Baking and humidifying integrated device and method thereof

Abstract

The invention discloses a baking and humidity equalizing integrated device and a method thereof, and relates to the technical field of agricultural product processing, the device comprises a shell, a baking cavity, a bearing assembly, a backflow cavity, a plurality of groups of micro-vortex airflow rings, a plurality of groups of phase change metal heat conduction substrates and a central control and learning assembly, the phase-change metal heat-conducting substrate sequentially comprises a heat-conducting plate, a phase-change layer and a heating layer and is fixedly connected with the corresponding micro-vortex airflow rings, the micro-vortex airflow rings jet vortex airflow through inclined micropores and independently adjust the airflow, and an inner circulating fan and a pressure detection unit are arranged in a backflow cavity to maintain constant-pressure air supply in a closed-loop mode; the multi-mode sensing assembly collects photoacoustic and acoustic observation data, the state reconstruction unit executes analysis domain alternate iteration reconstruction to output a space state diagram, the diagram topology binding unit constructs a state diagram, the strategy generation unit outputs a power and opening degree gear sequence, and the state diagram is stored in the storage unit. And the empirical memory unit indexes the multiplexing track and optimizes the control sequence under reward driving and constrained updating, so as to realize uniform dehydration and brittleness control.

Description

Technical Field

[0001] This invention relates to the field of agricultural product processing technology, specifically to an integrated baking and wetting device and method. Background Technology

[0002] Air fryers, hot air ovens, and drying equipment typically use hot air circulation combined with electric heating to achieve dehydration and cooking. Some devices incorporate temperature sensors, cavity humidity sensors, or preset program curves to control heating power and fan speed, thereby completing the baking process. Existing technologies also include solutions that automatically select cooking programs after identifying the type of food using a camera, as well as solutions that use closed-loop or semi-closed-loop control based on parameters such as weight, time, and temperature.

[0003] While the aforementioned existing technologies can achieve basic heating and drying control, they still have significant limitations: First, the controlled objects are mostly concentrated on macroscopic quantities such as cavity air temperature, time, or cavity humidity, making it difficult to reflect the internal moisture migration and changes in tissue structure of the food. This leads to problems such as excessively dry surface with internal dampness, localized charring and uncooked areas coexisting, and uneven crispness during baking. Second, the airflow supply is usually directly driven by a fan, lacking a measurable and maintainable air supply pressure boundary. This causes fluctuations in the jet and circulating airflow due to factors such as loading volume, cavity leakage, and oil mist blockage, resulting in variations between different batches. The first problem is the lack of consistency between different types of airflow. The second problem is the lack of consistency between different types of airflow. The third problem is that traditional jets are mostly direct or simple guides, which make it difficult to form a stable and controllable local rotating flow field near the bearing surface. The distribution of airflow and heat exchange conditions in space are difficult to adjust in a zoned manner. The fourth problem is that common heating structures are mostly single heating sources or simplified heat conduction paths. The spatial distribution and dynamic response of heat input are difficult to control precisely, and hot spots are easily generated. The fifth problem is that existing intelligent control relies on empirical curves or single model predictions. There is a lack of a mechanism to transform multimodal observation data into spatial state and use it for closed-loop decision-making. The control strategy is difficult to adapt stably to the differences in ingredients and real-time state changes.

[0004] Existing technologies achieve baking by adjusting heating power and fan speed, but they still have certain limitations. It is difficult to simultaneously meet the requirements of uniform moisture content, consistent crispness and process stability. Furthermore, the repeatability of control is insufficient when loading conditions change, and local over-baking or re-moistening is likely to occur.

[0005] Therefore, there is an urgent need for a baking and humidification integrated device and method that has constant pressure reflux gas supply conditions, can form vortex airflow in sections and cooperate with phase change heat conduction structure, and can reconstruct the spatial state of water content and brittleness based on multimodal observation and output adaptive control sequence, so as to achieve uniform dehydration and brittleness control in the baking process and improve batch consistency. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and to propose an integrated baking and moistening device and method to solve the above-mentioned problems.

[0007] The objective of this invention is achieved through the following technical solution: a baking and humidification integrated device, comprising a shell, a baking cavity disposed within the shell, a support component disposed within the baking cavity, a reflux cavity disposed between the baking cavity and the shell, multiple micro-vortex airflow rings communicating with the reflux cavity and uniformly distributed within the baking cavity, and a phase change metal heat-conducting substrate corresponding to each micro-vortex airflow ring, and a central control and learning component; each phase change metal heat-conducting substrate includes a heat-conducting plate, a phase change layer disposed on the heat-conducting plate, and a heating layer disposed on the phase change layer, with the outer end of each phase change metal heat-conducting substrate connected to the corresponding micro-vortex airflow ring. The flow ring is fixedly connected; multiple inclined micro-holes are opened on the end of the micro-vortex airflow ring near the supporting component, and each micro-vortex airflow ring is equipped with an independent airflow adjustment component; an internal circulation fan and a pressure detection unit are installed in the return cavity. The internal circulation fan is electrically connected to the central control and learning component and is adjusted by the central control and learning component under the pressure feedback output by the pressure detection unit to keep the pressure in the return cavity within a preset pressure range; the heating layer and the independent airflow adjustment component are both electrically connected to the central control and learning component; a multimodal sensing component is installed in the baking cavity, and the multimodal sensing component is electrically connected to the central control and learning component. The central control and learning component includes a state reconstruction unit, a graph topology binding unit, a policy generation unit, a policy update constraint unit, and an experience memory unit. The state reconstruction unit stores observation mapping parameters and is configured to generate a set of voxels corresponding to the distribution of the phase change metal thermally conductive substrate based on the observation data output by the multimodal sensing component and the observation mapping parameters, and to generate voxel-level water-bearing state estimates and voxel-level brittleness state estimates. The state reconstruction unit is configured to execute an alternating iterative reconstruction process in the analysis domain within each control cycle to output a spatial state diagram. The alternating iterative reconstruction process includes prior updates to the gradient representation of the voxel-level water-bearing state estimates and voxel-level brittleness state estimates based on the spatial discrete gradient operator, and data consistency updates to the voxel-level water-bearing state estimates and voxel-level brittleness state estimates based on the observation mapping parameters, with the prior updates and data consistency updates executed alternately. The graph topology binding unit is configured to use each phase change metal... The thermally conductive substrate and its fixedly connected micro-vortex airflow ring serve as graph nodes, and the adjacent arrangement relationship between phase change metal thermally conductive substrates and the recirculation coupling relationship formed by the micro-vortex airflow ring through the recirculation cavity serve as graph edges to generate a state diagram. The strategy generation unit is configured to output the power level sequence of the heating layer and the opening level sequence of the independent airflow adjustment component based on the state diagram to form a control sequence. The experience memory unit is configured to establish an index based on material type parameters, loading state parameters, initial moisture content state parameters, and initial brittleness state parameters, and store the state diagram sequence, power level sequence, opening level sequence, and evaluation data corresponding to the index. The strategy update constraint unit is configured to generate a reward amount based on the evaluation data of adjacent control cycles, and when updating the generation rules of the control sequence, restrict the ratio of the selection probability of the updated strategy for the existing control sequence to the selection probability of the previous strategy for the existing control sequence to a probability ratio range defined by the lower limit threshold and the upper limit threshold.

[0008] The reflux chamber is arranged around the baking chamber, and there are connecting channels between the reflux chamber and each micro-vortex airflow ring. The internal circulation fan is set in the circulation channel of the reflux chamber, and the reflux chamber and the circulation channel form a closed internal circulation loop.

[0009] The independent airflow regulation component includes a throttling valve and a drive actuator disposed within a micro-vortex airflow ring. The drive actuator is electrically connected to the central control and learning component, and the central control and learning component controls the opening degree of each throttling valve.

[0010] The inclined micro-holes are distributed circumferentially along the micro-vortex airflow ring, and the axis of the inclined micro-holes is inclined relative to the radial direction of the micro-vortex airflow ring, forming an outlet end on the side end face near the support component.

[0011] The multimodal sensing component includes a photoacoustic moisture content detection submodule and a structural acoustic fragility detection submodule. The photoacoustic moisture content detection submodule includes a pulsed near-infrared light source, an optical window located on the top of the baking cavity, and a photoacoustic pickup array. The structural acoustic fragility detection submodule includes an acoustic excitation unit and an acoustic receiving unit, which are located on the cavity wall of the baking cavity or on a phase change metal thermally conductive substrate.

[0012] The spatial discrete gradient operator is generated by the adjacency relationship of the voxel set, and the adjacency relationship of the voxel set is determined by the arrangement position of the phase change metal thermally conductive substrate and the arrangement position of the micro vortex airflow ring. The prior update performs denoising update on the gradient representation of the voxel-level water-bearing state estimate and the gradient representation of the voxel-level brittle state estimate respectively, and outputs the gradient target for data consistency update.

[0013] The node characteristics of the graph node include the amount of water content polymerization, the amount of brittle state polymerization, the amount of temperature polymerization, the amount of water content change, the amount of brittle state change, the power level of the previous control cycle, and the opening level of the previous control cycle for the voxel set corresponding to the phase change metal thermal conductive substrate.

[0014] The evaluation data includes moisture content deviation index, brittleness deviation index, moisture content uniformity index, brittleness uniformity index, energy consumption index, and time penalty index, and the reward amount is determined by the difference in evaluation data between adjacent control cycles.

[0015] Before generating the control sequence, the experience memory unit retrieves the power level sequence and the opening level sequence based on the index, and the strategy generation unit generates the power level sequence and the opening level sequence based on the state diagram and in combination with the retrieved power level sequence and the retrieved opening level sequence.

[0016] The central control and learning component executes a baking humidity control method, which includes steps S1 to S2: Step S: Drive the pulsed near-infrared light source and acquire the photoacoustic observation data output by the photoacoustic pickup array within the control cycle, and drive the acoustic excitation unit and acquire the acoustic observation data output by the acoustic receiving unit. Step S: The state reconstruction unit establishes a voxel set based on the observation mapping parameters, and generates voxel-level water content state estimates and voxel-level brittleness state estimates based on photoacoustic observation data and acoustic observation data. Step S: The spatial discrete gradient operator is established by the state reconstruction unit and the analysis domain alternating iterative reconstruction process is executed to output the spatial state diagram; Step S: The graph topology binding unit uses the phase change metal thermal conductive substrate and its fixedly connected micro vortex airflow ring as graph nodes, and uses the adjacent arrangement relationship and the backflow coupling relationship as graph edges to generate a state graph and output the state vector. Step S: The experience memory unit retrieves the power level sequence and the opening level sequence based on the index retrieval. The strategy generation unit outputs the power level sequence and the opening level sequence based on the state diagram and in combination with the power level sequence and the opening level sequence. It then drives each heating layer and each independent airflow adjustment component to operate according to the control sequence. The internal circulation fan is also driven to adjust under the pressure feedback output by the pressure detection unit to keep the pressure in the return chamber within the preset pressure range. Step S involves the policy update constraint unit generating a reward amount based on the evaluation data of adjacent control cycles, and limiting the ratio of the selected probability of the updated policy to the selected probability of the unupdated policy to the selected probability of the existing control sequence to a probability ratio range when updating the generation rules of the control sequence. The experience memory unit stores the state diagram sequence, power level sequence, opening level sequence and evaluation data.

[0017] The beneficial effects of this invention are: A reflux chamber is set up in the cavity structure to form a closed internal circulation loop. A pressure detection unit and an internal circulation fan are arranged in the reflux chamber and pressure feedback regulation is adopted to keep the pressure in the reflux chamber stable within the preset pressure range. This transforms the supply capacity of the air source into a measurable and calibrable physical boundary condition, avoiding insufficient air supply pressure difference or excessive fluctuation under different throttling openings of the micro vortex airflow ring. This ensures that the flow regulation of each micro vortex airflow ring has consistency and repeatability, and improves the calibrability and operational stability of the device under mass production conditions.

[0018] The phase change metal thermally conductive substrate and the micro-vortex airflow ring are fixedly connected and evenly distributed in the baking cavity, so that the heat conduction path and the jet path form a one-to-one corresponding control unit in space. This allows the heat input and convection organization to be coordinated and regulated in the same spatial unit, avoiding the problems of local overheating, local damping or inconsistent dryness and wetness between the surface and the interior caused by the separation of heat source and airflow source in traditional devices. It also provides a clear structural mapping basis for subsequent zoning state estimation and zoning strategy output.

[0019] The phase change metal thermal conductive substrate adopts a layered structure of heat-conducting plate, phase change layer and heating layer. The energy input of the heating layer first enters the phase change layer and is then transferred to the baking cavity through the heat-conducting plate. Structurally, it forms a buffer and shaping capability for heat flux, reduces the thermal shock caused by the power adjustment of the heating layer, reduces the instability caused by temperature overshoot such as rapid surface hardening and obstruction of internal water migration, and improves the controllability and predictability of the thermal field during the baking process.

[0020] Multiple inclined micro-holes are opened on the end face of the micro-vortex airflow ring near the support component. Through the geometric orientation of the micro-hole axis being relatively radially inclined, the jet flow has a tangential component and forms a stable rotating flow structure locally. Compared with the direct jet flow, the rotating flow is more likely to form uniform shear and heat transfer conditions in the near-surface layer, reducing the local strong scouring and hot spot concentration caused by direct jet, thereby improving the convection consistency of different regions on the same support plane and creating boundary conditions for achieving uniform dehydration and uniform embrittlement.

[0021] By introducing a multimodal sensing combination of a photoacoustic moisture content detection submodule and a structural acoustic brittleness detection submodule, the device can simultaneously acquire photoacoustic observation data related to moisture distribution and acoustic observation data related to changes in tissue structure. This avoids the control blind spots caused by using only temperature, time, or cavity humidity as indirect indicators, thus enabling the control system to have observable information sources directly facing the moisture content and brittleness states, and improving its adaptability to process differences under different materials and loading conditions.

[0022] The state reconstruction unit adopts an alternating iterative reconstruction process of the analysis domain, alternately executing prior updates and data consistency updates. This ensures that the reconstruction results satisfy the constraints of the observation data while maintaining the structural boundary characteristics and suppressing non-real fluctuations in the spatial discrete gradient representation, thereby obtaining a spatial state map corresponding to the distribution of the phase change metal thermally conductive substrate. This spatial state map is a unified expression of the voxel-level water-bearing state estimate and the voxel-level brittleness state estimate, which can provide a stable, continuous data foundation for calculating differential indices for zonal control and reduce control jitter caused by sensor noise or local anomalies.

[0023] The adjacency relationship of the spatial discrete gradient operator is determined by the arrangement of the phase change metal thermal conductive substrate and the arrangement of the micro vortex airflow ring. This ensures that the spatial coupling relationship in the reconstruction algorithm is consistent with the coupling relationship of the device structure. This avoids the state deviation caused by the inconsistency between the adjacency relationship assumed by the algorithm and the actual heat flow airflow coupling relationship from the source. It also ensures that the state estimation is isomorphic to the topology of the actuator, and enhances the engineering attributes of the algorithm at the device level, making it feasible and reproducible.

[0024] The graph topology binding unit uses a phase change metal thermally conductive substrate and its fixedly connected micro-vortex airflow ring as graph nodes, and generates a state diagram using adjacent arrangement relationships and recirculation coupling relationships as graph edges. This ensures a strict correspondence between the input structure of the control strategy and the actuator structure of the device. The power level sequence and opening level sequence output by the strategy generation unit can be directly mapped to the heating layer and independent airflow adjustment components, realizing an end-to-end closed loop from the state diagram to the actuator command. This avoids the problem of simply splicing together multiple control methods and failing to form an effective synergy.

[0025] The node characteristics of the graph nodes are uniformly defined so that they simultaneously include the amount of water content aggregation, the amount of brittleness aggregation, the amount of temperature aggregation, the amount of water content change, the amount of brittleness change, as well as the power level and opening level of the previous control cycle. This allows the strategy generation unit to not only utilize the current spatial distribution when making decisions, but also to utilize the process dynamic trend and control history, thereby reducing repeated trials under the same objective and improving the convergence speed and stability of the control sequence.

[0026] The evaluation system incorporates moisture content deviation, brittleness deviation, moisture content uniformity, brittleness uniformity, energy consumption, and time penalty into the evaluation data. The reward is generated by the difference in evaluation data between adjacent control cycles, which keeps the strategy update direction consistent with the comprehensive quality-efficiency constraint. This avoids unacceptable side effects caused by optimizing a single indicator and enables the baking process to achieve adaptive improvement while meeting multiple objective constraints.

[0027] The experience memory unit establishes an index and stores state diagram sequences, power level sequences, opening level sequences, and evaluation data according to material type parameters, loading state parameters, initial moisture content state parameters, and initial brittleness state parameters. Before generating the control sequence, it retrieves the power level sequence and opening level sequence and provides them to the strategy generation unit for use. This enables the system to quickly reuse historical effective control trajectories for similar working conditions, reduces repeated exploration, lowers the dependence on the actual number of baking cycles during the learning phase, and improves the availability and consistency of the device in actual use.

[0028] When updating the control sequence generation rules, the policy update constraint unit imposes a probability ratio range limit on the update magnitude, so that the selection probability of the updated policy and the policy before the update on the existing control sequence remains under controlled change, avoiding abnormal control sequence output caused by policy mutation, improving the safety and controllability of online update or phased update process, and enabling the device to iterate stably without introducing unpredictable risks in long-term use. Attached Figure Description

[0029] Figure 1 This is an overall structural diagram of the present invention; Figure 2 This is an exploded view of the entire invention; Figure 3 This is a front view of the present invention; Figure 4 For the present invention Figure 3 Sectional view of AA; Figure 5 For the present invention Figure 4 Enlarged view at point B in the middle; Figure 6 The local explosion of the present invention Figure 1 ; Figure 7 The local explosion of the present invention Figure 2 ; Figure 8 This is a structural diagram of the present invention.

[0030] Explanation of the labels in the diagram 1. Shell; 2. Baking cavity; 3. Supporting component; 4. Reflux cavity; 5. Micro-vortex airflow ring; 6. Heat-conducting plate; 7. Phase change layer; 8. Heating layer. Detailed Implementation

[0031] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0032] It should be noted that the directional concepts of left, right, up, down, front, back, inside, and outside in the following scheme are all relative directions, and will not be listed one by one here.

[0033] Example 1 like Figures 1 to 8 As shown, this embodiment provides a specific structural implementation of a baking and humidification integrated device. The device uses a shell 1 as a load-bearing and heat-insulating outer shell. A baking cavity 2 is formed inside the shell 1. The baking cavity 2 is used to accommodate the load-bearing component 3 and several sets of phase change metal heat-conducting substrates. A return cavity 4 is formed between the baking cavity 2 and the shell 1. The return cavity 4 serves as an annular air supply and return channel arranged around the baking cavity 2. The return cavity 4 and the baking cavity 2 are connected by multiple connecting channels, so that the return cavity 4 and each micro-vortex airflow ring 5 form a continuous gas communication relationship. A circulation channel is set inside the return cavity 4, and an internal circulation fan is set in the circulation channel. The return cavity 4 and the circulation channel together form a closed internal circulation loop, thereby forming a closed circulation air path structure inside the shell 1, which facilitates obtaining stable air supply conditions at different micro-vortex airflow rings 5.

[0034] Inside the baking cavity 2, the support component 3 is used to support the material to be baked. The support component 3 can be a wire basket structure or a perforated tray structure. The installation height of the support component 3 is kept at a predetermined distance from the inner wall of the baking cavity 2 to reserve the injection space of the micro vortex airflow ring 5 and ensure that the airflow can act on the space below and to the side of the support component 3. Through holes are opened on the support surface of the support component 3 so that the airflow from the micro vortex airflow ring 5 can penetrate the support component 3 and form a mixed flow field with the main circulating hot air in the baking cavity 2.

[0035] In this embodiment, the material to be baked can be loaded onto the support component 3 in a single-layer or multi-layer arrangement. In the single-layer arrangement, the material is spread along the support surface of the support component 3 while maintaining the spacing between each material, so that the rotating airflow generated by the micro-vortex airflow ring 5 can act on the upper surface, side surface, and lower surface of the material through the through holes of the support component 3. In an optional implementation, the central control and learning component applies an alternating sequence of opening levels to multiple micro-vortex airflow rings 5 ​​within the control cycle, so that the airflow forms a local shear flow field with directional switching on the support component 3. Thus, when the material is placed in a single layer and the friction between the material and the support surface meets the flipping threshold condition, the material is flipped or turned over by the airflow, thereby reducing the problem of uneven drying caused by heating or airflow from one side only.

[0036] In the multi-layer layered placement method, the bearing component 3 is arranged with multiple layered bearing surfaces along the vertical direction. The layered bearing surfaces are separated by spacers or columns to form a predetermined layer spacing, so that the airflow space of each layer of material is not blocked and the jet airflow can enter the gap between each layer. In the multi-layered placement method, in order to avoid insufficient flipping of the lower or inner layer of material due to airflow disturbance alone, in an optional implementation, the bearing component 3 is provided with an automatic flipping mechanism. The automatic flipping mechanism is electrically connected to the central control and learning component and is configured to drive at least one layered bearing surface or the flipping component on the bearing surface to flip at a preset flipping time, thereby realizing automatic flipping when multi-layered placement.

[0037] Furthermore, the baking process in this embodiment is performed in a segmented baking manner, which divides the overall baking process into at least two continuous baking stages, each corresponding to a different power level sequence and opening level sequence. In an optional implementation, the first baking stage uses a higher opening level and a medium power level to achieve rapid surface moisture migration and boundary layer disturbance, while the second baking stage uses a lower opening level and a stable power level to achieve uniform dehydration and brittleness formation. The central control and learning component synchronously switches the control commands for the heating layer 8 and each micro-vortex airflow ring 5 at the stage switching point, and triggers single-layer airflow flipping or triggers automatic flipping mechanism flipping when flipping is required, so that the segmented baking and flipping actions are executed collaboratively within the control cycle dimension.

[0038] Multiple sets of phase change metal thermal conductive substrates are evenly distributed within the baking cavity 2. Each set of substrates is positioned circumferentially or in a matrix along the baking cavity 2 to cover the projected area corresponding to the supporting component 3, avoiding the problem of insufficient thermal conduction support in local areas. Each set of substrates includes a heat-conducting plate 6, a phase change layer 7, and a heating layer 8. The heat-conducting plate 6 is a heat-bearing component thermally coupled to the supporting component 3. The heat-conducting plate 6 can be made of a high thermal conductivity metal plate and has an installation and positioning structure with the baking cavity 2. The upper surface of the heat-conducting plate 6 faces the supporting component 3, and the lower surface is attached to the phase change layer 7. The phase change layer 7 is a phase change heat storage material layer and forms a surface contact heat transfer path with the heat-conducting plate 6. The heating layer 8 is located on the side of the phase change layer 7 away from the heat-conducting plate 6. The heating layer 8 is a resistive heating component and is electrically connected to the central control and learning component. After being powered on, the heating layer 8 inputs heat to the phase change layer 7, and the phase change layer 7 then releases heat into the baking cavity 2 through the heat-conducting plate 6, so that the heat input has buffering and delay characteristics, thereby providing stable thermal boundary conditions for the supporting component 3. In terms of structural implementation, the interlayer fixing of the heat-conducting plate 6, the phase change layer 7 and the heating layer 8 can be formed by pressing assembly or interface thermal conductive filling material to form a continuous heat transfer interface, and the phase change layer 7 is sealed by a heat-resistant encapsulation frame to prevent leakage of the phase change material during the phase change process, which would affect the reliability of the device.

[0039] Each set of phase change metal thermally conductive substrates has a micro-vortex airflow ring 5 fixedly connected to its outer end. The micro-vortex airflow ring 5 is an annular pipe structure, and it is connected to the return cavity 4 through a connecting channel, allowing the constant pressure gas in the return cavity 4 to enter the micro-vortex airflow ring 5. The micro-vortex airflow ring 5 has multiple inclined micro-holes distributed along its circumference. The outlet end of the inclined micro-hole is located on the end face near the support component 3, and the axis of the inclined micro-hole is inclined relative to the radial direction of the micro-vortex airflow ring 5. This geometric orientation gives the fluid ejected from the micro-hole a tangential velocity component and forms a local rotating flow field in the vicinity of the micro-vortex airflow ring 5. To facilitate engineering implementation, the aperture of the inclined micro-holes is... The angle between the orifice axis and the radial direction can be selected from 0.3 mm to 1.2 mm, the angle between the orifice axis and the radial direction can be selected from 15 degrees to 45 degrees, the spacing of the micro-orifices along the circumferential direction can be selected from 2 mm to 8 mm, and the axial distance between the micro-vortex airflow ring 5 and the support component 3 can be selected from 8 mm to 30 mm. The above structural parameters are used to ensure that the jet maintains a velocity gradient that can form a rotating flow field near the support component 3 and avoids the jet from being too close to cause local strong scouring. The micro-vortex airflow ring 5 can be integrally formed from heat-resistant metal or heat-resistant composite material, and a sealing ring and locking structure are set at the connection with the connecting channel to ensure airtight communication between the return cavity 4 and the micro-vortex airflow ring 5.

[0040] Each micro-vortex airflow ring 5 is equipped with an independent airflow adjustment component, which includes a throttling valve and a drive actuator. The throttling valve is located in the communication channel between the return chamber 4 and the micro-vortex airflow ring 5 or at the air inlet of the micro-vortex airflow ring 5. The drive actuator is connected to the throttling valve and electrically connected to the central control and learning component. The central control and learning component controls the opening degree of each throttling valve, so that the air supply flow of each micro-vortex airflow ring 5 can be adjusted independently. The throttling valve can adopt a rotary valve core structure or a linear valve core structure. The drive actuator can adopt a stepper motor or a servo motor and be equipped with a position feedback encoder to ensure a repeatable correspondence between the opening degree of the throttling valve and the output command of the central control and learning component, which facilitates the subsequent mapping of the opening degree level with the control strategy.

[0041] To ensure that each micro-vortex airflow ring 5 maintains a usable air supply pressure differential under different opening conditions, a pressure detection unit is installed in the return chamber 4. The sampling end of the pressure detection unit is connected to the return chamber 4 and outputs the pressure signal from the return chamber 4 to the central control and learning component. The internal circulation fan is electrically connected to the central control and learning component. Based on the pressure feedback output by the pressure detection unit, the central control and learning component adjusts the speed or drive duty cycle of the internal circulation fan to maintain the pressure in the return chamber 4 within a preset pressure range, thereby forming a measurable constant pressure air source. In this embodiment, the preset pressure range is defined as the pressure difference relative to the pressure inside the baking cavity 2. Let the pressure in the return chamber 4 be... The pressure in baking cavity 2 is The pressure difference is The central control and learning component collects data via the pressure detection unit. For feedback quantity and maintain exist Within the interval, and During the assembly and calibration phase, parameters are written into the storage area of ​​the central control and learning component. To ensure the stability of differential pressure feedback, the pressure detection unit is equipped with a pressure tap in the reflux chamber 4 and connected to the pressure sensor through a heat-resistant pressure guide tube. At the same time, a comparison pressure tap is set in the baking chamber 2 and connected to the differential pressure sensor through a heat-resistant pressure guide tube, thereby directly outputting the differential pressure signal. The central control and learning component filters the differential pressure signal and outputs the fan control command to achieve closed-loop constant pressure control.

[0042] pressure difference It can be represented as: in, The pressure difference between the reflux chamber 4 and the baking chamber 2 is expressed in Pa (Pa). The pressure inside the reflux chamber 4 is expressed in Pa. The pressure inside the baking cavity 2 is expressed in Pa.

[0043] The central control and learning component collects data via the pressure detection unit. For feedback quantity and maintain Within a preset pressure range, preferably: in, This is the lower limit threshold of the pressure difference. This represents the upper limit threshold for differential pressure. In a set of optional implementations, Take 200Pa, A pressure of 600 Pa is used to ensure that the micro-vortex airflow ring 5 can still obtain a stable jet driving force when the opening of the throttling valve changes.

[0044] In the device structure of this embodiment, the heating layer 8 and each independent airflow regulating component are electrically connected to the central control and learning component. The central control and learning component outputs control commands to the power supply of each heating layer 8 and the opening degree of the throttling valve of each micro-vortex airflow ring 5. The power supply line of the heating layer 8 is equipped with a current detection unit to provide sampling signals for the power closed loop, and the drive actuator of the throttling valve is equipped with a position feedback encoder to provide sampling signals for the opening closed loop. The central control and learning component collects, verifies, and stores the sampling signals to form the data source for the control execution layer; at the same time, the central control and learning component receives the output of the pressure detection unit. The signal is collected and the internal circulation fan is adjusted to form a stable differential pressure air source in the return chamber 4, thereby ensuring that the opening change of the independent airflow adjustment component can correspond to the flow change of the micro vortex airflow ring 5 and can be repeated.

[0045] During one control cycle of the device's operation, the central control and learning component first acquires the output from the pressure detection unit. The signal is used to determine whether the pressure falls within a preset range. Less than Then increase the speed of the internal circulation fan. Greater than The internal circulation fan speed is reduced to bring the pressure difference in the return chamber 4 back to the preset pressure range. After the pressure difference in the return chamber 4 stabilizes, the central control and learning component outputs a valve position command to the corresponding drive actuator according to the current opening level and collects the valve position signal returned by the position feedback encoder to complete the opening closed-loop calibration. At the same time, it outputs a power supply command to the corresponding heating layer 8 according to the current power level and collects the current signal returned by the current detection unit to calculate the actual power of the heating layer 8. The actual power is output by the central control and learning component to the data recording module to form traceable execution data. In the above process, the data source includes the output of the differential pressure sensor. The sampling process includes sampling the opening degree output from the encoder that provides position feedback for the driven actuator and sampling the current output from the current detection unit. The processing also includes... The filtering and interval judgment, valve position calibration of opening degree sampling, and power conversion of current sampling are performed. The output results include internal circulation fan control commands, throttling valve opening commands, and heating layer 8 power supply commands, forming an execution dataset that can be called by subsequent control strategies.

[0046] Through the aforementioned structure and data closed loop, this embodiment achieves measurable constant-pressure gas supply to the reflux chamber 4, independent gas supply adjustment of the micro-vortex airflow ring 5, and layered thermal conductive structure configuration of the phase change metal thermal conductive substrate. This enables the space containing the support component 3 to have controlled gas injection conditions and controlled thermal boundary conditions. The arrangement and connection methods of the shell 1, baking cavity 2, support component 3, reflux chamber 4, micro-vortex airflow ring 5, heat-conducting plate 6, phase change layer 7, heating layer 8, pressure detection unit, internal circulation fan, throttling valve, and drive actuator complete the device manufacturing and assembly. The feedback control relationship establishes constant pressure control parameters, enabling the engineering implementation of this embodiment.

[0047] Example 2 like Figures 1 to 8 As shown, based on the device structure of Example 1, this example further describes the implementation method of the state reconstruction unit in the multimodal sensing component and the central control and learning component in the baking cavity 2, so as to realize the acquisition, calibration, mapping and analysis of multimodal observation data, and the alternating iterative reconstruction of the domain, and stably output the spatial state diagram corresponding to the distribution of the phase change metal thermal conductive substrate.

[0048] A multimodal sensing component is housed within the baking cavity 2 and electrically connected to the central control and learning component. The multimodal sensing component includes a photoacoustic moisture content detection submodule and a structural acoustic fragility detection submodule. The photoacoustic moisture content detection submodule includes a pulsed near-infrared light source, an optical window located at the top of the baking cavity 2, and a photoacoustic pickup array. The pulsed near-infrared light source is fixed inside the housing 1 and emits pulsed light towards the support component 3 within the baking cavity 2 through the optical window. The photoacoustic pickup array is fixed at the top or near the top of the baking cavity 2, facing the spatial area of ​​the support component 3. The channel layout of the photoacoustic pickup array is matched with the uniform distribution of the micro-vortex airflow ring 5 within the baking cavity 2. The structural acoustic brittleness detection submodule includes an acoustic excitation unit and an acoustic receiving unit. The acoustic excitation unit and the acoustic receiving unit are set on the cavity wall of the baking cavity 2 or on the corresponding structure of the phase change metal thermal conductive substrate, and are connected to the central control and learning component through heat-resistant leads. The acoustic excitation unit is used to apply acoustic excitation to the space where the supporting component 3 is located, and the acoustic receiving unit is used to collect acoustic signals related to the structural response of the material supported by the supporting component 3. The arrangement position of the acoustic excitation unit and the acoustic receiving unit establishes a fixed correspondence with the distribution position of the phase change metal thermal conductive substrate, so as to associate the acoustic observation data with the spatial region corresponding to the phase change metal thermal conductive substrate in subsequent observation mapping.

[0049] The central control and learning component includes a state reconstruction unit and stores observation mapping parameters in its memory. The observation mapping parameters are generated and written during the assembly and calibration stage. During the assembly and calibration stage, a calibration load is placed in the baking cavity 2 and the pulsed near-infrared light source, photoacoustic pickup array, acoustic excitation unit and acoustic receiving unit are driven to complete multiple sets of sampling. The central control and learning component generates observation mapping parameters after performing consistency verification on the sampled data. The observation mapping parameters are used to establish the mapping relationship between the observation data and the voxel set and are used for subsequent data consistency updates.

[0050] To align the spatial state diagram with the arrangement of the phase change metal thermal conductive substrates, the state reconstruction unit establishes a three-dimensional voxel set during the initialization phase, using the distribution position of the phase change metal thermal conductive substrates as geometric constraints. The voxel set covers the effective baking space above the bearing component 3, and divides the spatial region corresponding to each group of phase change metal thermal conductive substrates into several voxel subsets, so that each group of phase change metal thermal conductive substrates has a unique corresponding voxel subset index in the voxel set. After the voxel set is established, the state reconstruction unit defines the state vectors of the voxel-level water content state estimate and the voxel-level brittleness state estimate based on the voxel set, and synchronously forms the observation vectors by timestamps of the observation data output by the multimodal sensing components in this control cycle. The observation vectors include photoacoustic observation sub-vectors and acoustic observation sub-vectors. The central control and learning components perform channel gain correction, bandpass filtering and sampling alignment on the observation vectors to obtain the standardized observation vectors used for reconstruction.

[0051] In an optional implementation, when the load-bearing component 3 is loaded with materials in a multi-layered arrangement, the state reconstruction unit divides the voxel set vertically into multiple layered voxel subsets. Each layered voxel subset corresponds to an effective wind-receiving space above a layered load-bearing surface of the load-bearing component 3, and establishes a voxel index range for each layered voxel subset. The central control and learning component performs consistency constraint updates on each layered voxel subset based on the observation data of the multi-modal sensing component, so that the output spatial state map simultaneously includes the amount of water-containing state aggregation and the amount of brittle state aggregation in different layers. Thus, under the multi-layered arrangement, the spatial state map still maintains the correspondence between the distribution of the phase change metal thermal conductive substrate and the wind-receiving space of each layer of material.

[0052] In this embodiment, the observation mapping parameter is used to give the discrete observation relationship between the standardized observation vector and the state vector. The discrete observation relationship is represented in a unified linear observation form as follows: in, The standardized observation vector includes the standardized outputs of photoacoustic and acoustic sensors; The state vector is formed by concatenating the voxel-level water content state estimate and the voxel-level brittleness state estimate. Let be the observation mapping matrix determined by the observation mapping parameters, and its dimension is . ; This is for observing the noise term.

[0053] Observation mapping matrix The voxel subset index is structurally consistent with the distribution of phase change metal thermal conductive substrates, ensuring that the voxel subset corresponding to each group of phase change metal thermal conductive substrates is in... It has a fixed column index range, which facilitates the subsequent association of the reconstruction results with each group of phase change metal thermally conductive substrates to output a spatial state diagram.

[0054] For ease of expression, the observation vector and the state vector are concatenated as sub-vectors and written as follows: in, This is the photoacoustic observation sub-vector acquired and standardized by the photoacoustic water content detection sub-module. This is the acoustic observation subvector acquired and standardized by the structural acoustic fragility detection submodule; This is a vector of voxel-level water state estimates. This is a vector of voxel-level brittleness state estimates.

[0055] To perform alternating iterative reconstruction of the analysis domain, the state reconstruction unit establishes a spatial discrete gradient operator. The spatial discrete gradient operator is generated by the adjacency relationship of the voxel set. The adjacency relationship of the voxel set is determined by the arrangement position of the phase change metal thermal conductive substrate and the arrangement position of the micro vortex airflow ring 5. Specifically, the state reconstruction unit establishes adjacency edges between voxel subsets corresponding to adjacent phase change metal thermal conductive substrates, and establishes adjacency edges between voxel subsets corresponding to the micro vortex airflow ring 5 that forms a backflow coupling relationship through the backflow cavity 4. The above adjacency edges are mapped to an adjacency relationship table between voxels. The spatial discrete gradient operator performs a difference operation on the voxel state in the adjacency direction to form a gradient representation. The gradient representation includes the gradient representation of the voxel-level water-containing state estimate and the gradient representation of the voxel-level brittle state estimate.

[0056] The state reconstruction unit executes an alternating iterative reconstruction process of the analysis domain in each control cycle to output a spatial state map. The alternating iterative reconstruction process of the analysis domain consists of alternating prior updates and data consistency updates. The prior update performs denoising updates on the gradient representations of the voxel-level water-bearing state estimate and the voxel-level brittle state estimate, respectively, and outputs the gradient target for data consistency updates. The data consistency update performs consistency constraint updates on the voxel-level water-bearing state estimate and the voxel-level brittle state estimate based on the observation mapping parameters, and makes the updated state vector satisfy the observation vector constraint. For ease of implementation, at the beginning of the current control cycle, the state reconstruction unit uses the state vector obtained by back-calculating the spatial state map output by the previous control cycle as the initial value. The first control cycle uses a uniform initial value or a calibration load initial value as the initial value, and sets the iteration number, denoising intensity sequence, and consistency weight parameters.

[0057] In one optional implementation, the alternating iterative reconstruction process of the analysis domain consists of alternating "prior update sub-steps" and "data consistency update sub-steps," wherein the prior update sub-steps output gradient objectives. (See below), the data consistency update sub-step is performed given a gradient objective. Under the given conditions, the state vector is updated, which can be equivalently expressed as solving the following data consistency subproblem: in, Let the state vector be the one to be optimized. Describing the L2 norm, For consistency of observed data; It is a spatial discrete gradient operator; This is the consistency weight parameter.

[0058] In the prior update step, the state reconstruction unit first calculates the gradient representations of the voxel-level water content state estimate and the voxel-level brittleness state estimate based on the current iteration's state vector, and inputs them into the corresponding denoising update operators to obtain the gradient objective. The output form of the gradient objective is expressed as: in, This is the state vector from the previous iteration. This is the gradient target vector output by the current iteration; To update the operator for noise reduction, This is the denoising intensity parameter for this iteration.

[0059] The denoising update operator is implemented by using a computational structure that segments and thresholds the gradient representation and constrains the direction consistency, so that the gradient representation tends to be smooth in the non-boundary region and maintains the abrupt change characteristics of the gradient representation in the boundary region.

[0060] In the data consistency update step, the state reconstruction unit is in the observation mapping matrix With observation vector The state vector is updated under constraints so that the updated state vector is consistent with the observed vector in the observation domain and closely fits the gradient objective in the analysis domain. This is achieved by solving the following system of linear equations: in, For the observation mapping matrix transpose, For spatial discrete gradient operators transpose; For consistency weight parameters; Let be the current state vector to be solved. The gradient target is the output of the prior update step.

[0061] The state reconstruction unit no longer processes the photoacoustic observation sub-vector and the acoustic observation sub-vector of the observation vector separately, but instead integrates them uniformly. And call the same observation mapping matrix Consistency constraints are implemented to ensure that the voxel-level water content estimates and voxel-level brittleness estimates are jointly updated within the same iterative framework. The numerical solution for data consistency updates employs preconditional conjugate gradient iteration or a normal equation solver. The solver's stopping condition is limited by both a residual threshold and the maximum number of iterations. Upon completion, a new state vector is output. Then proceed to the next round of prior update steps.

[0062] After completing the prescribed number of iterations within a control cycle, the state reconstruction unit splits the final state vector according to the voxel set index to obtain voxel-level water content state estimates and voxel-level brittleness state estimates. These estimates are then aggregated according to the voxel subset indices corresponding to the phase change metal thermally conductive substrate to generate a spatial state diagram. The spatial state diagram, in its data structure, includes the aggregated water content and brittleness state values ​​for each voxel subset, maintaining a one-to-one correspondence with the phase change metal thermally conductive substrate for subsequent topology binding unit calls. Data sources within this control cycle include photoacoustic observation data output from the photoacoustic pickup array and acoustic observation data output from the acoustic receiving unit. The data processing includes sampling alignment, channel correction, observation vector construction, data consistency update under the constraint of the observation mapping matrix, gradient representation generation under the spatial discrete gradient operator, denoising and updating the output gradient target, and alternating iterative solution. The output results include voxel-level water content state estimates, voxel-level brittleness state estimates, and spatial state maps corresponding to the distribution of phase change metal thermally conductive substrates. After the spatial state map is output, it is written into the shared buffer of the central control and learning components, along with the control cycle timestamp, observation vector summary, and iterative residual summary, for use in the initial value selection and consistency verification of subsequent control cycles.

[0063] Example 3 like Figures 1 to 8 As shown, based on the constant pressure air source structure formed by the shell 1, baking cavity 2, bearing component 3, reflux cavity 4, micro-vortex airflow ring 5, heat conduction plate 6, phase change layer 7, heating layer 8, pressure detection unit, and internal circulation fan in Example 1, and based on the spatial state diagram output by the state reconstruction unit in Example 2, this embodiment further describes the collaborative implementation method of the topology binding unit, strategy generation unit, experience memory unit, and strategy update constraint unit in the central control and learning component. This realizes the state diagram construction, node feature definition, power level sequence and opening level sequence generation, indexed experience reuse, evaluation data calculation, reward quantity generation, and constrained strategy update, forming a closed-loop execution baking humidity control process.

[0064] In this embodiment, the central control and learning component establishes a binding table of "node-actuator-voxel set" during the assembly initialization phase. The binding table uses each group of phase change metal thermal conductive substrates as the primary index and binds the corresponding heating layer 8 channels, the corresponding micro-vortex airflow ring 5 and its independent airflow adjustment component channels, as well as the voxel set index corresponding to the spatial coverage area of ​​the phase change metal thermal conductive substrate in the spatial state diagram of Embodiment 2. The binding table serves as the basic data for generating the state diagram using the graph topology binding unit and remains unchanged throughout the entire device operation process. Simultaneously, the central control and learning component establishes an adjacent arrangement relationship table, where adjacent arrangements... The relationship table is determined based on the geometric arrangement of the phase change metal heat-conducting substrates in the baking cavity 2. Pairs of phase change metal heat-conducting substrates whose geometric distance meets the preset adjacency conditions are recorded as adjacent node pairs. The central control and learning component also establishes a recirculation coupling relationship table, which is determined based on the arrangement of the connecting channels between the micro vortex airflow ring 5 and the recirculation cavity 4. Pairs of micro vortex airflow rings 5 ​​that share the same recirculation branch section of the recirculation cavity 4 are recorded as recirculation coupling node pairs. The adjacent arrangement relationship table and the recirculation coupling relationship table together determine the graph edge set of the graph topology binding unit generating state graph, so that the graph node and graph edge are connected to the specific structure in a one-to-one correspondence.

[0065] In an optional implementation, when the carrier component 3 is loaded with materials in a multi-layered arrangement and an automatic flipping mechanism is provided on the carrier component 3, the central control and learning component further binds the flipping execution channel of the automatic flipping mechanism in the binding table. The flipping execution channel is used to receive flipping trigger commands and drive the automatic flipping mechanism to perform flipping movements. The flipping execution channel can be electrically connected to the drive actuator of the carrier component 3 and has position feedback or count feedback. It is used to write the flipping action into the execution dataset and align it with the control cycle timestamp, so that the automatic flipping during multi-layered arrangement can be recorded and reused in the form of repeatable execution data.

[0066] The graph topology binding unit receives the spatial state diagram output by the state reconstruction unit in each control cycle, and extracts the voxel set corresponding to each phase change metal thermal conductive substrate in the spatial state diagram as the node data source according to the binding table, thereby constructing the node feature vector of the graph node. The node feature vector is formed by splicing the following data in a fixed order: water content aggregation amount of voxel set, brittleness aggregation amount of voxel set, temperature aggregation amount of voxel set, water content change amount, brittleness change amount, power level of the previous control cycle, and opening level of the previous control cycle. Among them, the water content aggregation amount is calculated by the voxel-level water content estimate value in the voxel set according to the aggregation rule, the brittleness aggregation amount is calculated by the voxel-level brittleness estimate value in the voxel set according to the aggregation rule, and the temperature aggregation amount is obtained by power sampling of the heating layer 8 bound to the phase change metal thermal conductive substrate, temperature sampling of the heat conduction plate 6, and baking. The ambient temperature of the baking cavity 2 is obtained by sampling and fusion rules; the change in water content is determined by the difference between the aggregated water content of the current control cycle and the aggregated water content of the previous control cycle, and the change in brittleness is determined by the difference between the aggregated brittleness of the current control cycle and the aggregated brittleness of the previous control cycle; the power level and opening level of the previous control cycle are respectively mapped by the control commands of the heating layer 8 recorded by the central control and learning components and the control commands of the independent airflow adjustment components; after the node feature vector is generated, the graph topology binding unit generates a graph edge set according to the adjacent arrangement relationship table and the return coupling relationship table, and constructs a state graph with the graph node set and the graph edge set. The state graph is written into the shared buffer and the state vector is output. The state vector is a global representation of the state graph after summarizing the node features. The state vector maintains index consistency with each graph node and is used as input for the strategy generation unit.

[0067] The strategy generation unit receives a state vector in each control cycle and outputs a power level sequence and an opening level sequence. The power level sequence is the set of power levels for each heating layer 8 in the next control cycle, and the opening level sequence is the set of opening levels for each independent airflow regulating component in the next control cycle. To achieve feasible engineering control, the power level sequence is represented by a discrete level set, which is obtained by dividing the allowable power supply range of the heating layer 8 according to a preset number of levels. Each level corresponds to a set of repeatable power supply parameters. The opening level sequence is also represented by a discrete level set, which is obtained by dividing the full stroke opening of the throttling valve according to a preset number of levels. Each level corresponds to a set of repeatable valve position targets. The central control and learning component establishes a level-command mapping table for each level and maps the power level to the power supply duty cycle or current set value of the heating layer 8, and maps the opening level to the target pulse number or target position value of the driving actuator, so as to ensure that the output of the strategy generation unit can directly drive the execution layer.

[0068] In an optional implementation, for multi-layered placement, the strategy generation unit further outputs a flip trigger sequence. The flip trigger sequence is used to indicate whether the automatic flipping mechanism should perform a flip in the next control cycle. The central control and learning component maps the flip trigger sequence to the trigger command of the flip execution channel, and collects position feedback or count feedback after triggering to form a flip sample. This sample is used to record the correlation between the flipping action and the change in the spatial state diagram in the evaluation data calculation, so that the closed-loop process of "automatic flipping - state update - strategy adjustment" in multi-layered placement has a traceable data link.

[0069] The experience memory unit establishes an index based on material type parameters, loading state parameters, initial moisture content parameters, and initial brittleness parameters. Material type parameters are written by the input interface or recognition module of the central control and learning component. Loading state parameters are determined by weight sampling and load area occupancy sampling of the bearing component 3. Initial moisture content parameters and initial brittleness parameters are calculated from the spatial state diagram of the first control cycle at the start of control. The experience memory unit stores control trajectory records using the index as the key. Each control trajectory record includes at least a state diagram sequence, power level sequence, opening level sequence, and evaluation data. A timestamp and end marker are written to each control trajectory record for subsequent retrieval and updates. Before generating the control sequence in each control cycle, the experience memory unit... The system performs retrieval based on an index, employing hierarchical matching rules. First, it matches material type parameters and loading status parameters, then matches preset intervals of initial moisture content parameters and initial brittleness parameters, resulting in a retrieval power level sequence and a retrieval opening level sequence. When generating new power level and opening level sequences, the strategy generation unit uses these sequences as candidate sequence inputs, thus completing the level selection for the next control cycle within the candidate sequence set. The selection result is then output as a new power level sequence and a new opening level sequence. This candidate sequence input mechanism couples the output of the strategy generation unit with the index structure of the experience memory unit, preventing the strategy generation unit from outputting unexecutable level combinations when prior experience is lacking.

[0070] The evaluation data is calculated by the central control and learning component at the end of each control cycle. The evaluation data includes water content deviation index, brittleness deviation index, water content uniformity index, brittleness uniformity index, energy consumption index, and time penalty index. Specifically, the water content deviation index is obtained by weighted summation of the differences between the aggregated water content of each graph node and the preset target water content value; the brittleness deviation index is obtained by weighted summation of the differences between the aggregated brittleness of each graph node and the preset target brittleness value; the water content uniformity index is calculated by the discrete measure of the aggregated water content of each graph node; the brittleness uniformity index is calculated by the discrete measure of the aggregated brittleness of each graph node; the energy consumption index is obtained by the sum of the actual power sampling integral of each heating layer 8 and the power sampling integral of the internal circulation fan within this control cycle; and the time penalty index is determined by the deviation between the control cycle count and the preset completion time window.

[0071] To facilitate the standardized generation of reward amounts, the evaluation data is synthesized into a single evaluation value, which is expressed in a weighted sum form as follows: in, For a moment The overall evaluation value; As an indicator of moisture content deviation, It is an index of brittleness deviation. As an indicator of water content uniformity, As an indicator of brittleness uniformity, As an energy consumption indicator, As a time penalty indicator; The weight parameters for the corresponding indicators are written into the central control and learning component during the assembly calibration stage.

[0072] in, For a moment The amount of reward; The evaluation value from the previous control cycle. This is the evaluation value for the current control cycle; To prevent small constants with a denominator of zero (e.g.) ); The interval between adjacent control cycles; This represents the natural logarithm function.

[0073] When constructing the above reward amount, when the evaluation value improves ( )but It is positive; when the evaluation value deteriorates, it is negative. It is negative.

[0074] When the strategy update constraint unit updates the control sequence generation rules of the strategy generation unit, it adopts a constrained update method to limit the update magnitude. The control sequence generation rules of the strategy generation unit are stored in the form of parameterized strategy. The output of the parameterized strategy is the selection probability distribution of each candidate power level sequence and candidate opening level sequence.

[0075] In an alternative implementation, the policy update constraint unit can update the parameterized policy based on the proximal policy optimization algorithm (PPO), whose objective function can be expressed as: in, For strategy parameters The constrained update target value is represented; This represents the expected calculation of the control trajectory and time step for the sampled data; This indicates taking the smaller of the two values; Indicates will Cut off to interval Inside.

[0076] For ease of implementation, the ratio of the selection probability of the updated policy to that of the unupdated policy for the same control sequence can be defined as the probability ratio. : in, For the updated policy in state Next selection control sequence The probability, For the policy before the update in the state Next selection control sequence The probability of.

[0077] At the same time, the advantage function estimate can be... Defined as the difference between the discounted return and the value baseline: in, Used to characterize at time... Select the degree of advantage of the control sequence relative to the baseline; Indicates the time from Summation from the beginning of the trajectory until its termination; As a discount factor, satisfying ; For a moment The amount of reward; This is the state value function corresponding to the policy before the update.

[0078] In the specific implementation, the policy update constraint unit calculates the ratio of the probability of the updated policy selecting an existing control sequence to the probability of the policy selecting an existing control sequence before the update. This probability ratio is denoted as... and will Limited to the lower threshold With upper limit threshold Within a defined probability ratio range, this reduces the risk of policy mutation and improves the stability of the control sequence output. For strategy parameters, This is a control sequence (a combination of power level sequence and opening level sequence). For state vectors, The truncation parameter is half the width of the probability ratio interval.

[0079] In the closed-loop execution process of this embodiment, the central control and learning component completes the data flow closure according to the control cycle sequence: at the beginning of the control cycle, according to Embodiment 1, the pressure feedback output by the pressure detection unit is collected and the internal circulation fan is adjusted to keep the pressure in the return chamber 4 within the preset pressure range. Subsequently, according to Embodiment 2, the multimodal sensing component is driven to collect observation data and the spatial state diagram is output by the state reconstruction unit. Then, the graph topology binding unit generates the state diagram and outputs the state vector. The experience memory unit obtains the power level sequence and the opening level sequence based on the index retrieval. The strategy generation unit outputs the power level sequence and the opening level sequence based on the state vector and combined with the power level sequence and the opening level sequence. The central control and learning component drives each heating layer 8 and each independent airflow adjustment component to execute the control sequence accordingly and collects power sampling, valve position sampling and state diagram update results at the end of the control cycle. The evaluation data, evaluation value, and reward amount are calculated and written into the experience memory unit. When the update trigger condition is met, the strategy update constraint unit reads the control trajectory record in the experience memory unit, performs a constrained update, and writes the update result back to the strategy generation unit. The data sources in this embodiment include spatial state diagram, power sampling of heating layer 8, throttling valve opening sampling, internal circulation fan power sampling, and control cycle counting. The processing includes node feature construction, state diagram generation, index retrieval, candidate sequence generation, gear mapping and execution, evaluation data calculation, reward amount generation, trajectory storage, and constrained strategy update. The output results include the power gear sequence, opening gear sequence, updated control sequence generation rules, and control trajectory record bound to the index for the next control cycle. This enables the baking humidity control to quickly converge to a stable control sequence through experience reuse under the same material type and similar initial state.

[0080] Among the optional implementations, the update trigger condition can be set to the control cycle count reaching a preset update cycle, or the number of control trajectory records cumulatively written by the experience memory unit reaching a preset threshold, or the improvement of the comprehensive evaluation value being lower than a preset threshold in multiple consecutive control cycles. Those skilled in the art can select the appropriate condition based on the device's computing power, real-time performance, and stability requirements.

[0081] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A baking and moisture equalization integrated device, characterized in that, The system includes a housing (1), a baking cavity (2) disposed within the housing (1), a support component (3) disposed within the baking cavity (2), a reflux cavity (4) disposed between the baking cavity (2) and the housing (1), multiple micro-vortex airflow rings (5) connected to the reflux cavity (4) and uniformly distributed within the baking cavity (2), and a phase change metal heat-conducting substrate corresponding to each of the micro-vortex airflow rings (5), and a central control and learning component; each of the phase change metal heat-conducting substrates includes a heat-conducting plate (6), a phase change layer (7) disposed on the heat-conducting plate (6), and a heating layer (8) disposed on the phase change layer (7), with the outer end of each phase change metal heat-conducting substrate being fixed to the corresponding micro-vortex airflow ring (5). Fixed connection; multiple inclined micro-holes are opened on one end of the micro-vortex airflow ring (5) near the bearing component (3), and each micro-vortex airflow ring (5) is provided with an independent airflow adjustment component; an internal circulation fan and a pressure detection unit are provided in the return cavity (4), the internal circulation fan is electrically connected to the central control and learning component and is adjusted by the central control and learning component under the pressure feedback output by the pressure detection unit, so as to keep the pressure of the return cavity (4) within a preset pressure range; the heating layer (8) and the independent airflow adjustment component are both electrically connected to the central control and learning component; a multimodal sensing component is provided in the baking cavity (2), and the multimodal sensing component is electrically connected to the central control and learning component; The central control and learning component includes a state reconstruction unit, a graph topology binding unit, a policy generation unit, a policy update constraint unit, and an experience memory unit. The state reconstruction unit stores observation mapping parameters and is configured to generate a set of voxels corresponding to the distribution of the phase change metal thermal conductive substrate based on the observation data output by the multimodal sensing component and the observation mapping parameters, and to generate voxel-level water content state estimates and voxel-level brittleness state estimates. The state reconstruction unit is configured to execute an alternating iterative reconstruction process of the analysis domain in each control cycle to output a spatial state graph. The alternating iterative reconstruction process of the analysis domain includes prior updates to the gradient representation of the voxel-level water content state estimates and voxel-level brittleness state estimates based on the spatial discrete gradient operator, and data consistency updates to the voxel-level water content state estimates and voxel-level brittleness state estimates based on the observation mapping parameters, and the prior updates and data consistency updates are executed alternately. The graph topology binding unit is configured to use each phase change metal thermal conductive substrate... The micro-vortex airflow ring (5) and its fixed connection are used as graph nodes, and the adjacent arrangement relationship between the phase change metal heat-conducting substrates and the recirculation coupling relationship formed by the micro-vortex airflow ring (5) through the recirculation cavity (4) are used as graph edges to generate the state diagram; the strategy generation unit is configured to output the power level sequence of the heating layer (8) and the opening level sequence of the independent airflow adjustment component based on the state diagram and form a control sequence; the experience memory unit is configured to establish an index with material type parameters, loading state parameters, initial moisture content state parameters and initial brittleness state parameters, and store the state diagram sequence, power level sequence, opening level sequence and evaluation data corresponding to the index; the strategy update constraint unit is configured to generate a reward amount based on the evaluation data of adjacent control cycles, and when updating the generation rules of the control sequence, the ratio of the selection probability of the updated strategy for the existing control sequence to the selection probability of the previous strategy for the existing control sequence is limited to the probability ratio range defined by the lower limit threshold and the upper limit threshold.

2. The baking and moisture equalization integrated device as described in claim 1, characterized in that, The reflux chamber (4) is arranged around the baking chamber (2), and a communication channel is provided between the reflux chamber (4) and each of the micro vortex airflow rings (5). The internal circulation fan is set in the circulation channel of the reflux chamber (4), and the reflux chamber (4) and the circulation channel form a closed internal circulation loop.

3. The baking and moisture equalization integrated device as described in claim 1, characterized in that, The independent airflow regulating component includes a throttling valve and a drive actuator disposed within the micro vortex airflow ring (5). The drive actuator is electrically connected to the central control and learning component, and the central control and learning component controls the opening degree of each of the throttling valves.

4. The baking and moisture equalization integrated device as described in claim 1, characterized in that, The inclined micro-holes are distributed circumferentially along the micro-vortex airflow ring (5), and the hole axis of the inclined micro-holes is inclined relative to the radial direction of the micro-vortex airflow ring (5) and forms an outlet end on one side end face near the support component (3).

5. The baking and moisture equalization integrated device as described in claim 1, characterized in that, The multimodal sensing component includes a photoacoustic moisture content detection submodule and a structural acoustic fragility detection submodule; the photoacoustic moisture content detection submodule includes a pulsed near-infrared light source, an optical window disposed on the top of the baking cavity (2), and a photoacoustic pickup array; the structural acoustic fragility detection submodule includes an acoustic excitation unit and an acoustic receiving unit, and the acoustic excitation unit and the acoustic receiving unit are disposed on the cavity wall of the baking cavity (2) or on the phase change metal thermal conductive substrate.

6. The baking and moisture equalization integrated device as described in claim 1, characterized in that, The spatial discrete gradient operator is generated by the adjacency relationship of the voxel set, and the adjacency relationship of the voxel set is determined by the arrangement position of the phase change metal thermally conductive substrate and the arrangement position of the micro vortex airflow ring (5); the prior update performs denoising update on the gradient representation of the voxel-level water-containing state estimate and the gradient representation of the voxel-level brittle state estimate respectively and outputs the gradient target for the data consistency update.

7. The baking and moisture equalization integrated device as described in claim 1, characterized in that, The node features of the graph node include the amount of water-containing polymerization, the amount of brittle polymerization, the amount of temperature polymerization, the amount of water-containing change, the amount of brittle change, the power level of the previous control cycle, and the opening level of the previous control cycle of the voxel set corresponding to the phase change metal thermal conductive substrate.

8. The baking and moisture equalization integrated device as described in claim 1, characterized in that, The evaluation data includes moisture content deviation index, brittleness deviation index, moisture content uniformity index, brittleness uniformity index, energy consumption index, and time penalty index, and the reward amount is determined by the difference in the evaluation data between adjacent control cycles.

9. The baking and moisture equalization integrated device as described in claim 1, characterized in that, Before generating the control sequence, the experience memory unit retrieves the power level sequence and the opening level sequence based on the index, and the strategy generation unit generates the power level sequence and the opening level sequence based on the state diagram and in combination with the power level sequence and the opening level sequence.

10. The baking and moistening integrated apparatus according to any one of claims 1 to 9, characterized in that, The central control and learning component executes a baking humidity control method, which includes steps S1 to S6: Step S1: Drive the pulsed near-infrared light source and collect the photoacoustic observation data output by the photoacoustic pickup array within the control cycle, and drive the acoustic excitation unit and collect the acoustic observation data output by the acoustic receiving unit. Step S2: The state reconstruction unit establishes the voxel set based on the observation mapping parameters, and generates the voxel-level water content state estimate and the voxel-level brittleness state estimate based on the photoacoustic observation data and the acoustic observation data. Step S3: The state reconstruction unit establishes the spatial discrete gradient operator and executes the analysis domain alternating iterative reconstruction process to output the spatial state diagram; Step S4: The graph topology binding unit uses the phase change metal thermal conductive substrate and the micro vortex airflow ring (5) fixedly connected to it as graph nodes, and uses the adjacent arrangement relationship and the backflow coupling relationship as graph edges to generate the state graph and output the state vector. Step S5: The experience memory unit retrieves the power level sequence and the opening level sequence based on the index, and the strategy generation unit outputs the power level sequence and the opening level sequence based on the state diagram and in combination with the power level sequence and the opening level sequence. It also drives each heating layer (8) and each independent airflow adjustment component to operate according to the control sequence, and drives the internal circulation fan to adjust under the pressure feedback output by the pressure detection unit so that the pressure of the return chamber (4) is kept within the preset pressure range. Step S6: The strategy update constraint unit generates the reward amount based on the evaluation data of adjacent control cycles, and when updating the generation rule of the control sequence, the ratio of the selection probability of the updated strategy for the existing control sequence to the selection probability of the unupdated strategy for the existing control sequence is limited to the probability ratio range, and the experience memory unit stores the state diagram sequence, the power level sequence, the opening level sequence and the evaluation data.

Images