Real-time dynamic tuning system for microwave drying process

Abstract

The invention relates to the technical field of industrial control, and discloses a real-time dynamic tuning system for a microwave drying process, which is characterized in that temperature distribution, microwave power intensity, humidity gradient and material deformation parameters in a drying cavity are acquired in real time, and time-space alignment and noise suppression are carried out on data of a multi-source sensor based on a Kalman filtering algorithm; generating a drying state matrix, generating a three-dimensional regulation and control instruction containing microwave emission frequency, power partition and drying time according to the drying state matrix by adopting a mixed algorithm of model predictive control and fuzzy logic, and taking the deviation between a final moisture content detection value of the material and a model predictive value as a loss function through an online error back propagation model. The weight coefficient of the microwave-thermal-mass transfer coupling model is dynamically corrected, the microwave energy ratio is optimized based on the improved particle swarm optimization, and the three-dimensional gradient field intensity distribution matched with the material morphology is generated according to the three-dimensional regulation and control instruction, so that the efficiency of the microwave drying process during real-time dynamic regulation and optimization is improved.

Description

Technical Field

[0001] The present invention relates to the field of industrial control technology, and in particular to a real-time dynamic optimization system for a microwave drying process. Background Art

[0002] Microwave drying technology, due to its advantages such as rapid heating and high energy efficiency, has been widely used in food processing, pharmaceutical manufacturing, ceramic sintering, and other fields. However, existing technologies still face the following key bottlenecks in actual industrial applications: reliance on single sensor feedback (such as temperature) prevents real-time analysis of the interactive effects of the microwave field, heat conduction, and moisture migration; static models are unable to adapt to dynamic changes in material parameters such as dielectric properties and thermal conductivity; frequent manual parameter adjustments are required for different batches or types of materials; and there is a lack of real-time abnormality warning mechanisms. Localized overheating or abnormal energy consumption can easily lead to equipment damage and product quality issues.

[0003] While some existing research attempts to incorporate model predictive control or optimization algorithms, these technologies still face challenges such as slow algorithm convergence, imperfect multi-physics field coupling modeling, and lagging feedback mechanisms. Therefore, a microwave drying system capable of real-time dynamic optimization and adaptive multi-condition adaptation is urgently needed to address these technical bottlenecks. Summary of the Invention

[0004] The present invention provides a real-time dynamic tuning system for a microwave drying process, the main purpose of which is to solve the problem of low efficiency during the real-time dynamic tuning of the microwave drying process.

[0005] To achieve the above-mentioned object, the present invention provides a real-time dynamic tuning system for a microwave drying process, characterized in that the system includes a multi-source sensor group, a data fusion unit, an adaptive decision unit, a feedback compensation module, and a multi-channel microwave execution unit, wherein:

[0006] Multi-source sensor group, used to collect temperature distribution, microwave power intensity, humidity gradient and material deformation parameters in the drying chamber in real time;

[0007] The data fusion unit performs spatiotemporal alignment and noise suppression on multi-source sensor data based on the Kalman filter algorithm to generate a drying state matrix;

[0008] The adaptive decision-making unit uses a hybrid algorithm of model predictive control and fuzzy logic to generate three-dimensional control instructions including microwave emission frequency, power partition and drying time according to the drying state matrix;

[0009] The feedback compensation module uses an online error back propagation model to take the deviation between the final moisture content of the material and the model prediction value as a loss function to dynamically correct the weight coefficient of the microwave-thermodynamic-mass transfer coupling model;

[0010] The multi-channel microwave execution unit optimizes the microwave energy ratio based on the improved particle swarm algorithm, and generates a three-dimensional gradient field intensity distribution that matches the material morphology according to the three-dimensional control instructions.

[0011] Optionally, the data fusion unit further includes:

[0012] The spatiotemporal correlation module synchronizes the sampling data of multiple source sensors through timestamps;

[0013] The noise suppression module adopts the cascade structure of wavelet transform and Kalman filter to filter out the carbonization noise of temperature distribution and electromagnetic interference of microwave power intensity.

[0014] Optionally, the hybrid algorithm includes:

[0015] The drying trajectory of the next five control cycles is predicted based on the microwave-thermodynamic-mass transfer coupling model;

[0016] The rigid constraints of the microwave-thermodynamic-mass transfer coupling model are converted into fuzzy membership functions to dynamically adjust the power partition boundary conditions.

[0017] Optionally, the microwave-thermodynamic-mass transfer coupling model includes:

[0018] Physical field equations centered on material dielectric constant, thermal conductivity, and moisture diffusion rate;

[0019] The dynamic update mechanism of constraint conditions adjusts the equation parameters every 30 seconds according to the weight coefficient of the feedback compensation module.

[0020] Optionally, the feedback compensation module includes:

[0021] Dual-modal learning mechanism: historical data is used to train the model in the early stage of drying, and it switches to real-time deviation feedback mode in the late stage of drying;

[0022] The dynamic learning rate adjustment module automatically adjusts the neural network weight update step size according to the moisture content change rate.

[0023] Optionally, the multi-channel microwave execution unit includes:

[0024] Phase-adjustable magnetron array, supporting 0-360° continuous phase difference control;

[0025] The waveguide power divider has an independent X / Y / Z three-axis adjustment mechanism, and the power distribution error of each channel is ≤1.5%.

[0026] Optionally, the optimization process of the improved particle swarm optimization algorithm includes:

[0027] Taking the field intensity uniformity as the fitness function, the optimal phase-power combination is iteratively calculated;

[0028] The inertia weight decay factor is introduced to narrow the search range to the ±5% interval of the current optimal solution in the later iteration.

[0029] Optionally, the implementation of the three-dimensional gradient field intensity distribution includes:

[0030] Dynamically divide high / medium / low humidity areas based on material deformation parameters;

[0031] Match the microwave field intensity peak area with the spatial coordinates of the high humidity area.

[0032] Optionally, the system further includes:

[0033] The standing wave field reconstruction module reduces microwave reflection loss to below 0.8dB through reflected wave phase compensation;

[0034] Energy efficiency monitoring unit calculates the energy consumption ratio per unit of water evaporation in real time and displays the optimization curve.

[0035] Optionally, the energy efficiency monitoring unit includes:

[0036] Moisture-energy consumption correlation database, storing the optimal energy consumption thresholds for different types of materials;

[0037] Abnormal warning module, when the real-time energy consumption deviates from the preset energy consumption threshold, it triggers an audible and visual alarm and suspends the drying process.

[0038] The present invention significantly improves the reliability of sensor data through high-precision fusion and noise suppression of multi-source data; dynamically optimizes control parameters with the help of intelligent hybrid algorithms to enhance the uniformity and adaptability of the drying process; accurately matches the moisture distribution of materials through three-dimensional gradient field intensity to achieve targeted energy delivery and efficient utilization; combines a closed-loop self-learning mechanism to continuously correct model deviations to ensure the long-term stability of control instructions; utilizes reflection suppression technology to significantly reduce energy loss, and relies on real-time energy efficiency monitoring to intercept process anomalies. The overall system achieves shortened drying cycles, reduced energy consumption, and improved yield rates. It can be widely adapted to a variety of materials with significantly different dielectric properties, providing an efficient, robust, and intelligent solution for industrial microwave drying. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a system architecture diagram of a real-time dynamic optimization system for a microwave drying process provided by one embodiment of the present invention.

[0040] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0041] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0042] The terms used in the embodiments of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The singular forms "a," "an," "the," and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, and unless the context clearly indicates otherwise, "a plurality" generally includes at least two.

[0043] As used herein, the words "if" and "if" may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0044] In addition, the step sequence in the following method embodiments is only an example and not a strict limitation.

[0045] In practice, the server-side device deployed by the microwave drying process real-time dynamic tuning system may consist of one or more devices. The microwave drying process real-time dynamic tuning system can be implemented as a service instance, a virtual machine, or hardware devices. For example, the microwave drying process real-time dynamic tuning system can be implemented as a service instance deployed on one or more devices in a cloud node. Simply put, the microwave drying process real-time dynamic tuning system can be understood as software deployed on a cloud node, providing the microwave drying process real-time dynamic tuning system to each client. Alternatively, the microwave drying process real-time dynamic tuning system can be implemented as a virtual machine deployed on one or more devices in a cloud node. Application software for managing each client is installed in the virtual machine. Alternatively, the microwave drying process real-time dynamic tuning system can be implemented as a server-side device composed of multiple hardware devices of the same or different types, with one or more hardware devices configured to provide the microwave drying process real-time dynamic tuning system to each client.

[0046] In terms of implementation, the microwave drying process real-time dynamic tuning system and the user end are mutually compatible. Specifically, if the microwave drying process real-time dynamic tuning system is an application installed on a cloud service platform, the user end serves as a client that establishes a communication connection with the application. Alternatively, if the microwave drying process real-time dynamic tuning system is implemented as a website, the user end serves as a webpage. Alternatively, if the microwave drying process real-time dynamic tuning system is implemented as a cloud service platform, the user end serves as a mini-program within an instant messaging application.

[0047] like Figure 1 , which is a system architecture diagram of a real-time dynamic optimization system for a microwave drying process provided by an embodiment of the present invention.

[0048] The real-time dynamic optimization system 100 for a microwave drying process according to the present invention can be arranged in a cloud server. In terms of implementation, it can be used as one or more service devices, or it can be installed as an application on the cloud (such as a server of a mobile service operator, a server cluster, etc.), or it can be developed as a website. According to the functions implemented, the real-time dynamic optimization system 100 for a microwave drying process can include a multi-source sensor group 101, a data fusion unit 102, an adaptive decision unit 103, a feedback compensation module 104, and a multi-channel microwave execution unit 105. The module according to the present invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, which are stored in the memory of the electronic device.

[0049] In an embodiment of the present invention, in the real-time dynamic tuning system for a microwave drying process, each of the above modules can be implemented independently and called with other modules. The call here can be understood as a module that can connect to multiple modules of another type and provide corresponding services to the multiple modules connected to it. For example, the sharing evaluation module can call the same information acquisition module to obtain the information collected by the information acquisition module. Based on the above characteristics, in the real-time dynamic tuning system for a microwave drying process provided by an embodiment of the present invention, the scope of application of the real-time dynamic tuning system architecture of the microwave drying process can be adjusted by adding modules and directly calling them without modifying the program code, thereby realizing cluster-type horizontal expansion, so as to achieve the purpose of quickly and flexibly expanding the real-time dynamic tuning system for a microwave drying process. In actual applications, the above modules can be set in the same device or different devices, or they can be set in a virtual device, such as a service instance in a cloud server.

[0050] The following describes the various components and specific workflow of the real-time dynamic optimization system for microwave drying process in conjunction with specific embodiments:

[0051] The multi-source sensor group 101 is used to collect the temperature distribution, microwave power intensity, humidity gradient and material deformation parameters in the drying chamber in real time.

[0052] In the embodiment of the present invention, the multi-source sensor group is the data acquisition front end of the system, which consists of an infrared temperature sensor, a microwave power sensor, a humidity sensor and a material state detection unit, and is used to monitor the multi-dimensional physical parameters in the drying chamber in real time.

[0053] In detail, traditional drying systems often suffer from control deviations due to the one-sidedness of single sensor data. However, the dynamic coupling of temperature, humidity, microwave power and material deformation in the microwave drying process is extremely strong, and accurate perception requires cross-validation of multi-dimensional data.

[0054] Specifically, the infrared temperature sensor uses a non-contact infrared array (such as an 8×8 dot matrix) to scan the temperature distribution of the cavity cross section every 0.5 seconds; the microwave power sensor uses a directional coupler to capture the output power fluctuations of the magnetron in real time (range 0-5kW, accuracy ±0.5%); the humidity sensor uses a high-frequency capacitive probe, embedded under the material tray, to measure the humidity gradient (resolution 0.1% RH); the material state detection unit integrates a machine vision module, and captures deformation characteristics such as cracks and shrinkage on the material surface through a high-speed camera (200fps).

[0055] For example, when drying Chinese medicinal materials, the infrared sensor discovered that the temperature in a certain area rose abnormally to 120°C (the standard process is 95°C). Combined with the humidity sensor, it detected a sudden drop in humidity in the area, determined the risk of local carbonization, and triggered the subsequent control unit to intervene.

[0056] The data fusion unit 102 performs spatiotemporal alignment and noise suppression on the multi-source sensor data based on the Kalman filter algorithm to generate a drying state matrix.

[0057] In the embodiment of the present invention, the data fusion unit is responsible for performing spatiotemporal alignment and noise filtering on multi-source heterogeneous data to generate a standardized matrix representing the drying state.

[0058] Specifically, the differences in sampling frequency (e.g., 0.5 Hz for temperature and 1 Hz for humidity) and spatial resolution of different sensors will lead to data misalignment, and directly using the raw data will cause control command oscillation.

[0059] For example, when drying ceramic blanks, the microwave power sensor produces an instantaneous fluctuation of ±8% due to cavity reflection. After wavelet-Kalman cascade filtering, the signal standard deviation is reduced from 4.2% to 0.7%, ensuring the stability of the subsequent decision-making unit input.

[0060] In the embodiment of the present invention, the spatiotemporal correlation module synchronizes the sampling data of the multi-source sensors through timestamps.

[0061] The spatiotemporal correlation module is the core unit for preprocessing multi-source sensor data. Its function is to align asynchronously sampled data from different sensors (such as temperature, humidity, microwave power, and material deformation) to a unified time base and spatial coordinate system through precise timestamps and interpolation algorithms. This module addresses data misalignment caused by sampling frequency differences, transmission delays, or clock asynchrony across multiple sensors, ensuring that the data received by subsequent processing units is temporally consistent and spatially correlated.

[0062] In detail, in the microwave drying system, the data characteristics of each sensor are significantly different:

[0063] Differences in sampling frequency: For example, an infrared temperature sensor might capture a temperature distribution image at 0.5 Hz (once every 2 seconds), while a microwave power sensor records power fluctuations at 10 Hz (10 times per second).

[0064] Transmission delay: The image data of material deformation detected by machine vision may be delayed by 50ms due to transmission bandwidth limitations, while the data of capacitive humidity sensors is more real-time.

[0065] Spatial resolution difference: The 8×8 dot matrix data of the temperature sensor and the single-point measurement of the humidity sensor need to be spatially aligned.

[0066] If the raw data is used directly, the drying state matrix will contain noise caused by temporal and spatial mismatches, causing the adaptive decision-making unit to generate incorrect control instructions. For example, if the temperature data lags behind the humidity data by 3 seconds, the system may misjudge the drying stage and incorrectly increase the microwave power, causing carbonization of the material.

[0067] Furthermore, each sensor data packet is timestamped with a high-precision timestamp (using GPS or IEEE1588 protocol, with an accuracy of ±1ms). For example, each frame of an infrared temperature sensor is marked with the UTC time of acquisition (e.g., 12:00:00.000). Simultaneously, low-frequency data (e.g., a humidity sensor at 10Hz) is resampled to 10Hz using linear or spline interpolation, using the highest-frequency sensor (e.g., a microwave power sensor at 10Hz) as a benchmark. For example, the raw humidity data at t=0s, 1s, and 2s are 50%, 48%, and 45%, respectively. Interpolation generates intermediate values at t=0.1s, 0.2s, and 0.9s.

[0068] Specifically, spatially distributed sensors (such as infrared temperature arrays) are mapped to the 3D model of the drying chamber through coordinate transformation. For example, the pixel coordinates (i, j) of an 8×8 temperature array are converted to the chamber coordinates (x, y, z = 0.5 m).

[0069] For example, in the case of drying seaweed, the infrared sensor detected a temperature of 95°C at coordinates (0.2m, 0.3m) at t = 10.0s, while the humidity sensor measured a humidity of 12% at t = 10.2s. The spatiotemporal correlation module interpolated and inferred that the humidity should have been 12.5% at t = 10.0s. It then aligned the two data points to the same time point to ensure the accuracy of the drying state matrix.

[0070] In the embodiment of the present invention, the noise suppression module adopts a cascade structure of wavelet transform and Kalman filtering to filter out carbonization noise of temperature distribution and electromagnetic interference of microwave power intensity.

[0071] In detail, the noise suppression module is the key link in data cleaning. It adopts the cascade structure of "wavelet transform + Kalman filter" to filter out specific noise in the sensor signal in stages.

[0072] In detail, the wavelet transform extracts and eliminates the high-frequency carbonization noise (caused by local overheating and carbonization of the material) in the infrared temperature signal through multi-scale decomposition.

[0073] In detail, the Kalman filter suppresses random fluctuations in the microwave power signal by using state prediction and observation correction to detect low-frequency electromagnetic interference (from the magnetron or external equipment).

[0074] Specifically, the carbonization noise characteristic refers to the fact that local carbonization of the material will cause transient spikes in the temperature signal (such as soaring from 90°C to 150°C in 1 second). Traditional low-pass filters will smooth out the actual temperature changes, resulting in control delays.

[0075] Specifically, electromagnetic interference characteristics refer to the microwave power signal being affected by cavity reflection and power grid fluctuations, which manifests as low-frequency random drift (such as slow fluctuations of ±5%), and the real power trend needs to be dynamically tracked.

[0076] In detail, the wavelet transform decarbonization noise includes the following steps:

[0077] The temperature signal is decomposed into five layers using Daubechies5 wavelet basis to separate high-frequency detail coefficients (D1-D5) and low-frequency approximation coefficients (A5);

[0078] Hard threshold processing is applied to the D1-D3 layers (corresponding to frequencies > 10 Hz), and coefficients with amplitudes less than 0.5°C are removed to retain the true temperature slow variation characteristics;

[0079] The processed coefficients are used to reconstruct the temperature signal and eliminate the carbonization peak.

[0080] For example, when the infrared sensor detects that the temperature at a certain point jumps from 95°C to 130°C within 0.1 second (carbonization noise), the wavelet transform identifies that the mutation is located in the D1 layer (high frequency band). After threshold elimination, the reconstructed signal shows that the temperature rises steadily to 98°C.

[0081] In detail, Kalman filtering to suppress electromagnetic interference includes the following steps:

[0082] Assume that the microwave power P(k) varies linearly with time:

[0083] P(k)=P(k-1)+ΔT·v+w(k)

[0084] Where v is the rate of change of power and w(k) is the process noise.

[0085] In addition, the sensor measurement value Z(k)=P(k)+r(k), where r(k) is the observation noise.

[0086] In detail, the optimal power value is estimated through a prediction-correction cycle.

[0087] For example, the raw data of the microwave power sensor fluctuates as [3.0kW, 3.2kW, 2.8kW, 3.1kW] within 10 seconds, and the output after Kalman filtering is [3.0kW, 3.05kW, 3.02kW, 3.04kW], effectively smoothing out the random fluctuations.

[0088] For example, when drying instant coffee powder, the microwave power sensor produces ±8% fluctuation (2.4kW~2.6kW) due to standing wave reflection in the cavity, which is stabilized at 2.5kW±0.05kW after Kalman filtering; at the same time, the carbonization spike falsely reported by the infrared sensor (caused by local overheating of coffee oil) is filtered out by wavelet transform, avoiding the system from falsely triggering an emergency shutdown and ensuring continuous production.

[0089] In summary, the spatiotemporal correlation and noise suppression modules provide highly reliable input data for subsequent decision-making units through precise temporal-spatial synchronization and graded noise filtering. This combination of technologies not only solves the challenge of integrating multi-source, heterogeneous data but also significantly improves the system's robustness under complex operating conditions, laying the data foundation for real-time dynamic tuning.

[0090] The adaptive decision unit 103 adopts a hybrid algorithm of model predictive control and fuzzy logic to generate three-dimensional control instructions including microwave emission frequency, power partition and drying time according to the drying state matrix.

[0091] In an embodiment of the present invention, the drying state matrix is converted into multi-dimensional control instructions based on a hybrid algorithm of model predictive control and fuzzy logic.

[0092] In detail, single model predictive control is prone to divergence due to model mismatch in strongly nonlinear systems, while fuzzy logic lacks long-term prediction capabilities. The hybrid architecture can balance robustness and accuracy.

[0093] For example, when drying instant coffee powder, the material's wave absorption characteristics drop sharply from ε = 50 to ε = 15 as the moisture content decreases. Traditional MPC causes overheating due to fixed model parameters. The hybrid algorithm controls temperature fluctuations within ±2°C through fuzzy power constraints.

[0094] In an embodiment of the present invention, the hybrid algorithm includes:

[0095] The drying trajectory of the next five control cycles is predicted based on the microwave-thermodynamic-mass transfer coupling model;

[0096] The rigid constraints of the microwave-thermodynamic-mass transfer coupling model are converted into fuzzy membership functions to dynamically adjust the power partition boundary conditions.

[0097] Specifically, the microwave-thermodynamic-mass transfer coupling model is a multi-physics mathematical model that quantitatively describes the dynamic coupling of microwave energy absorption, heat transfer, and water evaporation by integrating electromagnetic, temperature, and moisture migration equations. Using the material's dielectric constant (ε), thermal conductivity (k), and moisture diffusion rate (D) as core parameters, the model constructs a system of partial differential equations to predict the temperature distribution, humidity gradient, and material moisture content within the drying chamber over the next 150 seconds (five control cycles, each 30 seconds).

[0098] Specifically, traditional drying control relies on empirical rules or a single physical field model, which cannot accurately characterize the nonlinear interaction of microwave energy, heat, and mass transfer: the dielectric constant ε of the material decays exponentially with decreasing moisture content (for example, from ε = 80 to ε = 10), resulting in dynamic changes in the microwave penetration depth; the speed at which heat is conducted from the surface to the interior of the material (determined by k) may lag behind the microwave heating rate, causing local overheating; and the moisture diffusion rate D is affected by both the temperature gradient and the pore structure, making it difficult to approximate with a linear equation.

[0099] Furthermore, without a coupled model, the prediction results will deviate significantly from the actual operating conditions. For example, when drying fruits and vegetables with high moisture content, ignoring the feedback of moisture migration on dielectric properties can lead to moisture content errors exceeding 20%.

[0100] Specifically, fuzzy membership functions are used to quantify the flexible boundaries of control constraints, transforming traditional "hard constraints" (such as "power ≤ 3kW") in model predictive control into "soft constraints" that allow for temporary violations. For example, when a sudden change in the dielectric properties of a material is detected, the power partition boundary can be dynamically expanded by 10%, avoiding control instability caused by strict limiting.

[0101] Specifically, the rigid constraints of model predictive control can lead to deterioration in system performance in the following scenarios: when the dielectric constant of the material deviates from the preset value due to batch differences, the fixed power limit may not be able to maintain the necessary temperature; when local carbonization of the material causes a temperature spike, forcing the power to drop sharply may disrupt the overall drying uniformity. In the trade-off between energy efficiency and drying speed, short-term power overlimits must be allowed to prioritize quality assurance.

[0102] For example, when drying instant coffee powder (ε drops sharply from 50 to 15), if the forced power is ≤3kW, the temperature in the central area will drop below 80°C, causing drying to stagnate; however, the fuzzy constraint allows the power to be temporarily increased to 3.3kW to compensate for dielectric losses.

[0103] For example, the power partition constraint is defined as a triangular membership function. When the membership degree μ ≥ 0.6, the constraint is considered to be "basically satisfied" and the control instruction is allowed to go out of bounds.

[0104] Furthermore, in the control instruction generation, the hard-constrained optimization problem is converted into a multi-objective optimization with membership weights.

[0105] In an embodiment of the present invention, the microwave-thermodynamic-mass transfer coupling model includes:

[0106] Physical field equations centered on material dielectric constant, thermal conductivity, and moisture diffusion rate;

[0107] The dynamic update mechanism of constraint conditions adjusts the equation parameters every 30 seconds according to the weight coefficient of the feedback compensation module.

[0108] In detail, the physics equations are constructed as follows:

[0109] Microwave field energy distribution modeling accurately describes the distribution of electromagnetic energy within and around the material by simulating the propagation, reflection, and absorption of microwaves within the drying chamber. Its core is to determine the intensity of microwave power density at different locations to guide targeted energy delivery.

[0110] Temperature field modeling quantifies the process of heat transfer within a material, including the rate at which microwave energy is converted to thermal energy, the delay effect of heat conduction from the surface to the interior, and the heat exchange between the material and the surrounding atmosphere. Microwave heating inherently operates from the inside out, but variations in thermal conductivity within materials can lead to uneven temperature distribution. For example, when drying high-density ceramic blanks, rapid surface heat dissipation and internal heat accumulation can lead to cracking. Accurate temperature field modeling can predict and mitigate these risks.

[0111] Specifically, mass transfer field modeling describes the dynamic process of moisture migration from the interior of a material to its surface and then evaporation into the air. This involves multiple mechanisms, including pore diffusion, capillary action, and vapor pressure gradients. The rate of moisture migration is nonlinearly affected by factors such as temperature and material structure. For example, moisture primarily diffuses in the liquid phase during the initial drying phase, then transitions to gaseous evaporation in the later stages, a process that cannot be accurately predicted by a single traditional model. Constructing a mass transfer field allows for identification of drying stages and optimization of moisture control strategies.

[0112] Specifically, multi-physics coupling integrates the interactions of microwave, temperature, and mass transfer fields, revealing the closed-loop relationship of "energy input-heat transfer-water migration." For example, microwave energy influences temperature distribution, temperature gradients drive water migration, and changes in water content in turn affect microwave absorption efficiency. Single-physics models cannot account for these complex interactions. For example, in fruit and vegetable drying, water loss causes a decrease in dielectric constant, reducing microwave absorptivity. Adjusting power alone while ignoring dielectric changes will lead to a sharp drop in drying efficiency.

[0113] Furthermore, microwave field distribution data is input into the temperature field model to calculate the local temperature rise; temperature data is input into the mass transfer field model to predict the water evaporation rate; and the moisture distribution data output by the mass transfer field is used to correct the dielectric parameters of the microwave field, forming a closed-loop optimization.

[0114] The coupled equations were further discretized using the finite element method, with a time step of 30 seconds and a spatial grid size of ≤5 mm. Initial boundary conditions were provided by the dry state matrix, and the moisture content (M) and temperature (T) distributions for the next five steps were iteratively calculated for each cycle.

[0115] The feedback compensation module 104 uses an online error back propagation model to dynamically correct the weight coefficient of the microwave-thermodynamic-mass transfer coupling model by taking the deviation between the final moisture content detection value of the material and the model prediction value as a loss function.

[0116] In the embodiment of the present invention, the model weights are dynamically corrected through an online BP neural network to achieve adaptive iteration of the process model.

[0117] Specifically, the dynamic update mechanism adjusts key parameters of the coupling model (such as ε, k, and D) every 30 seconds using weight coefficients provided by the feedback compensation module, ensuring that the model consistently tracks changes in the material's real-time properties. The weight coefficients are calculated using an online back-propagation (BP) neural network and reflect the deviation distribution between the model's predicted values and the actual measured values.

[0118] In detail, the dielectric and thermodynamic properties of the material show time-varying characteristics during the drying process: the dielectric constant decays exponentially with the decrease of moisture content M; the thermal conductivity changes due to the shrinkage of the material porosity; and the water diffusion rate is affected by the temperature gradient.

[0119] Specifically, static models trained offline cannot adapt to differences in material batches (such as natural fluctuations in the moisture content of agricultural products), and real-time feedback can solve the problem of model drift.

[0120] In detail, the bimodal learning mechanism refers to pre-training the network using historical data (100 sets of process parameters) in the early stage of drying (moisture content > 30%), and switching to online learning in the late stage (moisture content < 10%), with weights updated every 5 seconds.

[0121] In detail, the initial learning rate is set to 0.01, and when the moisture content change rate is detected to be >1% / min (sensitive stage), it is automatically reduced to 0.001 to prevent overshoot.

[0122] For example, when processing apple slices from different origins, the initial model's prediction error due to differences in fruit gum content reached 8%. After feedback compensation, the error converged to 1.2%, and the drying time was shortened by 15%.

[0123] In an embodiment of the present invention, the feedback compensation module includes:

[0124] Dual-modal learning mechanism: historical data is used to train the model in the early stage of drying, and it switches to real-time deviation feedback mode in the late stage of drying;

[0125] The dynamic learning rate adjustment module automatically adjusts the neural network weight update step size according to the moisture content change rate.

[0126] Specifically, the dual-modal learning mechanism is the core strategy of the feedback compensation module. It addresses the problem of data characteristic differences at different stages of the drying process by adjusting the learning mode of the neural network in stages:

[0127] Initial stage of drying (mode I): When the moisture content of the material is higher than the critical value (such as 30%), the historical process data is used to pre-train the neural network model and establish the basic weight parameters.

[0128] At the end of drying (mode II): when the moisture content drops below the critical value, it switches to the real-time deviation feedback mode and updates the model weights online based on the actual test data of the current batch of materials.

[0129] Specifically, the material state changes dramatically in the initial drying stage (0-30% moisture content), the amount of real-time data is insufficient, and direct online learning can easily lead to model overfitting; at the end of drying (<30% moisture content), the moisture migration rate drops sharply and the dielectric properties suddenly change, requiring high-sensitivity real-time correction; historical data covers the statistical characteristics of multiple batches of materials, which can improve the model generalization ability.

[0130] In detail, the mode switching logic includes:

[0131] Main trigger condition: real-time moisture content drops to a preset threshold (30% by default, configurable);

[0132] Auxiliary calibration: The moisture content fluctuation is less than ±1% within 5 consecutive sampling cycles. Ensure that the state is stable before switching to avoid false operation.

[0133] Specifically, 1,000 sets of typical drying cases were extracted from the process database, including different material types (dielectric constant 5-80), ambient temperature and humidity, microwave power configuration, etc.; a batch training mode was adopted to fuse multi-scene features through weighted averaging to prevent the model from being biased towards a specific type of material.

[0134] Specifically, the actual moisture content detection value is collected every 5 seconds and compared with the model prediction value to calculate the deviation; when the deviation exceeds the allowable range (such as >2%), the weight update is triggered, and the adjustment range is controlled by the dynamic learning rate module.

[0135] Specifically, when processing Chinese medicinal materials from different origins, the traditional single-model learning model fails to adaptively adjust model weights due to differences in initial moisture content (45% for Astragalus root and 28% for Angelica sinensis), resulting in a 12% prediction error at the end of the drying process. The dual-modal mechanism eliminates batch differences through initial pre-training and then fine-tunes the model at the end.

[0136] The multi-channel microwave execution unit 105 optimizes the microwave energy ratio based on the improved particle swarm algorithm, and generates a three-dimensional gradient field intensity distribution that matches the material morphology according to the three-dimensional control instructions.

[0137] In an embodiment of the present invention, the control instructions are converted into physical field regulation, and a three-dimensional gradient microwave field is generated through a magnetron array and a waveguide distributor.

[0138] In detail, a uniform microwave field can easily lead to overheating at the edge, while a gradient field can achieve on-demand energy distribution and match the moisture distribution of the material.

[0139] Specifically, the phase-adjustable magnetron array includes six groups of magnetrons (2.45 GHz, 0-5 kW adjustable) that generate a 0-360° phase difference through a phase shifter. For example, the phase difference between adjacent magnetrons is set to 120° to form a rotating standing wave field.

[0140] In detail, the improved particle swarm optimization algorithm takes field strength uniformity as the objective function, introduces inertia weight decay (from 0.9 to 0.4) after 50 iterations, focuses the search range to the ±5% neighborhood of the optimal solution, and compresses the calculation time from 120 seconds to 18 seconds.

[0141] For example, when drying seaweed sheets, the traditional uniform field results in an edge carbonization rate of 12%, while the gradient field intensity optimizes the center-edge energy flux density ratio from 1:1.8 to 1:1.2, and the carbonization rate drops to 3%.

[0142] In an embodiment of the present invention, the multi-channel microwave execution unit includes:

[0143] Phase-adjustable magnetron array, supporting 0-360° continuous phase difference control;

[0144] The waveguide power divider has an independent X / Y / Z three-axis adjustment mechanism, and the power distribution error of each channel is ≤1.5%.

[0145] Specifically, the phase-tunable magnetron array consists of multiple groups of independently controlled magnetrons, each of which can achieve continuous phase adjustment from 0° to 360° via electronic phase shifters. This design is similar to the sonic interference of different instruments in a symphony orchestra. By precisely controlling the phase superposition of microwave signals, energy is focused or dispersed in a specific direction within the drying chamber.

[0146] Specifically, traditional fixed-phase magnetrons generate only a fixed microwave pattern, unable to adapt to changes in the material's moisture distribution. For example, if the material shrinks, shifting the high-humidity area, the fixed phase will cause the microwave energy to continue acting in that location, resulting in uneven drying. The adjustable phase design allows for dynamic tracking of moisture changes, achieving the goal of "energy chasing moisture."

[0147] Specifically, a PIN diode phase shifter receives the phase parameters from the three-dimensional control instructions and adjusts the output phase of each magnetron group in real time. For example, if the instruction requires a 90° phase difference between adjacent magnetrons, a rotating microwave field is generated, causing the energy to periodically scan the drying cavity.

[0148] Specifically, an electromagnetic shielding isolation layer is set between the magnetrons to reduce the cross-coupling interference to below -40dB and ensure the phase control accuracy of ±2°.

[0149] Specifically, the waveguide power divider is a mechanical-electronic hybrid device that distributes total microwave power to different spatial regions according to preset proportions through independent X / Y / Z axis adjustment mechanisms. Its core function is to convert the "power partitioning" parameters in the control instructions into physical field distributions, ensuring that the power error in each channel is ≤1.5%.

[0150] Specifically, the energy efficiency of microwave drying is highly dependent on the accuracy of power distribution. For example, if a high-humidity area requires 70% power but is actually distributed 65%, drying time will be extended by 15%. If it is distributed 75%, localized carbonization may occur. A 1.5% error control balances efficiency and safety.

[0151] Specifically, the three-axis adjustment mechanism uses a stepper motor with a precision of 0.01° to move the impedance matching module in the waveguide according to control instructions, changing the energy proportion of each channel; the built-in power sensor monitors the output of each channel in real time. When a deviation >1% is detected, the fine-tuning motor is triggered to compensate for the error.

[0152] In an embodiment of the present invention, the optimization process of the improved particle swarm optimization algorithm includes:

[0153] Taking the field intensity uniformity as the fitness function, the optimal phase-power combination is iteratively calculated;

[0154] The inertia weight decay factor is introduced to narrow the search range to the ±5% interval of the current optimal solution in the later iteration.

[0155] Specifically, the improved particle swarm optimization (PSO) algorithm simulates the foraging behavior of bird flocks and searches for the optimal combination in the multidimensional parameter space of microwave frequency, phase, and power to maximize field intensity uniformity. Its core improvement lies in the introduction of an inertia weight decay factor, which enables the algorithm to focus on the area near the optimal solution in the later stages of the iteration, avoiding premature convergence.

[0156] In detail, traditional PSO has two major defects in microwave optimization: early iterations may waste resources exploring invalid areas; later iterations are prone to falling into suboptimal solutions.

[0157] In detail, the improved algorithm balances exploration and development efficiency by dynamically shrinking the search range.

[0158] Specifically, field strength uniformity = 1-(maximum field strength-minimum field strength) / average field strength, with the target value approaching 1. The initial weight is set to 0.9, decaying by 5% every 10 iterations and dropping to 0.4 after 50 iterations, forcing the particle swarm to converge toward the current optimal solution. 10% of the optimal particles are retained in each generation to prevent the loss of high-quality solutions.

[0159] In an embodiment of the present invention, the implementation of the three-dimensional gradient field intensity distribution includes:

[0160] Dynamically divide high / medium / low humidity areas based on material deformation parameters;

[0161] Match the spatial coordinates of the microwave field intensity peak area with the high humidity area.

[0162] Specifically, the 3D gradient field intensity distribution generates a non-uniform microwave field within the drying chamber based on real-time material deformation and humidity data, precisely spatially aligning areas of high field intensity with areas of high humidity. Essentially, this achieves targeted drying through spatial mapping of "energy-humidity."

[0163] Specifically, a uniform microwave field can lead to uneven drying and flooding—hotspots form at the edges of the drying cavity due to microwave reflection, while the center lacks energy. A gradient field intensity allows for dynamic energy focusing, ensuring that microwave energy precisely covers moisture-rich areas.

[0164] In detail, the humidity areas are divided as follows:

[0165] High humidity area: moisture content > 40%, marked in red;

[0166] Medium humidity zone: 20%-40%, marked in yellow;

[0167] Low humidity area: <20%, marked in green.

[0168] In detail, the field strength matching rules are as follows:

[0169] The red area is allocated a field strength of 500W / m 2 ;

[0170] Yellow area 300W / m 2 ;

[0171] Green area 100W / m 2 .

[0172] Furthermore, when the material shrinks and causes the coordinates to shift, the field intensity distribution coordinates are corrected according to the deformation ratio. For example, if the material shrinks by 5%, the field intensity peak area will move inward by 5%.

[0173] In the embodiment of the present invention, the standing wave field reconstruction module reduces the microwave reflection loss to below 0.8 dB by compensating the reflected wave phase.

[0174] Specifically, the standing wave field reconstruction module is an intelligent control system that dynamically compensates for the phase of microwave reflection waves to suppress energy reflection loss. Its core function is to detect the phase and amplitude of the reflected wave within the drying cavity in real time and adjust the phase of the transmitted wave inversely to create destructive interference at the cavity port. This reduces reflection loss from the industry-standard 2-3dB to below 0.8dB.

[0175] In detail, cavity standing waves can cause more than 30% of energy to be reflected, which not only wastes energy but also may damage the magnetron.

[0176] Specifically, a PIN diode phase shifter is installed at the end of the waveguide to detect the phase of the reflected wave in real time (with an accuracy of ±5°), and the phase of the transmitted wave is dynamically adjusted to make it in antiphase with the reflected wave, thus achieving interference cancellation.

[0177] In the embodiment of the present invention, the energy efficiency monitoring unit calculates the energy consumption ratio per unit water evaporation in real time and displays an optimization curve.

[0178] Specifically, the energy efficiency monitoring unit is the system's core data analysis platform. It calculates the energy consumption per unit of water evaporation in real time and, combined with historical data, generates energy efficiency optimization curves, providing a quantitative basis for process improvement. It consists of two submodules: a moisture-energy consumption correlation database and an anomaly warning module.

[0179] Specifically, the optimal energy consumption of different materials (such as wood and medicinal materials) varies significantly, and dynamic monitoring is required to prevent over-consumption.

[0180] In detail, the multi-level response mechanism includes:

[0181] Level 1 warning (deviation of 15%-20%): Triggers a yellow warning light and buzzer, and an optimization suggestion pops up on the HMI interface;

[0182] Level 2 shutdown (deviation > 20%): The red alarm flashes, the microwave power is immediately cut off, and emergency air cooling is started;

[0183] For example, during the drying process of certain medicinal materials, condensation on the humidity sensor caused the detection value to be falsely low (the moisture content was displayed as 25%, but the actual value was 38%). The system detected that the energy consumption ratio had abnormally increased to 1.8kWh (threshold 1.4kWh), triggering a second-level shutdown. The fault analysis report accurately located the sensor failure and avoided the expansion of material carbonization losses.

[0184] In an embodiment of the present invention, the energy efficiency monitoring unit includes:

[0185] Moisture-energy consumption correlation database, storing the optimal energy consumption thresholds for different types of materials;

[0186] Abnormal warning module, when the real-time energy consumption deviates from the preset energy consumption threshold, it triggers an audible and visual alarm and suspends the drying process.

[0187] Specifically, the moisture-energy consumption correlation database is a knowledge base that stores the optimal energy consumption thresholds for different materials. It covers correlated data in over 300 dimensions, including dielectric properties, thermodynamic parameters, and process constraints. It supports fuzzy matching and adaptive recommendations. For example, the standard energy consumption for wood is 1.2 kWh / kg-H2O. If the real-time value exceeds 1.5 kWh, an alert is triggered.

[0188] Furthermore, after each drying batch is completed, the actual energy consumption is automatically associated with the material characteristics and stored, and the database is continuously optimized; manual marking of special cases is supported (such as high-sugar materials requiring additional energy consumption compensation).

[0189] In detail, the abnormal warning module adopts the sliding t-test algorithm. When the energy consumption of three consecutive sampling points deviates from the threshold by 15%, it is determined to be abnormal and the shutdown protection is initiated.

[0190] In detail, the abnormal warning module is a safety protection system that triggers a graded response when energy efficiency is abnormal by comparing actual energy consumption with database thresholds in real time. It includes a three-stage response including sound and light alarm, process suspension, and root cause analysis.

[0191] Specifically, threshold deviation detection refers to the real-time calculation of the energy consumption ratio, and an abnormality is determined when five consecutive sampling points exceed the threshold by ±15%; a sliding window algorithm is used to eliminate instantaneous interference.

[0192] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0193] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to achieve optimal results.

[0194] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A real-time dynamic tuning system for microwave drying process, characterized in that: The system includes a multi-source sensor group, a data fusion unit, an adaptive decision unit, a feedback compensation module, and a multi-channel microwave execution unit, wherein: Multi-source sensor group, used to collect temperature distribution, microwave power intensity, humidity gradient and material deformation parameters in the drying chamber in real time; The data fusion unit performs spatiotemporal alignment and noise suppression on multi-source sensor data based on the Kalman filter algorithm to generate a drying state matrix; The adaptive decision-making unit uses a hybrid algorithm of model predictive control and fuzzy logic to generate three-dimensional control instructions including microwave emission frequency, power partition and drying time according to the drying state matrix; The feedback compensation module uses an online error back propagation model to take the deviation between the final moisture content of the material and the model prediction value as a loss function to dynamically correct the weight coefficient of the microwave-thermodynamic-mass transfer coupling model; The multi-channel microwave execution unit optimizes the microwave energy ratio based on the improved particle swarm algorithm, and generates a three-dimensional gradient field intensity distribution that matches the material morphology according to the three-dimensional control instructions.

2. The real-time dynamic optimization system for microwave drying process according to claim 1, characterized in that: The data fusion unit further comprises: The spatiotemporal correlation module synchronizes the sampling data of multiple source sensors through timestamps; The noise suppression module adopts the cascade structure of wavelet transform and Kalman filter to filter out the carbonization noise of temperature distribution and electromagnetic interference of microwave power intensity.

3. The real-time dynamic optimization system for microwave drying process according to claim 1, characterized in that: The hybrid algorithm includes: The drying trajectory of the next five control cycles is predicted based on the microwave-thermodynamic-mass transfer coupling model; The rigid constraints of the microwave-thermodynamic-mass transfer coupling model are converted into fuzzy membership functions to dynamically adjust the power partition boundary conditions.

4. The real-time dynamic optimization system for microwave drying process according to claim 3, characterized in that: The microwave-thermodynamic-mass transfer coupling model includes: Physical field equations centered on material dielectric constant, thermal conductivity, and moisture diffusion rate; The dynamic update mechanism of constraint conditions adjusts the equation parameters every 30 seconds according to the weight coefficient of the feedback compensation module.

5. The real-time dynamic optimization system for microwave drying process according to claim 1, characterized in that: The feedback compensation module includes: Dual-modal learning mechanism: historical data is used to train the model in the early stage of drying, and it switches to real-time deviation feedback mode in the late stage of drying; The dynamic learning rate adjustment module automatically adjusts the neural network weight update step size according to the moisture content change rate.

6. The real-time dynamic optimization system for microwave drying process according to claim 1, characterized in that: The multi-channel microwave execution unit includes: Phase-adjustable magnetron array, supporting 0-360° continuous phase difference control; The waveguide power divider has an independent X / Y / Z three-axis adjustment mechanism, and the power distribution error of each channel is ≤1.5%.

7. The real-time dynamic optimization system for microwave drying process according to claim 6, characterized in that: The optimization process of the improved particle swarm algorithm includes: Taking the field intensity uniformity as the fitness function, the optimal phase-power combination is iteratively calculated; The inertia weight decay factor is introduced to shrink the search range to the ±5% interval of the current optimal solution in the later iteration.

8. The real-time dynamic optimization system for microwave drying process according to claim 1, characterized in that: The realization of the three-dimensional gradient field intensity distribution includes: Dynamically divide high / medium / low humidity areas based on material deformation parameters; Match the microwave field intensity peak area with the spatial coordinates of the high humidity area.

9. The real-time dynamic optimization system for microwave drying process according to any one of claims 1 to 8, characterized in that: Also includes: The standing wave field reconstruction module reduces microwave reflection loss to below 0.8dB through reflected wave phase compensation; Energy efficiency monitoring unit calculates the energy consumption ratio per unit of water evaporation in real time and displays the optimization curve.

10. The real-time dynamic optimization system for microwave drying process according to claim 9, characterized in that: The energy efficiency monitoring unit includes: Moisture-energy consumption correlation database, storing the optimal energy consumption thresholds for different types of materials; Abnormal warning module, when the real-time energy consumption deviates from the preset energy consumption threshold, it triggers an audible and visual alarm and suspends the drying process.

Images