing model to invert the initial moisture content distribution field.

[0031] Secondly, based on multi-point temperature and pressure sequences, the moving least squares method is used to fit the instantaneous temperature gradient field and pressure distribution along the thickness direction of the slab.

[0032] Furthermore, the collected physical data is processed through moving average filtering and normalization, then aggregated to edge computing nodes and packaged to generate standardized timber state feature vectors: Where n is the total number of slab space partitions. Let be the dielectric constant of the i-th partition. This represents the temperature gradient along the thickness direction, expressed in Kelvin per meter. This is the measured pressure in the core layer, in Pascals. This sets the current operating frequency of the high-frequency power supply, in Hertz (Hz). The voltage is set for the current operation of the electrode plate, in volts, providing a holographic input for subsequent decisions.

[0033] S2, the reaction principle-driven digital twin module is the technical core of this invention, used to reveal the physicochemical evolution laws hidden inside wood. Traditional data models lack generalization ability; therefore, this invention constructs a hybrid modeling architecture that embeds physical priors, elevating empirical knowledge to a computable digital mechanism.

[0034] This invention employs a Physics-Informed Neural Network (PINN) architecture, incorporating Fick's second law describing moisture migration and the Arrhenius equation describing adhesive crosslinking as soft constraints into the network loss function. The network input is a feature vector uploaded by the sensing module, and the output is a set of reaction process indicators covering the entire slab, including predicted core layer moisture content, adhesive curing degree index, and residual stress risk factor.

[0035] The PINN architecture includes a shared coding network, a water migration physical branch, a solidification kinetics physical branch, and a physical constraint decoding layer.

[0036] S21. The shared coding network maps feature vectors to a high-dimensional latent space, extracting a general representation covering temperature, electric field, and material properties.

[0037] S22, the physical branch of water migration receives latent space features and outputs the water content decay rate of each partition. During network training and inference, the following differentiable form of the unsteady water diffusion equation is enforced: Where M is the local moisture content of the wood (wet basis), in kilograms per kilogram, and t is the hot-pressing time, in seconds. The moisture content is expressed as the rate of change over time in kilograms per second, and D is the apparent moisture diffusion coefficient of wood, expressed as square meters per second. The Laplace operator (second gradient) for the spatial distribution of moisture content, in kilograms per kilogram per square meter. Z represents the water vapor flux density along the thickness direction of the slab, in kilograms per square meter per second; z represents the spatial coordinate along the thickness direction of the slab, in meters. Let be the gradient of water vapor flux along the thickness direction, in kilograms per cubic meter per second. The system calculates the residual of the above equation by automatic differentiation and adds it as a physical loss term to the total loss to ensure that the network output does not violate the law of conservation of mass.

[0038] S23. The curing kinetics branch predicts activation energy and reaction rate constants based on latent space characteristics, and calculates real-time degree of curing using the Arrhenius relation: Where C is the curing degree index of the adhesive layer, with a value ranging from 0 to 1. The percentage change in degree of cure over time, expressed in seconds. This is the pre-exponential factor for the curing reaction, expressed in seconds. The activation energy of the crosslinking reaction of the adhesive is expressed in joules per mole. is the universal gas constant, with a value of 8.314, and the unit is joules per mole per Kelvin. T is the thermodynamic absolute temperature of the gel layer, and the unit is Kelvin. This represents the order of the curing reaction.

[0039] S24, the physical constraint decoding layer, integrates the results of the two branches, outputs a set of reaction process indicators, and simultaneously calculates the uniformity thermogram of the entire slab. If moisture accumulation or solidification stagnation occurs in a certain area, the system immediately generates an abnormal warning signal and embeds it into the feature flow.

[0040] S3, the multi-branch collaborative decision-making module, is the core of generating the optimal process strategy. This invention adopts a hierarchical multi-agent deep reinforcement learning framework to replace the traditional single-road-condition feedback control, achieving dynamic matching between response demand and execution capability.

[0041] This module includes a state assessment branch, a thermodynamic modulation branch, a mechanical modulation branch, and a collaborative arbitration network.

[0042] S31. The state assessment branch receives the reaction process index set in real time, quantifies the deviation vector of the current slab from the ideal curing state, and outputs the quality potential gradient, indicating the physical quantities that should be compensated first to improve the bonding strength.

[0043] S32, the thermodynamic modulation branch, focuses on the decision-making of energy input form. Its input includes the deviation vector of the slab from the ideal solidification state and dielectric loss feedback. Internally, it integrates an attention-enhanced recurrent neural network to dynamically output the operating frequency of the high-frequency power supply and the plate driving voltage. To prevent local overheating, a miniature thermal balance solver is connected to the end of the network to verify the theoretical temperature rise rate under the current energy input. If it exceeds the safety threshold, a reverse gradient correction output is generated.

[0044] S33, the mechanical modulation branch determines the vacuum degree and hydraulic servo current based on the progress of curing degree and quality potential gradient. It adopts a conditional residual network and uses prior knowledge such as wood type and veneer thickness as dynamic parameters for conditional batch normalization, so that the same network can adaptively handle different working conditions from thin wood veneer to thick board integration.

[0045] S34. The collaborative arbitration network calculates the coupling matching degree between the thermodynamic and mechanical branch outputs. For example, when the high-frequency is lowered to enhance penetration, the arbitration network automatically prompts the mechanical branch to appropriately increase the vacuum level to match the increased water evaporation and prevent vapor pressure buildup leading to bubbling. The final output is a joint modulation command matrix. f is the target operating frequency of the high-frequency power supply at the decision output, in Hertz; U is the target driving voltage of the plates at the decision output, in Volts; and V is the target vacuum level of the vacuum chamber at the decision output, in Pascals. The target control current of the hydraulic servo system is expressed in amperes, which is used to achieve global optimization of multiple parameters rather than local optimization at a single point.

[0046] The multi-branch collaborative decision-making module has built-in operating condition discrimination logic. When it is determined that the core layer inversion water content is more than 15% higher than that of the surface layer, the penetration enhancement modulation strategy is automatically triggered to lower the high-frequency power supply operating frequency and simultaneously increase the electrode driving voltage, thereby realizing abnormal handling of multiple parameters.

[0047] S4, the hierarchical electrical signal modulation module, acts as a bridge, translating digital instructions into physical actions. This invention breaks down the data silos between traditional industrial software and equipment control layers, achieving penetrating instruction distribution based on a five-layer industrial internet architecture.

[0048] Specifically, the joint modulation instruction matrix is ​​first parsed into device semantic instructions at the L3 operating layer, and then sent to the programmable logic controller at the L1 layer via Time-Sensitive Networking (TSN) and OPC UA protocol.

[0049] At the L1 level, the system discretizes continuous decision variables into specific electrical signal operation sequences: For high-frequency power supplies, the operating frequency of the high-frequency power supply is converted into a frequency control word that can be called by a digital signal processor, and the adjustment resolution is accurate to 0.1 MHz. For plate control, the pulse width modulation duty cycle of the insulated gate bipolar transistor is dynamically rewritten according to the target driving voltage of the plate. At the same time, the plate spacing is adjusted in 1 mm steps through the lead screw actuator to achieve fine matching of electric field energy density. For hydraulic and vacuum systems, the hydraulic servo current and decision vacuum level are mapped to the real-time current or frequency setpoint of the proportional valve and frequency converter.

[0050] During execution, the system collects the deviation between the actual feedback value of the actuator and the instruction setting value every 500 milliseconds. If the deviation is greater than the allowable tolerance, it triggers local PID fine-tuning and reports the fine-tuning record as a working condition disturbance event to the cloud, ensuring a high degree of consistency between digital instructions and physical actions.

[0051] S5, the cross-process closed-loop self-evolution module is used to extend the optimization of individual processes to the quality improvement of the entire process, and form a lifelong learning mechanism for the system.

[0052] The round-mouth inspection module (corresponding to the finished product inspection of wood) uses stress wave tomography and hyperspectral imaging to perform non-destructive re-inspection of the hot-pressed board and extract the actual quality vector including elastic modulus, bonding strength and internal defect distribution.

[0053] The parameter closed-loop adjustment module calculates the comprehensive reward value by utilizing the difference between the actual quality vector, including the internal defect distribution, and the expected quality. ,in To enhance the overall learning reward value, The weighting factor is the finished product quality factor, where Quality is a quantitative indicator of the overall finished product quality of the board material. The weighting factor for energy consumption is denoted as , where Energy is a quantitative indicator of energy consumption per unit sheet material in the hot pressing process. The reward value is the weighting coefficient for the defect penalty term, and "Defect" is a quantitative indicator of the risk of internal defects in the board material. Based on this reward value, the system uses a strategy gradient algorithm in the background to periodically fine-tune the network weights of the multi-branch collaborative decision-making module. More importantly, the system uses the environmental parameter requirements of the curing process as a feedforward signal, pushing it to the downstream L3 layer system at the moment the board material is depressurized, guiding the dynamic setting of the temperature and humidity slow-release curve of the curing chamber, thereby dissipating residual stress.

[0054] Ultimately, the updated reaction principle model and decision network are redeployed to the edge, forming a spiraling closed loop of production, learning, optimization, and reproduction.

[0055] In this embodiment, the integration of coaxial array dielectric sensing and thin-film implanted thermal measurement is realized. For the first time, millisecond-level transparent monitoring of the internal moisture field and temperature field of wood is achieved in a hot-pressed sealed mold, overcoming the blind defects of traditional methods that rely solely on experience or external temperature to infer the internal state.

[0056] By embedding Fick's law and the Arrhenius equation into a differentiable form in a neural network, a digital twin with clear physical meaning was constructed. This enabled the system to output control commands that conform to the basic laws of thermodynamics when faced with extreme non-standard raw materials such as fast-growing materials and high-moisture materials, thus avoiding the absurd decisions that are easily generated by pure data-driven models.

[0057] A multi-branch decision architecture that decouples and coordinates thermodynamics and mechanics was designed. By guiding the dynamic optimization of parameters through the mass potential gradient, the problem of regulation conflict caused by strong coupling of multiple physical fields such as high-frequency heating, vacuum extraction and hydraulic pressurization was solved, realizing a qualitative change from single-parameter feedback to multi-parameter joint self-consistency.

[0058] By relying on the industrial internet hierarchical architecture, data penetration and control loop between cloud, edge and terminal have been achieved, and the hot pressing process and subsequent curing process have been seamlessly connected, which has brought the consistency and reliability of the board bonding quality to a new level. It has responded to the pain points of insufficient core drying and surface charring that have long existed in the industry from the mechanism level.

[0059] Accordingly, such as Figure 2 As shown, this embodiment of the invention also provides an intelligent optimization method for wood deep processing parameters based on the Industrial Internet, applied to an intelligent optimization system for wood deep processing parameters based on the Industrial Internet of Things in this embodiment of the invention. The method includes: S10 synchronously acquires dielectric response signals, thermodynamic signals, and equipment-end electrical signals of wood blanks during hot pressing through multi-source heterogeneous sensing units, and performs spatiotemporal alignment and preprocessing on the multi-source data. S20 is a process twin engine that uses time-sensitive networking to upload pre-processed multi-dimensional data streams to the cloud industrial internet platform via an edge industrial gateway. S30 utilizes a process twin engine to construct a coupled reaction model of moisture migration and adhesive curing within the slab, and inverts the global reaction uniformity and curing degree indices in real time. S40, based on real-time inversion indicators, solves the joint modulation strategy of high-frequency power supply frequency, plate spacing, vacuum degree and hydraulic pressure through built-in multi-objective reinforcement learning algorithm; S50 breaks down the joint modulation strategy into execution instructions at different levels, and sends them down to the field equipment layer to complete the dynamic modulation of electrical signals and process execution. S60 collects the actual status feedback value of the slab after execution, performs online incremental updates to the cloud-based coupling model and algorithm weights, and outputs collaborative control parameters to upstream and downstream processes.

[0060] The principle and effects of the above technical solution are as follows: This invention provides an intelligent optimization method for wood deep processing parameters based on the Industrial Internet, which includes converting the hidden physicochemical changes in wood processing into high-frequency adjustable electrical signal commands, and realizing cross-level closed loop in a cloud-edge-device collaborative architecture. Since it no longer relies on fixed process curves, the system can autonomously compensate for the heterogeneous fluctuations of raw materials, thereby eliminating persistent problems such as bubbling, delamination, and insufficient core drying at the source, significantly improving the mechanical consistency and energy utilization efficiency of the finished product.

[0061] In one embodiment, S10 includes: S101, eight sets of coaxial array dielectric probes are arranged between the upper and lower plates of a high-frequency vacuum hot press. The excitation frequency of the probe center is set to 1 MHz. The relative permittivity and dielectric loss tangent of eight sections in the length direction of the slab are continuously acquired with a sampling period of 200 ms. The measurement error is controlled within 2%. S102, a thin-film platinum resistance thermometer and a miniature pressure thin-film sensor are respectively implanted near the surface and core layers in the thickness direction of the slab. The temperature sampling accuracy is set to 0.1 ℃ and the pressure sampling accuracy is set to 0.01 MPa, so as to simultaneously capture the evolution of the temperature field and pressure field during the hot pressing process.

[0062] The principle and effect of the above technical solution are as follows: the dipole polarization effect of water molecules is most significant at a frequency of 1 MHz, thus it can keenly reflect the microscopic distribution difference between free water and adsorbed water inside wood, providing high signal-to-noise ratio boundary condition input for digital models.

[0063] In one embodiment, S20 includes: S201 deploys an industrial gateway supporting a precision time protocol at the edge, timestamps dielectric and temperature data at the microsecond level, and uses a moving average filtering algorithm to remove abnormal jump points caused by electromagnetic interference. S202 publishes the encapsulated data object to the cloud data lake via the OPC UA protocol through a time-sensitive network with an end-to-end delay of no more than 5 ms, while caching the raw waveforms of the most recent 72 hours for model backtracking verification.

[0064] The principle and effect of the above technical solution are as follows: the microsecond-level timestamp ensures the spatiotemporal consistency of multi-physical quantity sampling, enabling the subsequently constructed three-dimensional digital twin to truly restore the spatiotemporal reaction process inside the slab, avoiding control lag caused by asynchronous data.

[0065] In one embodiment, S30 includes: S301, based on Fick's second law describing the law of moisture migration and Darcy's law reflecting heat and mass transfer, constructs a partial differential equation for the moisture content gradient field inside the slab, and solves the boundary conditions of the equation in real time by combining the dielectric constant inversion results. S302 introduces the Arrhenius equation, which describes the crosslinking kinetics of adhesives, and uses the internal temperature field of the slab as input to calculate the curing degree index of each zone at different hot pressing times. The index is calibrated to the range of 0% to 100%. S303 spatially superimposes the moisture content gradient field and the solidification index to output a global reaction uniformity thermogram, and extracts the core layer hysteresis coefficient and the surface layer overheating risk value as the state input of the reinforcement learning agent.

[0066] The principle and effect of the above technical solution are as follows: by using a hybrid modeling approach that combines physical equations and data-driven methods, the shortcomings of simple data models, such as lack of mechanistic support and poor generalization, are overcome, enabling the system to maintain accurate reasoning ability when faced with new tree species or specifications.

[0067] In one embodiment, S40 includes: S401 defines the state space, which includes the core layer moisture content, dielectric loss change rate, and curing index, and defines the action space, which includes the high-frequency power supply frequency, electrode drive voltage, vacuum pump frequency, and servo valve current. S402 sets the reward function with zero defects after the slab exits the press as the primary goal, assigning a weight of 60% to quality, 25% to energy consumption, and 15% to production efficiency. S403 employs a dual-delay deep deterministic strategy gradient algorithm for online solution, outputting the optimal joint modulation matrix once every 1 second.

[0068] In one embodiment, S50 includes: S501 parses the joint modulation matrix sent from the cloud into a set of device control commands at the L2 monitoring layer, where the high-frequency power supply operating frequency is limited to between 3 and 8.5 MHz, with a resolution of 0.1 MHz; S502 transmits instruction sets through industrial Ethernet to the PLC in the L1 layer and the high-frequency inverter unit in the L0 layer, and dynamically adjusts the electrode spacing in 1 mm increments within the range of 50~150 mm. S503 synchronously limits the output frequency of the vacuum pump inverter to the range of 20~50 Hz, and dynamically corrects the control current of the proportional valve of the hydraulic system according to the modulation matrix, with a current adjustment step of 0.01 A.

[0069] In one embodiment, S60 includes: S601, after hot pressing, uses a stress wave non-destructive testing instrument at a sampling rate of 10 kHz to collect the internal wave velocity distribution of the finished product, back-calculate the residual stress level, and transmit the deviation back to the cloud in real time; S602: When the measured residual stress is higher than the preset threshold of 50 με, the model fine-tuning is automatically triggered, and the diffusion coefficient weight in the digital twin engine is updated using the small-batch gradient descent method. S603 packages and sends the optimized temperature and humidity parameters of the health room to the L3 operations layer to guide the subsequent process connection for at least 24 hours.

[0070] In one specific embodiment, in S40, if the inverted water content of the core layer is more than 15% higher than that of the surface layer, a penetration-enhanced modulation strategy is adopted, and the specific steps are as follows: SA41: The high-frequency power supply operating frequency is forcibly reduced to 3.2 MHz to enhance the penetration depth of electromagnetic waves into the high-humidity core layer.

[0071] SA42: While keeping the surface temperature below 110 ℃, increase the plate drive voltage by 15% of the rated value to increase the volume heating power.

[0072] SA43: Simultaneously reduce the vacuum level by 10% to appropriately slow down the evaporation rate of surface moisture, forcing heat to be conducted to the core layer.

[0073] SA44: Continuously monitor the convergence trend of the dielectric loss tangent value, and resume normal control when the difference in water content between the core layer and the surface layer is reduced to within 5%.

[0074] The penetration-enhanced modulation strategy effectively solves the problem of uneven curing caused by slow heat transfer in the core layer of thick plates, enabling thick plates that are difficult to press using conventional processes to be formed in one step.

[0075] In step S40, if the curing index of the adhesive is still below 30% after holding the pressure for 5 minutes, a temperature-increase and pressure-boosting modulation strategy is adopted. The specific steps are as follows: SB41: Switch the high-frequency excitation to pulse intermittent mode and set the duty cycle to 70% to suppress the artifact of adhesive pre-curing caused by local overheating.

[0076] SB42: The hydraulic pressure is increased by 0.15 MPa by a servo valve to promote the penetration and wetting of the adhesive in the wood conduits.

[0077] SB43: Based on the temperature sensitivity of the Arrhenius equation, the overall target temperature of the slab is increased by 5 °C to activate the curing agent.

[0078] SB44: Extend the predicted time for the pressure holding phase and feed it back to the L3 production scheduling system to recalculate the cycle time.

[0079] The temperature-increasing and pressure-boosting modulation strategy significantly shortens the process time required for cold bonding and improves production line efficiency through the synergy of physical pressure and chemical activation.

[0080] In S40, if the dielectric loss tangent exhibits abnormal oscillations with an amplitude greater than 0.05, an intermittent homogeneous modulation strategy is adopted. The specific steps are as follows: SC41: Immediately stop the high-frequency power supply output and maintain the vacuum pump at full speed for 3 seconds to expel the free water vapor accumulated inside the slab.

[0081] SC42: After power is restored, run in low-energy mode at 40% of rated power for 10 pulse cycles and observe the stability of the dielectric signal.

[0082] SC43: Gradually increase the power in steps until the dielectric oscillation amplitude drops below 0.02, and record the unique dielectric characteristic code of this batch of wood.

[0083] SC44: Store the feature code in the feature fingerprint library on the edge side, so that subsequent slabs in the same batch can directly call the initial parameters.

[0084] The intermittent homogeneous modulation strategy can intelligently identify and resolve abnormal discharges caused by internal cavities in the wood or resin accumulation, ensuring the safety of equipment operation and the stability of the process.

[0085] In one specific embodiment, in S50, if the temperature rise rate of the slab core layer exceeds 0.5 ℃ / s within 30 seconds after the command is executed, a gentle slope and current limiting execution mode is adopted, and the specific steps are as follows: SA51: Immediately limits the maximum output current of the high-frequency inverter unit to no more than 1 / 4 of the rated value.

[0086] SA52: Increase the electrode spacing to more than 120 mm to dilute the electromagnetic energy density per unit volume.

[0087] SA53: Activate the built-in circulating cooling water valve of the hot press plate to increase the cooling water flow rate to 15 L / min.

[0088] SA54: After the core layer temperature rise rate returns to the safe range below 0.2 ℃ / s, the preset modulation matrix is ​​gradually restored.

[0089] The gentle slope flow restriction mode prevents wood pyrolysis and carbonization caused by local energy accumulation, thus providing dual protection for both equipment and process.

[0090] In S50, if the deviation between the vacuum feedback value and the set value continues to be greater than 2 kPa, a fault-tolerant redundant execution mode is adopted. The specific steps are as follows: SB51: Automatically identifies vacuum pipeline leaks or vacuum pump performance degradation, issues local audible and visual warnings without interrupting the current hot pressing cycle.

[0091] SB52: Proportional compensation for high-frequency power supply output power, with a compensation coefficient of 0.5% / min of the deviation value.

[0092] SB53: Temporarily call up the backup drying kiln data and mark this batch of slabs as objects requiring enhanced curing.

[0093] SB54: After hot pressing is completed, a vacuum system maintenance work order is automatically generated and pushed to the equipment management terminal.

[0094] The fault-tolerant redundancy execution mode enables smooth degradation of faulty equipment without sacrificing shift productivity, demonstrating the superiority of software and hardware decoupling under the industrial internet architecture.

[0095] In S50, if the standard deviation of the pressure in each zone of the multi-cylinder hot press is greater than 0.05 MPa, then the differential pressure equalization execution mode is adopted, and the specific steps are as follows: SC51: Locks the specific zone numbers for high and low pressure, and generates a zone pressure regulation matrix.

[0096] SC52: Issues a negative bias current command to the high-pressure zone servo valve, with a step size of 0.02 A.

[0097] SC53: Synchronously issues positive current bias commands to low-voltage zones to ensure that the total voltage remains constant and is not affected by zone adjustments.

[0098] SC54: Iterate the above bias process with a period of 500 ms until the standard deviation of the pressure in the entire region converges to within 0.01 MPa.

[0099] The differential pressure equalization execution mode eliminates the edge effect commonly found in large hot press units, ensuring that the bonding strength of large-format engineered wood panels is highly consistent throughout.

[0100] In one specific embodiment, in S60, if the residual stress detected by the stress wave is higher than 80 με, a deep introspection update mode is adopted, and the specific steps are as follows: SA61: Freeze the water diffusion coefficient in the current digital twin model and prohibit it from participating in subsequent training to avoid error propagation.

[0101] SA62: Extract the full-cycle dielectric and temperature sequence of the abnormal slab, label it as a high-value negative sample, and store it in the cloud training pool.

[0102] SA63: It uses an adversarial generative network to expand the training set for similar abnormal conditions and fine-tunes the reinforcement learning policy network with a learning rate of 5%.

[0103] SA64: After fine-tuning, perform at least 1000 policy replay verification steps in a virtual simulation environment. Only after confirming that there is no performance degradation can the policy be relaunched.

[0104] The deep introspective update mode gives the system the ability to learn and evolve autonomously from occasional defects. As the running time increases, the system's process insight will continuously approach the theoretical limit.

[0105] In S60, if the finished product quality indicators of the 10 consecutive slabs are all better than the historical best values, then the knowledge distillation sinking mode is adopted, and the specific steps are as follows: SB61: The decision logic of the large cloud-based reinforcement learning model under this condition is compressed into a lightweight decision tree model through knowledge distillation.

[0106] SB62: Lightweight models are remotely burned into the local inference chip of the edge industrial gateway via OTA.

[0107] SB63: When there is jitter or disconnection in the cloud link, the edge side can seamlessly take over the control tasks and achieve offline autonomous operation.

[0108] SB64: After the link is restored, the offline operation data will be encrypted and transmitted back to complete the cloud-edge status synchronization.

[0109] The knowledge distillation and sinking mode breaks down the last mile of cloud-edge collaboration, preserving the wisdom of the large cloud model while giving individual devices strong on-site survivability and response capabilities.

[0110] It should be noted that the intelligent optimization system for wood deep processing parameters based on the Industrial Internet provided in this embodiment of the invention is used to execute all process steps of the intelligent optimization method for wood deep processing parameters based on the Industrial Internet in the above embodiment. The control principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0111] This invention also provides an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as an intelligent optimization program for wood deep processing parameters based on the Industrial Internet. When the processor executes the computer program, it implements the steps in each of the above embodiments of an intelligent optimization method for wood deep processing parameters based on the Industrial Internet. Alternatively, when the processor executes the computer program, it implements the functions of each module or unit in the above system embodiments.

[0112] For example, the computer program may be divided into one or more modules or units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules or units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0113] The electronic device may be a vehicle-mounted controller deployed on the robot body, an industrial field edge computing gateway, or a remote cloud platform, etc. The electronic device may include, but is not limited to, processors and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include a data acquisition unit for connecting a multi-antenna array, a radio frequency transceiver assembly, an inertial measurement module, network communication equipment, a control bus, etc.

[0114] The processor can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (OPGs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting various parts of the entire electronic device via various interfaces and lines.

[0115] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store topology maps, envelope reference parameters, state machine configuration tables, etc., created according to system usage. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart memory cards, secure digital cards, flash memory cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0116] Wherein, if the modules or units integrated into the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory, random access memory, electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0117] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An intelligent optimization system for wood deep processing parameters based on the Industrial Internet, characterized in that, The system includes an online sensing module for multiple dielectric and thermal parameters, a reaction principle-driven digital twin module, a multi-branch collaborative decision-making module, a hierarchical electrical signal modulation module, and a cross-process closed-loop self-evolution module; The dielectric thermal multi-parameter online sensing module is used to perform multi-dimensional in-situ measurements of slabs under high-frequency vacuum hot pressing environment with sub-second time resolution, and convert the physical response into a standardized wood state feature vector. The reaction principle-driven digital twin module is used in a hybrid modeling architecture based on physical priors to receive feature vectors and output a set of reaction process indicators for the entire slab in real time. The multi-branch collaborative decision-making module is used to generate a joint modulation instruction matrix based on a set of reaction process indicators through a hierarchical multi-agent deep reinforcement learning framework. The hierarchical electrical signal modulation module is used to decompose the joint modulation instruction matrix into execution instructions of different levels and issue them based on the hierarchical architecture of the Industrial Internet, so as to complete the dynamic modulation and process execution of electrical signals. The cross-process closed-loop self-evolution module is used to collect the actual state feedback value of the slab, update the weights of the coupled reaction model and decision algorithm online, and output collaborative control parameters to upstream and downstream processes.

2. The system as described in claim 1, characterized in that, The online sensing module for multiple dielectric and thermal parameters includes a flexible implantable sensing unit, an edge signal conditioning unit, and a high-precision synchronous acquisition card. The flexible implantable sensing unit is arranged in a matrix along the board surface, specifically including a coaxial array dielectric probe, a thin film platinum resistance thermometer, and a miniature pressure-sensitive film; the coaxial array dielectric probe operates at a 1 MHz excitation frequency and is sensitive to the polarization relaxation of free water and adsorbed water in the wood; the thin film sensor is attached to the gap between the veneers in a non-invasive manner. The edge signal conditioning unit uses lock-in amplification technology to extract the imaginary part of the dielectric signal and suppress common-mode interference caused by high-frequency electromagnetic fields. The synchronous acquisition card supports hardware triggering to ensure that the sampling points of all channels are strictly synchronized with the power frequency cycle of the high-frequency power supply, thus ensuring strict alignment of multiple physical quantities in the spatiotemporal dimensions.

3. The system as described in claim 1, characterized in that, The dielectric thermal multi-parameter online sensing module specifically scans the slab in sections using a coaxial array probe to obtain the relative permittivity and dielectric loss tangent of each sub-region. It then obtains the initial moisture content distribution field through dielectric hybridization model inversion. Based on multi-point temperature and pressure sequences, it uses the moving least squares method to fit the instantaneous temperature gradient field and pressure distribution along the thickness direction of the slab. After the collected physical data is processed by moving average filtering and normalization, it is aggregated to generate a standardized wood state feature vector containing dielectric constant, dielectric loss tangent, thickness temperature gradient, core layer measured pressure, high-frequency power supply operating frequency setting, and electrode operating voltage setting.

4. The system as described in claim 1, characterized in that, The reaction principle-driven digital twin module adopts a physical information neural network architecture, incorporating Fick's second law describing water migration and Arrhenius's equation describing adhesive crosslinking as soft constraints into the network loss function; The physical information neural network architecture includes a shared encoding network, a water migration physical branch, a solidification dynamics physical branch, and a physical constraint decoding layer; Shared coding networks map feature vectors to a high-dimensional latent space; The physical branch of water migration outputs the rate of water content decay in each zone and uses the residual of the unsteady water diffusion equation as the physical loss term by automatic differentiation. The curing kinetics branch predicts activation energy and reaction rate constant based on latent space characteristics, and calculates real-time degree of curing by combining Arrhenius relation; The physical constraint decoding layer integrates the results of the two branches and outputs a set of reaction process indicators, including the predicted core layer moisture content, the adhesive layer curing degree index, and the residual stress risk factor, while simultaneously calculating the uniformity thermogram of the entire slab.

5. The system as described in claim 1, characterized in that, The multi-branch collaborative decision-making module includes a state assessment branch, a thermodynamic modulation branch, a mechanical modulation branch, and a collaborative arbitration network; The state assessment branch receives the reaction process index set in real time, quantifies the deviation vector of the slab from the ideal solidification state, and outputs the quality potential gradient. The thermodynamic modulation branch input includes a deviation vector and dielectric loss feedback, and integrates an attention-enhanced recurrent neural network. It dynamically outputs the operating frequency of the high-frequency power supply and the plate driving voltage. The network end is connected to a micro thermal balance solver to verify the theoretical temperature rise rate. The mechanical modulation branch determines the vacuum degree and hydraulic servo current based on the progress of curing degree and quality potential gradient, and uses a conditional residual network to treat wood species and veneer thickness as dynamic parameters. The collaborative arbitration network calculates the coupling matching degree between the thermodynamic and mechanical branch outputs and outputs a joint modulation command matrix containing the target operating frequency of the high-frequency power supply, the target driving voltage of the electrode plate, the target vacuum degree, and the target control current of the hydraulic servo system.

6. The system as described in claim 5, characterized in that, The multi-branch collaborative decision-making module has built-in operating condition discrimination logic. When it is determined that the core layer inversion water content is more than 15 percent higher than the surface layer, the penetration enhancement modulation strategy is automatically triggered to lower the high-frequency power supply operating frequency and simultaneously increase the electrode driving voltage, so as to realize the abnormal handling of multi-parameter linkage. In the multi-branch collaborative decision-making module, if the adhesive curing degree index is still lower than the preset threshold after the pressure holding time is preset, a temperature-increasing and pressure-boosting modulation strategy is adopted to switch the high-frequency excitation to the pulse intermittent mode, increase the hydraulic pressure, and adjust the overall target temperature of the slab according to the temperature sensitivity to activate the curing agent activity. If the dielectric loss tangent value exhibits abnormal amplitude oscillation, an intermittent homogeneous modulation strategy is adopted. The high-frequency power output is paused and the vacuum pump is kept running at full speed to expel the internally accumulated free water vapor. Then, the power supply is restored in a low-energy mode and the power is increased stepwise until the dielectric signal is stable. The unique dielectric feature code is recorded and stored in the edge-side feature fingerprint database.

7. The system as described in claim 1, characterized in that, The hierarchical electrical signal modulation module realizes instruction penetration and distribution based on the five-layer industrial internet architecture. The joint modulation instruction matrix is ​​parsed into device semantic instructions at the operation layer, and then distributed to the programmable logic controller via time-sensitive network and OPC UA protocol. For high-frequency power supplies, the operating frequency is converted into a frequency control word, and the adjustment resolution is accurate to 0.1 MHz. For plate control, the pulse width modulation duty cycle of the insulated gate bipolar transistor is dynamically rewritten, and the plate spacing is adjusted in 1 mm increments via a lead screw actuator. For hydraulic and vacuum systems, the servo current and vacuum level are mapped to the real-time current or frequency setpoint of the proportional valve and frequency converter. During execution, the deviation between the actual feedback value and the instruction setting value is collected every 500 milliseconds. If the deviation is greater than the allowable tolerance, local PID fine-tuning is triggered, and the fine-tuning record is reported as a working condition disturbance event.

8. The system as described in claim 7, characterized in that, If the temperature rise rate of the core layer of the slab exceeds the safety threshold within a unit time during the execution of instructions, the hierarchical electrical signal modulation module adopts a gentle slope current limiting execution mode to limit the maximum output current of the high-frequency inverter unit, increase the electrode spacing to dilute the electromagnetic energy density, and activate the circulating cooling water valve built into the hot press plate for cooling protection. If the deviation between the vacuum feedback value and the set value continues to exceed the preset deviation value, the fault-tolerant redundant execution mode will be adopted to automatically identify and issue an early warning, compensate the high-frequency power supply output power proportionally, mark the batch of slabs as objects to be strengthened and cured, and automatically generate a maintenance work order after hot pressing is completed. If the pressure standard deviation of each zone of the multi-cylinder hot press is greater than the preset value, the pressure difference equalization execution mode is adopted to generate a zone pressure regulation matrix. Current bias commands are issued to the high-pressure zone and the low-pressure zone respectively to ensure that the total pressure is constant until the pressure standard deviation of the entire zone converges to the allowable range.

9. The system as described in claim 1, characterized in that, The cross-process closed-loop self-evolution module includes a circular port detection module and a parameter closed-loop adjustment module; The circular opening detection module extracts the actual mass vector containing elastic modulus, bonding strength and internal defect distribution through stress wave tomography and hyperspectral imaging. The parameter closed-loop adjustment module calculates the reinforcement learning comprehensive reward value by using the difference between the actual quality vector and the expected quality. The reward value is composed of a weighted sum of the finished product quality item, energy consumption item, and defect penalty item. Based on the reward value, the system uses a strategy gradient algorithm in the background to periodically fine-tune the network weights of the multi-branch collaborative decision-making module. At the same time, it pushes the environmental parameter requirements of the health care process as a feedforward signal to the downstream system to guide the dynamic setting of the temperature and humidity slow-release curve of the health care room. If the residual stress detected by the stress wave is higher than the preset threshold, the cross-process closed-loop self-evolution module adopts a deep self-reflection update mode, freezes the moisture diffusion coefficient in the digital twin model, extracts the full-cycle sequence of abnormal slabs and labels it as high-value negative samples, expands the training set using an adversarial generative network and fine-tunes the strategy network, and is re-launched after verification.

10. The intelligent optimization system for wood deep processing parameters based on the Industrial Internet according to claim 1, wherein the system is configured to execute an intelligent optimization method for wood deep processing parameters, characterized in that, The methods include: S10 synchronously acquires dielectric response signals, thermodynamic signals, and equipment-end electrical signals of wood blanks during hot pressing through multi-source heterogeneous sensing units, and performs spatiotemporal alignment and preprocessing on the multi-source data. S20 is a process twin engine that uses time-sensitive networking to upload pre-processed multi-dimensional data streams to the cloud industrial internet platform via an edge industrial gateway. S30 utilizes a process twin engine to construct a coupled reaction model of moisture migration and adhesive curing within the slab, and inverts the global reaction uniformity and curing degree indices in real time. S40, based on real-time inversion indicators, solves the joint modulation strategy of high-frequency power supply frequency, plate spacing, vacuum degree and hydraulic pressure through built-in multi-objective reinforcement learning algorithm; S50 breaks down the joint modulation strategy into execution instructions at different levels, and sends them down to the field equipment layer to complete the dynamic modulation of electrical signals and process execution. S60 collects the actual status feedback value of the slab after execution, performs online incremental updates to the cloud-based coupling model and algorithm weights, and outputs collaborative control parameters to upstream and downstream processes.

Images