Multi-stage flexible intelligent drying method and system for drying meat products

Abstract

The application discloses a multi-stage flexible intelligent drying method and system for drying meat products, and belongs to the field of food drying processing. In order to solve the problem that the existing drying process cannot adapt to the difference in physical properties of raw meat and the instability of product quality, the application constructs a drying kinetics model with the variety, physical property parameters and drying target moisture content of raw meat as input, and with the adjustment of infrared wavelength and / or infrared power, infrared heating temperature, infrared radiation distance, hot air speed, drying time and dehumidification strategy parameters in the multi-stage infrared combined hot air drying program as output; through digital twin model pre-training, real test data transfer learning fine-tuning, an accurate model is obtained; the physical property parameters of the raw meat to be dried are collected, and the drying program output by the model is executed. The application can intelligently adapt to different raw meat to improve drying uniformity and nutrient retention rate, and is suitable for industrialized flexible drying of meat products.

Description

Technical Field

[0001] This invention relates to the field of food drying and processing. More specifically, this invention relates to a multi-stage flexible intelligent drying method and system for drying meat products. Background Technology

[0002] Dried meat is a traditional specialty meat product in my country, represented by beef jerky, pork jerky, mutton jerky, and chicken jerky. It holds an important position in the snack and tourist food markets due to its rich nutrition, strong flavor, long shelf life, and portability. The drying process is the most critical step in dried meat processing, and also the most time-consuming and energy-intensive. The quality of the drying process directly determines the uniformity of moisture content, texture, taste, nutrient retention, and flavor of the dried meat product. However, current industrial-scale drying technology for dried meat still faces common industry challenges such as low drying efficiency, unstable quality, and the inability to precisely adapt various types of dried meat to the characteristics of different meat types. These challenges severely restrict the transformation and upgrading of the traditional dried meat industry towards high efficiency, high quality, and high added value.

[0003] Currently, hot air drying is the most widely used drying method in the industrial production of dried meat. Hot air drying relies on convection heat transfer of hot air, with heat conducted from the surface of the material to its interior. It has advantages such as low equipment cost and simple operation. However, this method has significant inherent drawbacks: due to slow heat transfer and long moisture migration paths, the drying cycle is typically 8–12 hours; prolonged high-temperature heating easily leads to rapid evaporation of surface moisture, forming a hard crust that hinders the migration of internal moisture, resulting in uneven dehydration (dry on the outside, wet on the inside); simultaneously, prolonged high-temperature treatment also causes excessive denaturation of muscle protein, significant loss of heat-sensitive B vitamins and other nutrients (loss rate can reach 25%–40%), and volatilization of flavor compounds, resulting in a hard texture, poor chewiness, and a severe decline in nutritional and flavor quality.

[0004] To overcome the limitations of single hot air drying, researchers have attempted to introduce infrared radiation technology into the field of meat drying. Chinese patent CN103385279B discloses a method for drying meat products. This method uses both 6.1–6.3 μm far-infrared and 2.8–3.3 μm mid-infrared radiation for heating, supplemented by hot air convection, to reduce the moisture content of meat strips from the initial value to 40% in two steps. Its drying speed is approximately twice that of traditional hot air drying, and energy consumption is reduced by about 30%. However, in this method, regardless of whether it is beef, pork, or other livestock and poultry meat, the key process parameters such as the far-infrared and mid-infrared wavelength ranges, radiation heating temperature, and hot air velocity are all fixed and uniform values. No differentiated adjustments are made based on the significant differences in the inherent physical properties of different types of raw meat, such as fat content, fiber structure, protein content, and water retention. For pork with high fat content, a uniform heating temperature may cause fat to melt, surface oil to appear, and oxidation and rancidity. For beef with coarse muscle fibers and high water retention, a uniform hot air velocity may not be able to effectively establish the driving force for the migration of internal moisture to the outside, resulting in a low dehydration rate and uneven drying. In actual production, it is difficult to meet the quality requirements of different varieties of dried meat.

[0005] To further improve the segmented control capability of the drying process, Chinese patent CN104054798A discloses a meat product drying device. This device divides the internal space of the shell into a drying zone and a cooling zone along the longitudinal axis. The drying zone is further divided along the longitudinal axis into a first drying zone (pre-dehydration zone), a second drying zone (dehydration zone), and a third drying zone (forced dehydration zone). Multiple infrared generators are installed along the longitudinal axis in each drying zone to heat the meat products to be dried in segments, achieving continuous production in conjunction with a conveying device. This method structurally achieves segmented control of the drying process and adjusts the heating conditions of each segment by differentiating the configuration of infrared generators in different zones, thus improving drying efficiency to some extent. However, although the device divides the drying process into three zones, the infrared heating parameters of each zone remain preset fixed values. Its segmented control is based solely on the position of the material in the conveying direction, without considering the differences in muscle tissue structure between different types of raw meat, or multi-dimensional physical properties such as fat percentage, fiber content, protein content, initial moisture content, and target moisture content as input variables to drive the dynamic generation of drying program parameters. When switching between different types of raw meat on the production line, operators still need to manually readjust the infrared heating parameters of each area based on their experience. This not only results in a long debugging cycle and high switching costs, but also makes it difficult to guarantee the consistency of product quality between batches.

[0006] In summary, existing meat drying technologies generally suffer from the following shortcomings: The application of digital twin models in the food drying field is lacking. The setting of drying process parameters relies heavily on human experience, lacking the technical means to establish a quantifiable mapping relationship and intelligently match between the multidimensional physical properties of raw meat (fat percentage, fiber content, protein content, initial moisture content, target moisture content) and drying program parameters. Furthermore, it is impossible to automatically generate targeted mid-infrared-hot air synergistic drying programs based on the intrinsic physical property differences of different varieties of raw meat. This leads to quality problems such as uneven drying, excessively hard texture, nutrient loss, and flavor distortion when multiple varieties of meat are produced on the same production line, making it difficult to achieve high-quality, standardized, and flexible industrial production of multiple varieties of meat. In addition, although some studies in recent years have attempted to use data-driven intelligent models to predict or optimize drying process parameters, the training of such models usually requires large-scale real-world drying test data covering different varieties and a wide range of physical property combinations as labeled samples. However, real-world infrared-hot air synergistic drying experiments not only involve high material costs and long single-experiment cycles of several hours or even more than ten hours, but also suffer from natural fluctuations in physical properties such as fat percentage, fiber content, and initial moisture content of the raw meat due to different batches of raw materials. This makes it difficult to obtain a sufficient number of comprehensive and accurately labeled training datasets within limited time and cost. This fundamentally limits the training effect and generalization ability of the intelligent drying model, becoming another key bottleneck restricting the industrial application of this technology. Summary of the Invention

[0007] One object of the present invention is to solve at least the above-mentioned problems and to provide at least the advantages that will be described later.

[0008] Another objective of this invention is to provide a multi-stage flexible intelligent drying method for dried meat products. This method constructs a drying dynamics model through digital twin pre-training and transfer learning fine-tuning, accurately mapping the raw meat varieties and multi-dimensional physical property parameters with the parameters of a four-stage mid-infrared-hot air synergistic drying program. This enables intelligent generation and adaptive matching of drying programs for multiple varieties of dried meat, significantly improving drying uniformity, nutrient retention rate, and overall quality consistency.

[0009] To achieve these objectives and other advantages according to the present invention, a multi-stage flexible intelligent drying method for drying meat products is provided, comprising: S1. Construct a drying kinetic model. The drying kinetic model takes the variety of raw meat, the physical properties of raw meat, and the target moisture content as inputs, and outputs a set of parameters for a four-stage mid-infrared-hot air synergistic drying program. The physical properties of raw meat include the thickness, fat percentage, fiber content, protein content, and initial moisture content. The four-stage mid-infrared-hot air synergistic drying program includes a preheating stage, a rapid dehydration stage, a uniform drying stage, and a constant temperature color fixing stage, which are connected sequentially. The parameter set for each stage includes the mid-infrared wavelength and / or mid-infrared power density, mid-infrared heating temperature, mid-infrared radiation distance, hot air velocity, drying time, and dehumidification strategy. S2. Using a digital twin model constructed based on the physical field simulation of mid-infrared-hot air synergistic drying, generate a four-segment mid-infrared-hot air synergistic drying program and its simulated physical performance index corresponding to different varieties, different physical property parameters and different drying target moisture content combinations. Based on the simulated physical performance index, determine the drying program with optimal physical performance and pre-train the initial drying kinetic model. S3. Using real mid-infrared-hot air synergistic drying test data as samples, the optimal drying procedure determined based on the overall comprehensive score of multi-dimensional quality indicators is used to fine-tune the pre-trained drying kinetics model through transfer learning to obtain an accurate drying kinetics model. S4. Obtain the variety of raw meat to be dried, the physical property parameters of the raw meat, and the target moisture content for drying. Execute the four-segment mid-infrared-hot air synergistic drying program parameter set output by the precise drying kinetic model obtained in step S3 until the target moisture content is reached. Among them, the simulated physical performance indicators in step S2 include water activity, moisture uniformity, drying uniformity, hot air efficiency, mid-infrared utilization rate, and total drying time; the multi-dimensional quality indicators in step S3 include shear force, chewiness, protein retention rate, nutrient retention rate, and fat oxidation value.

[0010] Preferably, the multi-stage flexible intelligent drying method for dried meat products further includes online evolutionary training of the precise drying kinetics model obtained in step S3, specifically: During the drying process, for each batch of raw meat to be dried, the measured values ​​of physical performance indicators and multi-dimensional quality indicators fed back from the actual drying process are collected to calculate the reward function value. The precise drying kinetic model is updated using a deep deterministic strategy gradient algorithm, so that the set of parameters of the four-segment mid-infrared-hot air synergistic drying program output by the precise drying kinetic model adaptively evolves with the increase of production batches. Among them, the reward function is the weighted sum of the weighted comprehensive score of multi-dimensional quality indicators and the reciprocal of the total drying time.

[0011] Preferably, the construction of the digital twin model in step S2 includes: Establish a coupled set of equations for the mid-infrared radiation heat transfer model, the hot air convection mass transfer model, the moisture diffusion model, and the solid mechanics contraction model; Using the raw meat variety and physical property parameters as input, and the four-stage synergistic drying process parameter set as adjustable boundary conditions, the spatiotemporal distribution of the temperature field, humidity field, water activity field and stress field of the raw meat during the drying process is obtained by solving the coupled equation set through finite element method. The spatiotemporal distributions of the temperature field, humidity field, water activity field, and stress field are statistically or integrally calculated to obtain the simulated physical performance indicators, including: water activity obtained from the spatial average of the water activity field; water uniformity obtained from the spatial variation coefficient of the humidity field; drying uniformity obtained from the time difference of different regions reaching the critical moisture content; hot air efficiency obtained from the ratio of surface moisture evaporation to the total heat input of hot air; mid-infrared utilization rate obtained from the ratio of infrared radiation energy absorbed by the raw meat to the total emitted radiation energy; and total drying time obtained from the simulation time taken to reduce the moisture content to the target value.

[0012] Preferably, the coupled equation set includes: The mid-infrared radiation heat transfer model is based on Lambert-Beer's law and describes the exponential decay of infrared radiation along the thickness direction inside the raw meat and the heat source term. The hot air convection mass transfer model is based on Newton's cooling formula and Fick's surface diffusion law, which couples hot air with the convective heat transfer and moisture evaporation flux on the meat surface. The moisture diffusion model adopts Fick's second law, with the moisture content gradient as the driving force, and introduces temperature and shrinkage correction terms; The solid mechanics shrinkage model is based on the linear viscoelastic constitutive relation, which equates the volume shrinkage caused by water loss to strain and applies it to the stress field.

[0013] Preferably, the pre-training of the initial drying kinetics model in step S2 specifically includes: For each set of virtual raw meat varieties and physical property parameter combinations, the parameters of each segment of the four-segment infrared-hot air synergistic drying program are used as decision variables, and the simulated physical performance index is used as the optimization objective. The Pareto front is solved using a multi-objective optimization algorithm, and then the unique drying program with the best physical performance is selected from the Pareto front based on the weighted comprehensive score as the label of the virtual sample. The initial drying kinetic model is pre-trained using a supervised learning approach. The variety and physical properties of the raw meat are used as inputs, and the parameters of each segment of the drying process with optimal physical performance are used as output labels. By minimizing the mean absolute error loss function, the weights and biases of the initial drying kinetic model are iteratively updated until convergence, thus completing the pre-training of the initial drying kinetic model.

[0014] Preferably, the overall comprehensive score based on multi-dimensional quality indicators in step S3 is calculated in the following way: Shear force, chewability, protein retention rate, nutrient retention rate, and fat oxidation value were normalized and then weighted and summed according to preset weight coefficients to obtain the overall score. The weight coefficients were determined by the analytic hierarchy process, with protein retention rate having the highest weight and fat oxidation value having the second highest weight.

[0015] Preferably, the method for obtaining the actual mid-infrared-hot air synergistic drying test data in step S3 includes: Raw meat samples of different varieties and different combinations of physical property parameters were selected. On a mid-infrared-hot air co-drying equipment, multiple drying tests with different combinations of drying program parameters were performed on each group of raw meat samples to obtain the measured values ​​of multi-dimensional quality indicators of the final dried products of each test group. The overall score is calculated based on the measured values ​​of the multi-dimensional quality indicators. The drying program corresponding to the group with the highest overall score from multiple tests of the same raw meat sample is selected as the label of the optimal drying program for the raw meat sample.

[0016] Preferably, in step S3, the pre-trained drying kinetics model is fine-tuned using transfer learning, specifically as follows: The drying dynamics model pre-trained in step S2 is used as the source domain model, and the network structure and pre-trained weights of the source domain model remain unchanged. The feature extraction layer of the source domain model is frozen so that its weights are not updated during fine-tuning, in order to preserve the basic mapping relationship between the physical property parameters learned from the digital twin virtual samples and the drying procedure. Using the variety and physical properties of raw meat from real drying test data as input, and the parameters of each segment of the corresponding optimal quality drying program as output labels, only the weights of the output layer or the last N fully connected layers of the source domain model are updated, so that the model is transferred and adapted from the mapping relationship of optimal physical performance to the mapping relationship of optimal quality, thus obtaining the accurate drying kinetic model; wherein, in the transfer learning fine-tuning, the unfrozen layers are fine-tuned with a learning rate lower than that of the pre-training stage in step S2, and iterative training is performed by minimizing the mean square error loss function until the loss converges on the validation set of real test data.

[0017] Preferably, in step S4, when executing the four-segment mid-infrared-hot air synergistic drying program parameter set output by the precise drying kinetic model, the following process control rules are also included: During any of the preheating, rapid dehydration, uniform drying and constant temperature color fixing sections, the actual temperature inside the cavity or on the surface of the material is collected in real time by a temperature sensor. When the actual temperature in this segment continuously exceeds the upper limit deviation value of the program-set temperature for this segment, and the cumulative duration exceeds the preset duration threshold, the program-set drying time for this segment will be shortened by a preset ratio based on the original value. When the actual temperature in this segment is consistently lower than the lower limit deviation value of the program-set temperature for this segment, and the cumulative duration exceeds the preset duration threshold, the program-set drying time for this segment will be extended by the preset ratio based on the original value. After adjustment, continue operating this section according to the revised drying time; The upper and lower deviation values ​​are 2-3℃, and the preset ratio is 5-8%.

[0018] The present invention further claims a system for the multi-stage flexible intelligent drying method for dried meat products, comprising: The drying chamber is used to hold the raw meat to be dried and to provide a sealed and insulated environment for the drying process. A mid-infrared heating device is installed inside the drying chamber to generate mid-infrared radiation for internal heating of the raw meat. The infrared wavelength and / or mid-infrared power density, mid-infrared heating temperature and mid-infrared radiation distance are all adjustable. A hot air device is installed in the drying chamber to generate circulating hot air for external dehydration of the raw meat by convection, and the hot air speed is adjustable. A dehumidification device, connected to the drying chamber, is used to remove moisture from the chamber according to a dehumidification strategy. Temperature sensors are used to collect the actual temperature inside the drying chamber or on the surface of the raw meat during the drying process in real time; The control system is connected to the mid-infrared heating device, hot air device, dehumidification device, and temperature sensor; the control system has a built-in precise drying kinetic model, which is constructed and trained according to steps S1 to S3 of the method; The control system is configured as follows: Obtain the variety and physical property parameters of the raw meat to be dried, including the thickness, fat percentage, fiber content, protein content, initial moisture content, and target moisture content of the raw meat; The variety and physical property parameters are input into the precise drying kinetic model, and the precise drying kinetic model outputs a set of parameters for a four-stage mid-infrared-hot air synergistic drying program. The four-stage mid-infrared-hot air synergistic drying program includes a preheating stage, a rapid dehydration stage, a uniform drying stage, and a constant temperature color fixing stage connected in sequence. The parameter set for each stage includes mid-infrared wavelength and / or mid-infrared power density, mid-infrared heating temperature, mid-infrared radiation distance, hot air velocity, drying time, and dehumidification strategy. According to the four-stage mid-infrared-hot air synergistic drying program parameter set, the mid-infrared heating device and the hot air device are controlled to operate in stages, and the dehumidification device is controlled to execute the dehumidification strategy until the raw meat reaches the target moisture content.

[0019] The present invention has at least the following beneficial effects: Firstly, this invention generates virtual samples using a digital twin model to pre-train the drying kinetics model, and then fine-tunes it using transfer learning with real drying test data. This constructs a precise drying kinetics model that can automatically output a four-stage mid-infrared-hot air synergistic drying program based on the raw meat variety and multi-dimensional physical property parameters such as fat percentage and fiber content. This eliminates the reliance on manual experience in traditional drying, realizes the intelligent generation and adaptive matching of drying programs for multiple varieties of dried meat, and significantly improves drying uniformity and overall quality consistency. Secondly, the digital twin model of this invention couples four types of physical equations: mid-infrared radiation heat transfer, hot air convection mass transfer, moisture diffusion, and solid mechanical contraction. By solving the finite element method, the spatiotemporal distributions of temperature field, humidity field, water activity field, and stress field are obtained, and simulated physical performance indicators such as water activity, moisture uniformity, drying uniformity, hot air efficiency, mid-infrared utilization rate, and total drying time are derived. Through multi-physics field coupled simulation, the drying physical behavior under different combinations of physical property parameters can be accurately predicted, providing reliable optimal labels for physical performance for pre-training, greatly reducing the trial and error cost of real experiments, and enabling the model to grasp the intrinsic laws of heat and mass transfer, ensuring a high starting point for subsequent transfer learning. Thirdly, this invention utilizes a deep deterministic strategy gradient algorithm to perform online evolutionary training on the model, using multi-dimensional quality indicators and drying efficiency as rewards, so that the drying process continuously and adaptively optimizes as the number of production batches increases, realizing the self-evolution of the model in the actual production environment, and the more batches, the more stable the quality. Fourth, the present invention effectively compensates for the quality deviation caused by temperature fluctuations during actual drying by dynamically adjusting the drying time when the temperature exceeds the limit, thus ensuring the precise controllability of the moisture content at the end of each stage.

[0020] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the multi-stage flexible intelligent drying method for dried meat products according to the present invention; Figure 2 This is a flowchart illustrating the digital twin pre-training process of the present invention; Figure 3 This is a flowchart illustrating the transfer learning fine-tuning process of a real drying experiment according to the present invention. Figure 4 This is a data flow diagram of the drying kinetics model of the present invention. Detailed Implementation

[0022] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0023] like Figures 1-4 This invention provides a multi-stage flexible intelligent drying method for dried meat products, comprising: S1. Construct a drying kinetic model. The drying kinetic model takes the variety of raw meat, the physical properties of the raw meat, and the target moisture content as inputs, and outputs a set of parameters for a four-stage mid-infrared-hot air synergistic drying program. The physical properties of the raw meat include its thickness, fat percentage, fiber content, protein content, and initial moisture content. The four-stage mid-infrared-hot air synergistic drying program includes a preheating stage, a rapid dehydration stage, a uniform drying stage, and a constant temperature color-fixing stage, which are connected sequentially. The parameter set for each stage includes mid-infrared wavelength and / or mid-infrared power density, mid-infrared heating temperature, mid-infrared radiation distance, hot air velocity, drying time, and dehumidification strategy. Among them, the preheating stage is used to uniformly raise the surface temperature of the raw meat to the moisture activation temperature; the rapid dehydration stage is used to drive the rapid vaporization and migration of internal moisture through mid-infrared deep heating; the uniform drying stage is used to maintain the surface moisture and prevent crusting; and the constant temperature color-fixing stage is used to complete the Maillard reaction and final dehydration at low water activity. S2. Using a digital twin model constructed based on the physical field simulation of mid-infrared-hot air synergistic drying, generate a four-segment mid-infrared-hot air synergistic drying program and its simulated physical performance index corresponding to different varieties, different physical property parameters and different drying target moisture content combinations. Based on the simulated physical performance index, determine the drying program with optimal physical performance and pre-train the initial drying kinetic model. S3. Using real mid-infrared-hot air synergistic drying test data as samples, the optimal drying procedure determined based on the overall comprehensive score of multi-dimensional quality indicators is used to fine-tune the pre-trained drying kinetics model through transfer learning to obtain an accurate drying kinetics model. S4. Obtain the variety of raw meat to be dried, the physical property parameters of the raw meat, and the target moisture content for drying. Execute the four-segment mid-infrared-hot air synergistic drying program parameter set output by the precise drying kinetic model obtained in step S3 until the target moisture content is reached. Among them, the simulated physical performance indicators in step S2 include water activity, moisture uniformity, drying uniformity, hot air efficiency, mid-infrared utilization rate, and total drying time; the multi-dimensional quality indicators in step S3 include shear force, chewiness, protein retention rate, nutrient retention rate, and fat oxidation value.

[0024] The drying kinetics model employs a sub-model parameter system, consisting of a first sub-model, a second sub-model, a third sub-model, and a fourth sub-model, each corresponding to a preheating stage, a rapid dehydration stage, a uniform drying stage, and a constant temperature color-fixing stage, respectively. Each sub-model takes the aforementioned variety and physical property parameters as input and independently outputs the parameter set for its corresponding stage. The parameter set for the four-stage infrared-hot air synergistic drying program is formed by sequentially concatenating the outputs of each sub-model. The input layer of each sub-model receives a (number of variety categories + 6)-dimensional feature vector, including the variety code of the raw meat and six physical property parameters: thickness, fat percentage, fiber content, protein content, initial moisture content, and target moisture content. All input variables are Z-score standardized before entering the network. After the input layer, three fully connected hidden layers are stacked sequentially. The first hidden layer has 64 neurons, the second hidden layer expands to 128 neurons to enhance feature expression, and the third hidden layer is compressed back to 64 neurons. Each hidden layer is followed by a ReLU activation function, where the activation value is the maximum value between the input value and zero. To prevent overfitting, a Dropout layer is inserted after each hidden layer. During the pre-training phase, neuron connections are randomly dropped with a probability of 0.2, and this probability is reduced to 0.1 during the transfer learning fine-tuning phase. Simultaneously, a coefficient of 1×10⁻⁶ is applied to the weights of the fully connected layers. -4 The L2 regularization penalty term is applied. Weight initialization employs a He normal distribution initialization strategy, where elements of each layer's weight matrix are randomly sampled from a normal distribution with a mean of zero and a variance equal to half the number of neurons in the previous layer. Biases are initialized to zero. The output layer is 8-dimensional, corresponding to eight parameters: mid-infrared wavelength, mid-infrared power density, mid-infrared heating temperature, mid-infrared radiation distance, hot air velocity, drying time, dehumidification threshold, and humidity hysteresis. The first seven continuous parameters are mapped to the 0-1 range using a Sigmoid activation function, and then further mapped to their corresponding physical ranges through linear inverse normalization. The humidity hysteresis is directly output linearly.

[0025] In terms of training strategy, the Adam optimizer is used in the pre-training phase, with a learning rate set to 1×10. -3 The batch size is 64, the maximum number of iterations is no more than 500, and early stopping is triggered when the mean absolute error loss on the validation set no longer decreases for 20 consecutive iterations. During the transfer learning fine-tuning phase, the same Adam optimizer is maintained, but the learning rate is reduced to 1×10⁻⁶. -4To accommodate the fine-tuning needs of small-sample real-world experimental data, the batch size was reduced to 16 to enhance the randomness of gradient updates, and the early stopping patience value was set to 10 rounds. During fine-tuning, the weights of the first three hidden layers were frozen to preserve the basic mapping relationship between the physical property parameters learned from the digital twin virtual samples and the drying procedure. Only the weights of the output layer or the last 2-3 fully connected layers were updated, allowing the model to smoothly migrate from the mapping with optimal physical performance to the mapping with optimal quality. In the online evolution phase, the actor-critic architecture in the deep deterministic strategy gradient algorithm was adopted to continuously update the model, with the actor network learning rate maintained at 1×10⁻⁶. -4 The critics' network learning rate was set to 1×10. -3 Both methods use the Adam optimizer, with discount factors ranging from 0.9 to 0.99, and the soft update coefficient of the target network is set to 1 × 10⁻⁶. -3 Each step of gradient update is performed using only a single batch of actual production feedback data, thereby enabling the model to adaptively evolve as the number of production batches increases.

[0026] The state 's' in the deep deterministic strategy gradient algorithm is defined as a vector of inherent physical property parameters of the raw meat obtained before the start of drying for the current batch. Specifically, it includes: the unique thermal code of the raw meat variety, raw meat thickness, fat percentage, fiber content, protein content, initial moisture content, and target moisture content. 's' is a d-dimensional vector composed of the raw meat variety code and the thickness, fat percentage, fiber content, protein content, initial moisture content, and target moisture content. Action 'a' is defined as the normalized vector of all adjustable parameters of the four-segment mid-infrared-hot air synergistic drying program output by the precise drying kinetic model. Each segment contains eight parameters: mid-infrared wavelength, mid-infrared power density, mid-infrared heating temperature, mid-infrared radiation distance, hot air velocity, drying time, dehumidification threshold, and humidity hysteresis (or equivalent time-loop control parameters). The four segments form a 32-dimensional continuous action space. The actual control parameters are obtained by inverse normalization of the normalized values ​​corresponding to all adjustable parameters of the four-segment drying program. The deep deterministic strategy gradient algorithm updates once after each batch of drying is completed. It takes the physical property parameters of the raw meat in the current batch as the state, the drying program parameters output by the model as the action, calculates the reward function value with the measured quality index and drying time of the batch, generates an experience sample and stores it in the playback cache, and randomly samples a small batch of samples from the cache to update the gradient of the actor-critic network.

[0027] The type of raw meat is determined by the name of the variety when it is a single type of meat, such as pork, beef, chicken, or mutton. When the raw meat is a mixture of two or more types of meat, the variety is referred to as mixed meat. Preferably, the mixing ratio can be included as an additional physical property parameter in the input for mixed meat.

[0028] Preferably, after generating virtual samples using a digital twin model, the virtual sample data is first cleaned. For raw meat varieties with existing real experimental data, the value range and distribution characteristics of key parameters (including mid-infrared power density, mid-infrared heating temperature, hot air velocity, total drying time, and moisture uniformity) in the real experiments for that variety are statistically analyzed. For virtual samples of the same variety, if any key parameter exceeds the reasonable expansion range determined by the real experimental data for that variety, it is judged as an unreasonable virtual sample and is removed. The reasonable expansion range is defined as: the value range obtained from the real experimental data is expanded by 1.2 to 1.5 times. At the same time, all virtual samples that violate physical common sense are removed, including those with mid-infrared wavelength in the range of 2.8 to 3.1 μm, mid-infrared power density in the range of 0.08 to 0.18 W / cm², mid-infrared heating temperature in the range of room temperature to 100℃, hot air velocity in the range of 0 to 3.0 m / s, drying time generally not exceeding 10 hours, and simulated water activity value Aw in the range of 0.2 to 0.95. For new varieties for which no real experimental data is available, basic cleaning is performed based solely on the aforementioned physical principles and literature experience. The cleaned virtual sample data is then used to pre-train the initial drying kinetics model.

[0029] In one of the technical solutions, the multi-stage flexible intelligent drying method for dried meat products further includes online evolutionary training of the precise drying kinetics model obtained in step S3, specifically: During the drying process, for each batch of raw meat to be dried, the measured values ​​of physical performance indicators and multi-dimensional quality indicators fed back from the actual drying process are collected to calculate the reward function value. The precise drying kinetic model is updated using a deep deterministic strategy gradient algorithm, so that the set of parameters of the four-segment mid-infrared-hot air synergistic drying program output by the precise drying kinetic model adaptively evolves with the increase of production batches. Among them, the reward function is the weighted sum of the weighted comprehensive score of multi-dimensional quality indicators and the reciprocal of the total drying time.

[0030] The hyperparameters of the deep deterministic policy gradient algorithm are set as follows: the actor network learning rate is 0.0001~0.001, the critic network learning rate is 0.001~0.01, the discount factor is 0.9~0.99, and the target network soft update coefficient is 0.001~0.01; in the reward function, the coefficient of the weighted comprehensive score of the multi-dimensional quality indicators is 0.6~0.8, the coefficient of the reciprocal of the total drying time is 0.2~0.4, and the sum of the two is 1.

[0031] like Figure 2 In one of the technical solutions, the construction of the digital twin model in step S2 includes: Establish a coupled set of equations for the mid-infrared radiation heat transfer model, the hot air convection mass transfer model, the moisture diffusion model, and the solid mechanics contraction model; Using the raw meat variety and physical property parameters as input, and the four-stage synergistic drying process parameter set as adjustable boundary conditions, the spatiotemporal distribution of the temperature field, humidity field, water activity field and stress field of the raw meat during the drying process is obtained by solving the coupled equation set through finite element method. The spatiotemporal distributions of the temperature field, humidity field, water activity field, and stress field are statistically or integrally calculated to obtain the simulated physical performance indicators, including: water activity obtained from the spatial average of the water activity field; water uniformity obtained from the spatial variation coefficient of the humidity field; drying uniformity obtained from the time difference of different regions reaching the critical moisture content; hot air efficiency obtained from the ratio of surface moisture evaporation to the total heat input of hot air; mid-infrared utilization rate obtained from the ratio of infrared radiation energy absorbed by the raw meat to the total emitted radiation energy; and total drying time obtained from the simulation time taken to reduce the moisture content to the target value.

[0032] In one of the technical solutions, the set of coupled equations includes: The mid-infrared radiation heat transfer model is based on Lambert-Beer's law and describes the exponential decay of infrared radiation along the thickness direction inside the raw meat and the heat source term. The hot air convection mass transfer model is based on Newton's cooling formula and Fick's surface diffusion law, which couples hot air with the convective heat transfer and moisture evaporation flux on the meat surface. The moisture diffusion model adopts Fick's second law, with the moisture content gradient as the driving force, and introduces temperature and shrinkage correction terms; The solid mechanics shrinkage model is based on the linear viscoelastic constitutive relation, which equates the volume shrinkage caused by water loss to strain and applies it to the stress field.

[0033] Mid-infrared radiation heat transfer model: The attenuation of mid-infrared radiation along the thickness direction (x-axis, origin at the irradiated surface) within the raw meat follows the Lambert-Beer law, and the volumetric heat source q generated by its absorption is... IR (x) is: Where I0 is the effective infrared radiation intensity reaching the meat surface (W / m²). 2 ); PIR is the mid-infrared radiation power (set by the drying program); Arad is the effective irradiated area (determined by sample size and the relationship between radiation sources); α is the meat surface absorptivity (usually taken as 0.90~0.95, which can be obtained by FTIR reflectance measurement); μ is the linear absorption coefficient of the raw meat to mid-infrared radiation (m). -1 For a given mid-infrared wavelength, , κ λ λ is the extinction coefficient of the meat sample at this wavelength; λ is the wavelength; μ can also be determined by Fourier transform infrared spectroscopy (FTIR) transmission experiments, or by consulting literature on the optical properties of meat. Typical values ​​are around 10.4 ~10 5 m −1 ; x is the depth (m) from the irradiated surface into the flesh. This heat source term is directly added to the internal source term of the energy conservation equation, i.e.: , where ρ and c p and k represent the density, specific heat capacity, and thermal conductivity of the raw meat, which can be found in the Food Engineering Handbook or determined experimentally using methods such as DSC and hot wire method.

[0034] Hot air convection mass transfer model: Hot air and the surface of the raw meat (x=0) simultaneously undergo convective heat transfer and convective mass transfer, providing energy and mass boundary conditions. Convective heat transfer boundary conditions: k is the thermal conductivity of the raw meat (W / (m·k)); h is the convective heat transfer coefficient (W / (m·k)). 2 The value of h is calculated using the forced convection criterion correlation. In the range of 40–80°C hot air and 1–5 m / s wind speed, the typical value of h is approximately 15–60 W / (m³). 2 ·k);T air T represents the hot air temperature. s The surface temperature of the raw meat is determined by the finite element method, which solves the entire temperature field. The temperature node values ​​at the x=0 boundary are then extracted. v The latent heat of vaporization of water is approximated by a constant of 2.45 × 10⁻⁶. 6 J / kg; Jm is the surface moisture evaporation mass transfer flux (kg / (m²)). 2 ·s), based on Fick's surface diffusion law and the expression of the convective mass transfer coefficient: ;k m The convective mass transfer coefficient was obtained through the Chilton-Colburn analogy. The mass concentration of water vapor at the surface of the meat (kg / m³) 3 ); The mass concentration of water vapor in the mainstream of hot air.

[0035] Moisture diffusion model: Moisture migration within raw meat follows Fick's second law, with temperature and shrinkage corrections incorporated. Governing equations (in Lagrange dry matter coordinates): To eliminate mesh deformation caused by shrinkage, Lagrangian coordinates based on dry matter mass are often used in simulations, eliminating the need for additional convection terms. M represents the dry-basis moisture content (kg water / kg dry matter). Where D... eff It is the effective diffusion coefficient. ;D0 pre-exponential factor (m 2 / s), E a Given the diffusion activation energy (J / mol), R = 8.314 J / (mol·K), D0 and E aObtained through inverse analysis of isothermal drying experiments (e.g., fitting the drying curve with Crank analytical solutions); f M (M) is a moisture content-dependent function, commonly in power-law form. n is an empirical constant, usually 1 ≤ n ≤ 3; f η (η) is the shrinkage correction factor, where η is the volume shrinkage rate: ,β V The volume shrinkage coefficient (m) 3 / kg dry matter), obtained by linear regression of sample volume at different moisture contents, commonly corrected as follows: The surface is coupled with the hot air convection mass transfer model: ;where ρ s Dry matter density (kg dry matter / m³) 3 The initial volume is V0), and the symmetry center plane or ground surface uses a zero-flux boundary.

[0036] Solid mechanics shrinkage model: The volume shrinkage of raw meat during the drying process is driven by moisture loss, and the stress-strain behavior is described using a linear viscoelastic constitutive relation. Shrinkage strain: Assuming isotropic shrinkage, the shrinkage strain tensor ε sh It has a linear relationship with changes in moisture content: Where I is the second-order unit tensor, and M0 is the initial dry-basis moisture content of the raw meat. The linear viscoelastic constitutive equation adopts the Boltzmann superposition integral form. For isotropic materials: ; where e mech For mechanical deviatoric strain: The total strain ε can be decomposed into mechanical strain and contraction strain: ε = ε mech +ε sh , Mechanical volumetric strain: The shear relaxation modulus G(t) and volume relaxation modulus K(t) are expressed using Prony series, and only 2-3 terms are needed for simplification. Mechanical equilibrium: neglecting inertial forces, the stress field satisfies: Boundary condition: There are no external forces on the free surface, i.e. , where n is the outward normal unit vector of the surface; if the sample is constrained by a fixture, the corresponding nodal displacement is set to zero. Coupling feedback: contraction leads to sample thickness... The changes in temperature and humidity fields affect the infrared penetration depth and the characteristic length of the convective boundary layer. In the numerical solution, mechanical strain is fed back to the heat and mass transfer equations through a dynamic mesh or deformable geometry module, achieving complete coupling. The above models together constitute the coupled equation set for the digital twin simulation. The spatiotemporal distributions of the temperature field, humidity field (water content field), water activity field, and stress field can be solved using the finite element method (using general-purpose finite element software such as ANSYS), thereby calculating the simulated physical performance indicators.

[0037] The specific process of solving the problem using the finite element method is as follows: Define geometric and material parameters. Geometric parameters: Raw meat is usually simplified to a two-dimensional (sliced) or three-dimensional (cubic prism) region, with the thickness direction being the main direction of change; Material parameters: density ρ, specific heat capacity c. p Thermal conductivity k, infrared absorption coefficient μ, moisture diffusivity D eff (T,M), volumetric shrinkage coefficient βV, viscoelastic parameters (Prony series). Based on the established coupled partial differential equations, energy is conserved (temperature field): Moisture diffusion (water content field): Solid mechanics (stress / strain field): The standard Galerkin finite element method is adopted: the solution domain is divided into quadrilateral / hexahedral meshes (local refinement is required due to the large gradient in the thickness direction); temperature T, moisture content M, and displacement u are used as nodal degrees of freedom, and shape functions N are used respectively. i Interpolation; multiply each equation by a weighting function (the shape function itself), integrate over the element, and obtain the residual equation. Time discretization: use backward Euler (implicit) or generalized -α method to ensure the stability of the coupled system. Use either monolithic or staggered iteration methods. Since temperature and moisture content fields strongly influence mechanical strain (contraction), and contraction in turn affects geometric thickness (and thus infrared penetration and convection), staggered iteration is used. Finally, extract simulated physical performance indicators from the solution results: the water activity at each node is calculated by M and T through an isothermal adsorption model, and the average water activity of the entire meat block is reported. Define moisture homogeneity as calculated by the spatial coefficient of variation (CV) of the moisture content field M(x) at the final time: , Moisture uniformity is expressed as 1 − CV. Drying uniformity is calculated using the time difference between different regions of the meat block reaching the critical moisture content. In the finite element post-processing, the time t at which each node reaches the critical moisture content is recorded. i ,calculate: Drying uniformity = 1 - (Δt / total drying time). Critical moisture content refers to the average moisture content of the material when the drying rate transitions from a constant rate to a decreasing rate during the drying process. Hot air efficiency is calculated as the ratio of heat absorbed by surface moisture evaporation to the total heat input of the hot air. Mid-infrared utilization rate is calculated as the ratio of infrared energy absorbed by the meat to the total infrared energy emitted. Total drying time refers to the time from 0°C until the average moisture content of the meat first drops to the target moisture content.

[0038] In one of the technical solutions, the pre-training of the initial drying kinetics model in step S2 specifically includes: For each set of virtual raw meat varieties and physical property parameter combinations, the parameters of each segment of the four-segment infrared-hot air synergistic drying program are used as decision variables, and the simulated physical performance index is used as the optimization objective. The Pareto front is solved using a multi-objective optimization algorithm, and then the unique drying program with the best physical performance is selected from the Pareto front based on the weighted comprehensive score as the label of the virtual sample. The initial drying kinetic model is pre-trained using a supervised learning approach. The variety and physical properties of the raw meat are used as inputs, and the parameters of each segment of the drying process with optimal physical performance are used as output labels. By minimizing the mean absolute error loss function, the weights and biases of the initial drying kinetic model are iteratively updated until convergence, thus completing the pre-training of the initial drying kinetic model.

[0039] The formula for calculating the weighted comprehensive score in the drying process that selects the unique physical performance-optimized drying procedure from the Pareto front based on the weighted comprehensive score is: S = A1 × (1 - water activity / upper limit of water activity) + A2 × moisture uniformity + A3 × drying uniformity + A4 × hot air efficiency + A5 × mid-infrared utilization rate + A6 × (1 / total drying time), where A1~A6 are weighting coefficients, and A2+A3≥0.4, A1+A6≥0.3. Each weight is determined and normalized by the entropy weight method, and the upper limit of water activity is taken as 0.7.

[0040] The kinetic targets are: water activity ≤ 0.7; lipid oxidation value ≤ 0.3 mg MDA / kg; and moisture homogeneity ≥ 95%.

[0041] In one of the technical solutions, the overall comprehensive score based on multi-dimensional quality indicators in step S3 is calculated in the following way: Shear force, chewiness, protein retention rate, nutrient retention rate, and lipid oxidation value were normalized and then weighted and summed according to preset weighting coefficients to obtain a total comprehensive score. The weighting coefficients were determined using the analytic hierarchy process (AHP), with protein retention rate having the highest weight, followed by lipid oxidation value. Specifically, the preset weighting coefficients were: protein retention rate 0.35–0.45, lipid oxidation value 0.25–0.35, shear force 0.10–0.15, chewiness 0.10–0.15, and nutrient retention rate 0.05–0.10, with the sum of all weights being 1.

[0042] In one of the technical solutions, the method for obtaining the real mid-infrared-hot air synergistic drying test data in step S3 includes: Raw meat samples of different varieties and different combinations of physical property parameters were selected. In a mid-infrared-hot air co-drying equipment, drying tests with multiple combinations of different drying program parameters were performed on each group of raw meat samples to obtain the measured values ​​of multi-dimensional quality indicators of the final dried products of each test group. Multiple combinations of different drying program parameters were generated by the central composite design method. The overall score is calculated based on the measured values ​​of the multi-dimensional quality indicators. The drying program corresponding to the group with the highest overall score from multiple tests of the same raw meat sample is selected as the label of the optimal drying program for the raw meat sample.

[0043] In one of the technical solutions, step S3 involves fine-tuning the pre-trained drying dynamics model using transfer learning, specifically as follows: The drying dynamics model pre-trained in step S2 is used as the source domain model, and the network structure and pre-trained weights of the source domain model remain unchanged. The feature extraction layer of the source domain model is frozen so that its weights are not updated during fine-tuning, in order to preserve the basic mapping relationship between the physical property parameters learned from the digital twin virtual samples and the drying procedure. Using the variety and physical property parameters of raw meat from real drying test data as input, and the parameters of each segment of the corresponding optimal quality drying program as output labels, only the weights of the output layer or the last N fully connected layers of the source domain model are updated, so that the model is adapted from the mapping relationship of optimal physical performance to the mapping relationship of optimal quality, thus obtaining the accurate drying kinetic model; the N of the last N fully connected layers takes the value of 2 or 3. In the transfer learning fine-tuning process, a fine-tuning learning rate lower than that of the pre-training stage in step S2 is used for the unfrozen layers. The fine-tuning learning rate is generally taken as 1 / 10 to 1 / 5 of the learning rate of the pre-training stage in step S2. Iterative training is performed by minimizing the mean square error loss function until the loss converges on the validation set of real experimental data.

[0044] Before fine-tuning the transfer learning, Gaussian noise with a mean of 0 and a standard deviation equal to the measurement error of the physical property parameters is applied to the physical property parameters in the real drying test data, along with random masking, in order to expand the sample size and enhance the generalization ability.

[0045] In one of the technical solutions, when executing the four-segment mid-infrared-hot air synergistic drying program parameter set output by the precise drying kinetic model in step S4, the following process control rules are also included: During any of the preheating, rapid dehydration, uniform drying and constant temperature color fixing sections, the actual temperature inside the cavity or on the surface of the material is collected in real time by a temperature sensor. When the actual temperature in this segment continuously exceeds the upper limit deviation value of the program-set temperature for this segment, and the cumulative duration exceeds the preset duration threshold, the program-set drying time for this segment will be shortened by a preset ratio based on the original value. When the actual temperature within a segment remains below the lower limit deviation value of the program-set temperature for that segment, and the cumulative duration exceeds the preset duration threshold, the program-set drying time for that segment is extended by the preset proportion based on the original value. The cumulative duration is the duration during which the actual temperature continuously exceeds the upper or lower limit deviation value on a single occasion, starting from the start time of the current segment. The preset duration threshold is 120~300s. After adjustment, continue operating this section according to the revised drying time; The upper and lower deviation values ​​are 2-3℃, and the preset ratio is 5-8%.

[0046] like Figure 1 The present invention further claims a system for the multi-stage flexible intelligent drying method for dried meat products, comprising: The drying chamber is used to hold the raw meat to be dried and provide a sealed, insulated environment for the drying process. Inside the drying chamber is a material-carrying and conveying structure for holding the raw meat. This structure consists of a movable multi-layer rack, a breathable mesh panel, and material trays. The rack is a stainless steel frame with 10-20 layers, and the layer spacing can be flexibly set according to the material thickness. The mesh panel is made of stainless steel woven mesh, offering good air permeability and uniform heating, ensuring hot air penetration and infrared radiation. The rack is equipped with casters at the bottom, allowing it to be pushed into / pull out of the chamber as a whole for easy loading, unloading, and cleaning. It can achieve continuous conveying or static loading, ensuring that the material is laid flat and of uniform thickness in a single layer, resulting in higher consistency in the drying process. A mid-infrared heating device is installed inside the drying chamber to generate mid-infrared radiation for internal heating of the raw meat. The infrared wavelength and / or mid-infrared power density, mid-infrared heating temperature and mid-infrared radiation distance are all adjustable. A hot air device is installed in the drying chamber to generate circulating hot air for external dehydration of the raw meat by convection, and the hot air speed is adjustable. A dehumidification device, connected to the drying chamber, is used to remove moisture from the chamber according to a dehumidification strategy. Temperature sensors are used to collect the actual temperature inside the drying chamber or on the surface of the raw meat during the drying process in real time; The control system is connected to the mid-infrared heating device, hot air device, dehumidification device, and temperature sensor; the control system has a built-in precise drying kinetic model, which is constructed and trained according to steps S1 to S3 of the method; The control system is configured as follows: Obtain the variety and physical property parameters of the raw meat to be dried, including the thickness, fat percentage, fiber content, protein content, initial moisture content, and target moisture content of the raw meat; The variety and physical property parameters are input into the precise drying kinetic model, and the precise drying kinetic model outputs a set of parameters for a four-stage mid-infrared-hot air synergistic drying program. The four-stage mid-infrared-hot air synergistic drying program includes a preheating stage, a rapid dehydration stage, a uniform drying stage, and a constant temperature color fixing stage connected in sequence. The parameter set for each stage includes mid-infrared wavelength and / or mid-infrared power density, mid-infrared heating temperature, mid-infrared radiation distance, hot air velocity, drying time, and dehumidification strategy. According to the four-stage mid-infrared-hot air synergistic drying program parameter set, the mid-infrared heating device and the hot air device are controlled to operate in stages, and the dehumidification device is controlled to execute the dehumidification strategy until the raw meat reaches the target moisture content.

[0047] Specifically, the control system is divided into a computer-based module (model calculation) and an external control device (execution). The control system includes: a host computer module for running the precise drying kinetic model and outputting drying program parameters; and a slave control module located outside the drying chamber, connected to the host computer module via wired or wireless communication, for receiving drying program parameters and executing segmented control. The drying kinetic model and online evolutionary training module are deployed on a regular personal computer, industrial computer, or cloud server (host computer), sending drying program parameters to the slave computer via interfaces (such as HTTP API, serial port, Modbus, etc.). The drying chamber is equipped with an embedded external controller, responsible for sensor acquisition, actuator driving, local PID closed-loop regulation, and communication with the host computer. The controller receives four-segment drying program parameters (time, temperature, power, wind speed, dehumidification strategy, etc.) from the computer, independently executes sequential control (via PLC, microcontroller, or embedded industrial control board), collects sensor data such as temperature and humidity, and transmits it back to the computer, and can store execution logs locally. The communication methods for both devices can be wired (RS485 / RS232, Ethernet (TCP / IP), suitable for factory environments) or wireless (Wi-Fi, LoRa, 4G / 5G), facilitating remote monitoring and cross-workshop deployment. Communication protocols can include Modbus RTU / TCP, MQTT, or a custom simple protocol. To prevent loss of control due to communication interruptions, the external control device should have the ability to locally store the most recent drying program and be equipped with a watchdog timer. Once communication is restored, the drying process data should be automatically uploaded for online evolutionary training of the model.

[0048] like Figure 1As shown, in the system, the mid-infrared heating device includes multiple sets of quartz mid-infrared heating tubes and ceramic infrared radiation plates. The emission wavelength is continuously adjustable within the range of 2.8~3.1μm, the radiation distance is adjustable within the range of 60~430mm, and the heating temperature is adjustable within the range of room temperature to 200℃. It is equipped with a dual monitoring module consisting of a PT100 thermocouple and an infrared temperature probe. The mid-infrared heating tubes are symmetrically arranged vertically and / or horizontally along the drying chamber. The hot air device includes a high-temperature resistant variable frequency circulating fan, a uniform air duct, and a return air duct. The variable frequency circulating fan is installed at the top of the drying chamber, and the wind speed is adjustable within the range of 0.5~3.0m / s. The uniform air duct adopts a perforated plate structure.

[0049] Example 1 (Dried Beef) Using commercially available fresh beef hind leg meat as raw material, beef muscle fibers are coarse and tightly arranged, with a strong water-holding capacity (the initial moisture content of this batch was measured to be 74.7%). Traditional hot air drying easily forms a dry layer on the surface, which hinders the migration of internal moisture to the outside, resulting in a dry exterior and a wet interior and an excessively hard texture.

[0050] Raw material preparation and physical property parameter acquisition: Fresh beef hind leg meat was selected, visible fascia and fat were removed, and the meat was sliced ​​into 8 mm thick slices along the muscle fiber direction. Physical property parameters of the raw meat were determined: the initial moisture content was 74.7% (corresponding to approximately 2.95 kg water / kg dry matter on a dry basis) determined by gravimetric method; the fat content was 4.8% determined by Soxhlet extraction; the protein content was 22.3% determined by Kjeldahl nitrogen determination method; and the fiber content was 18.6% determined by fiber separation method (there is component overlap between crude fiber and protein). The target moisture content was set at 25% (wet basis). The raw meat variety was designated as beef, coded as A.

[0051] Drying kinetics model output drying program: Input the beef and its physical properties into the precise drying kinetics model. The model outputs a four-stage mid-infrared-hot air synergistic drying program as follows, and the parameters of each stage are shown in Table 1.

[0052] Table 1. Beef Drying Parameters Note: The dehumidification strategy in the table uses time-cycle control. The reference settings for humidity control are as follows: Rapid dehydration stage: dehumidification threshold 65%RH, hysteresis 5%RH (corresponding to dehumidification for 30s every 8min); Uniform drying stage: dehumidification threshold 70%RH, hysteresis 5%RH (corresponding to dehumidification for 30s every 20min); Constant temperature color fixing stage: dehumidification threshold 72%RH, hysteresis 8%RH (corresponding to 6min on, 4min off).

[0053] As shown in Table 1, for beef drying, Example 1, based on the parameters output by the precise drying kinetic model, implemented a deep-penetration, slow dehydration strategy: Mid-infrared radiation was used to directly heat the interior of the material, causing moisture vaporization and generating internal pressure that drove it outward migration. Simultaneously, the hot air velocity and temperature were controlled to prevent the surface moisture evaporation rate from exceeding the internal moisture replenishment rate, thus maintaining a non-crusted surface throughout the drying process and providing a channel for continuous internal moisture overflow. Specifically, the preheating stage aimed to gently raise the temperature and activate internal moisture, avoiding sudden surface heating. The rapid dehydration stage used a 2.90 μm wavelength precisely matched to the broad absorption peak of water molecules, with moderate power density and a short radiation distance, allowing mid-infrared radiation to penetrate directly to the deep layers of muscle fibers. The internal moisture rapidly vaporized, generating pressure that drove moisture to migrate from the inside out; simultaneously, the hot air velocity was controlled at 2.0 m / s, only assisting in removing surface moisture to prevent excessively rapid surface evaporation and the formation of a hard crust. The uniform drying section appropriately reduces the infrared intensity and extends the drying time, allowing sufficient time for internal moisture to diffuse to the surface, maintaining the surface in a consistently moist state, thus preventing crust formation and ensuring that the drying front advances uniformly inward. The constant temperature color-fixing section continues to reduce the moisture content to the target value while simultaneously using the Maillard reaction to develop the characteristic good color and flavor of beef jerky.

[0054] Key controls during the drying process: In the first 30 minutes of the rapid dehydration phase, infrared thermal imaging monitoring ensured the surface temperature of the meat slices remained consistently between 55 and 60°C, with no localized hot spots observed. During the uniform drying phase, samples were taken every 20 minutes to observe the cross-section of the meat slices: the cross-section was moist, and there was no obvious darkening of the hard shell layer, indicating that internal moisture was continuously migrating outwards, and the drying front was advancing uniformly inwards. No temperature over-limit adjustments were triggered throughout the process (actual temperature fluctuation ≤ ±2°C), and the drying process was stable.

[0055] Drying Results and Quality Evaluation: Total drying time was 270 minutes (approximately 4.5 hours). The final product had a moisture content of 25.9% (wet basis), with a soft, non-dry surface and uniform internal moisture. Moisture uniformity: the standard deviation / mean of moisture content in different locations within the same tray was 3.2% (moisture uniformity 96.8%), with no external dryness and internal wetness. Shear force: 128.54 N (measured using a texture analyzer with a cylindrical probe along the muscle fiber direction); hardness: 96.72 N, within the suitable chewing range, neither too high due to surface hardening (traditional hot air drying often >210 N, making it difficult to chew) nor too soft, thus maintaining the proper chewiness of beef jerky. Chewability: 55.08 N, moderate. Nutrient retention (vitamins): 91.2%. Protein retention: 92.3%, significantly higher than the 65-75% of traditional methods. This is because slow dehydration avoids severe protein denaturation caused by high surface temperatures. Fat oxidation value: 0.28 mg MDA / kg, below the limit of 0.5 mg MDA / kg. Cross-sectional observation: The interior of the meat slices is uniformly brownish-red, without any white dry core or dark wet core, indicating that moisture is drained smoothly.

[0056] A control experiment was conducted using the same beef raw material and conventional hot air drying (constant temperature 65℃, air velocity 2.5m / s, no infrared radiation). Results: After 8 hours of drying, the moisture content was still 29.6% (not reaching the target), a distinct hard crust had formed on the surface, shear force reached 255.98N, and hardness was 119.36N (exceeding the upper limit of 210 N, resulting in a tough and chewy texture). The protein retention rate was only 68%. The parameters and strategies of Example 1 were significantly superior to the control in key indicators such as drying time, shear force, and protein retention rate.

[0057] Example 1 successfully achieved deep penetration, slow dehydration, flexible and intelligent drying of beef by combining infrared wavelength, power density, wind speed, and long-term uniform speed segment in a four-segment program output by a precise drying kinetic model. This effectively prevented surface crusting, ensured smooth escape of internal moisture, and controlled the final product shear force at 128.54N and hardness at 96.72N, resulting in a significant improvement in overall quality.

[0058] Example 2 (Pork Drying) The product is made from commercially available fresh pork hind leg meat. Pork muscle fibers are finer, and the fat content is significantly higher than that of beef (the actual fat content of this batch was 18.6%). It also has a high proportion of unsaturated fatty acids, which makes it very easy to melt and release oil on the surface when heated, and accelerates oxidation and rancidity.

[0059] Raw material preparation and physical property parameter acquisition: Fresh pork hind leg meat was selected, and visible fascia and epidermis were removed. The meat was then sliced ​​into 7 mm thick pieces along the muscle fiber direction. Physical property parameters of the raw meat were determined: the initial moisture content was 73.8% (approximately 2.82 kg water / kg dry matter on a dry basis) determined by gravimetric method; the fat content was 18.6% determined by Soxhlet extraction; the protein content was 19.5% determined by Kjeldahl nitrogen determination; and the fiber content was 12.3% determined by fiber separation method. The target moisture content was set at 18.0% (wet basis). The raw meat variety was designated as pork, coded as B.

[0060] The drying kinetics model outputs the drying program: Input the "pork" and the aforementioned six physical properties into the precise drying kinetics model. The four-stage mid-infrared-hot air synergistic drying program output by the model is shown in Table 2.

[0061] Table 2 Pork Drying Parameters Note: The dehumidification strategy in the table uses time-cycle control. The reference settings for humidity control are as follows: Rapid dehydration stage: dehumidification threshold 63%RH, hysteresis 5%RH (corresponding to dehumidification for 35s every 6min); Uniform drying stage: dehumidification threshold 68%RH, hysteresis 5%RH (corresponding to dehumidification for 30s every 10min); Constant temperature color fixing stage: dehumidification threshold 73%RH, hysteresis 6%RH (corresponding to 4min on, 2min off).

[0062] As shown in Table 2, for pork drying, this embodiment employs a low-temperature oil control and strong convection strategy based on parameters output from a precise drying kinetic model: a lower mid-infrared power density and a larger radiation distance are used to avoid localized overheating that could cause fat melting; a wavelength of 2.8–2.95 μm is selected to avoid the fat's strong absorption heat-sensitive region (3.2–3.5 μm); simultaneously, the hot air velocity is increased (2.0–2.6 m / s) to utilize forced convection to promptly remove surface-permeated oil and moisture, thereby inhibiting fat oxidation and maintaining the product's refreshing taste. Specifically, in the preheating stage (42℃ / 15 min): the lowest power density (0.09 W / cm³) is used in the preheating stage. 2 The maximum radiation distance (430 mm) is controlled to ensure gentle heating and prevent premature melting of fat. The rapid dehydration phase (59℃ / 75 min) uses a 2.85 μm wavelength to avoid the strong absorption peak of fat in the 3.2~3.5 μm range, reducing direct absorption of infrared energy by fat; the power density is controlled at 0.13 W / cm². 2 (lower than 0.16 W / cm² for dried beef) 2The radiation distance is 350mm (significantly longer than the 200mm for beef) to reduce infrared energy deposition per unit volume and prevent fat overheating. Simultaneously, the hot air velocity is increased to 2.4m / s to forcibly disperse surface-permeable grease, and the dehumidification frequency is increased to 35s dehumidification every 6 minutes to promptly remove high-humidity, oily exhaust gases. In the uniform drying section (57℃ / 105 min): a 2.95 μm wavelength is used, and the drying time is appropriately extended (105 min) to allow internal moisture to slowly diffuse to the surface, preventing surface grease concentration due to excessively rapid evaporation; the hot air velocity is further increased to 2.6 m / s to enhance the convection scouring effect. In the constant temperature color fixing section (55℃ / 45 min): the power density is adjusted back to 0.09 W / cm³. 2 The radiation distance is shortened to 280 mm, color fixation is completed at a lower temperature, the Maillard reaction is suppressed from excessive production of potentially harmful substances, and the lipid oxidation value is kept at a low level.

[0063] Key controls during the drying process: During the rapid dehydration phase, infrared thermal imaging monitored the surface temperature of the meat slices, maintaining it at a stable 56-58℃ with no localized hot spots observed (traditional fixed-parameter infrared drying often produces hot spots >65℃, leading to fat liquefaction and seepage). The humidity sensor inside the chamber showed relative humidity controlled between 63% and 75%, with no obvious oil mist condensation observed in the exhaust air (indicating that surface oil was effectively carried away by the hot air and did not evaporate into the chamber). During the uniform drying phase, meat slices were observed every 30 minutes: the surface was dry with no visible oil sheen, and the cross-section showed no fat liquefaction cavities, indicating the effectiveness of the low-temperature oil control strategy. The actual temperature fluctuation throughout the process was ≤±2℃, and the time correction rule was not triggered.

[0064] Drying Results and Quality Evaluation: Total drying time: 240 min (4.0 h). Final product moisture content: 18.9% (wet basis), surface dry and non-greasy, internal meat tender. Moisture uniformity: Standard deviation / mean of moisture content in different locations within the same tray = 3.5%, no dry exterior and wet interior. Shear force: 78.46 N, hardness: 61.37 N (pork muscle fibers are relatively fine, suitable shear force is 70~110 N), significantly lower than the 150~190 N of traditional hot air drying, resulting in a moderately firm texture. Fat oxidation value (TBARS value): 0.19 mg MDA / kg (determined by thiobarbituric acid colorimetric method, far below the limit of 0.5 mg MDA / kg, and significantly lower than the 0.4~0.6 mg MDA / kg commonly observed in traditional infrared drying), product without rancid odor, normal color. Protein retention: 91.8% (approximately 60~70% for traditional hot air drying). Nutritional retention rate (vitamins): 92.5%. Cross-sectional observation: The meat is uniform in texture, with no cavities or oil spots formed after fat liquefaction, indicating that the fat structure is basically intact.

[0065] Control 1 (conventional hot air drying): constant temperature 65℃, air velocity 2.5 m / s, no infrared radiation. Results: after 9.5 hours of drying, the moisture content was still 23% (not up to standard), the surface was hard, shear force was 119.86 N, hardness was 106.73 N, fat oxidation value was 0.42 mg MDA / kg (slight rancidity was observed), and protein retention was 65%. Control 2 (using beef drying parameters, i.e., higher power, near-field infrared): the beef drying parameters in Example 1 were used (power density of 0.16 W / cm² in the rapid dehydration section, radiation distance of 200 mm, wavelength of 2.90 μm, hot air velocity of 2.0 m / s). Results: at 40 minutes of drying, the surface of the meat slices showed obvious oil production, and infrared thermal imaging showed a local temperature of 68℃; the final product had a moisture content of 22.3%, but the fat oxidation value was as high as 0.58 mg MDA / kg (seriously exceeding the standard), there were oil spots on the surface, and the chewing test showed a greasy feeling, indicating that the quality was unqualified.

[0066] Example 2 successfully achieved low-temperature oil control and strong convection flexible intelligent drying of pork through a four-stage program output by a precise drying kinetic model. This program features a low infrared power density, a long radiation distance, wavelengths that avoid the heat-sensitive area of ​​fat, and a high hot air velocity. It effectively inhibited fat melting and oxidative rancidity, resulting in a final product with a fat oxidation value as low as 0.19 mg MDA / kg, a shear force of 78.46 N, and a hardness of 61.37 N, demonstrating excellent overall quality.

[0067] Example 3 (Dried Chicken) Made from commercially available fresh chicken breast. Chicken meat has fine muscle fibers and low fat content (2.8% fat content measured in this batch), but high protein content (approximately 23.5%), and its tissue is loose and porous, making it extremely heat-sensitive. Traditional hot air drying or excessively high-power infrared heating can easily cause excessive denaturation of surface proteins, resulting in a charred, hardened texture with a woody feel, and at the same time, rapid moisture evaporation makes the meat dry and tough.

[0068] Raw material preparation and physical property parameter acquisition: Fresh chicken breast was selected, visible fascia and fat were removed, and the meat was sliced ​​into 6 mm thick slices along the muscle fiber direction. Physical property parameters of the raw meat were determined: the initial moisture content was 75.5% (approximately 3.08 kg water / kg dry matter on a dry basis) determined by gravimetric method; the fat content was 2.8% determined by Soxhlet extraction; the protein content was 23.5% determined by Kjeldahl nitrogen determination method; and the fiber content was 8.5% determined by fiber separation method. The target moisture content was set at 18.0% (wet basis). The raw meat variety was designated as chicken meat, coded as C.

[0069] The drying kinetics model outputs the drying program: Input the "chicken" and the aforementioned six physical properties into the precise drying kinetics model. The four-stage mid-infrared-hot air synergistic drying program output by the model is shown in Table 3.

[0070] Table 3 Chicken Drying Parameters Note: The dehumidification strategy in the table uses time-cycle control. The reference settings for humidity control are as follows: Rapid dehydration stage: dehumidification threshold 62%RH, hysteresis 4%RH (corresponding to dehumidification for 45s every 5min); Uniform drying stage: dehumidification threshold 67%RH, hysteresis 5%RH (corresponding to dehumidification for 40s every 9min); Constant temperature color fixing stage: dehumidification threshold 72%RH, hysteresis 6%RH (corresponding to 3min on, 2min off).

[0071] As can be seen from the data in Table 3, for chicken drying, Example 3, based on the parameters output by the precise drying kinetic model, implemented a flexible, mild, and short-term strategy: using a lower mid-infrared power density (0.08~0.12 W / cm²). 2 The system features a suitable radiation distance (220~320 mm) and a wavelength of 2.88~2.98 μm (balancing protein protection and moisture absorption), combined with a high hot air velocity (2.2~2.8 m / s) and intensive dehumidification, achieving rapid dehydration while avoiding localized overheating, preventing excessive protein denaturation and water loss, and maintaining tender meat texture. Specifically, during the preheating stage (40℃ / 12 min): a minimum power density of 0.08 W / cm³ is used. 2 The radiation distance is 320 mm, and the hot air velocity is 2.2 m / s, gently awakening internal moisture and preventing premature denaturation of surface proteins. The rapid dehydration stage (57℃ / 65 min) uses a 2.88 μm wavelength, which is near the absorption peak of water molecules, while avoiding over-excitation of proteins; the power density is only 0.10 W / cm³. 2 The radiation distance is 260 mm, primarily using shallow penetration to prevent excessive shrinkage of the protein network due to deeper heat penetration. The hot air velocity is increased to 2.5 m / s, and dehumidification is intensified to 45 s every 5 minutes, forcibly removing surface moisture and reducing the time moisture remains on the surface, thereby lowering the risk of protein thermal denaturation. In the uniform drying section (55℃ / 95 min): a 2.98 μm wavelength is used, and the power density is further reduced to 0.09 W / cm². 2 The radiation distance is 240 mm, and the hot air velocity is increased to 2.8 m / s. This stage utilizes strong convection to quickly remove moisture diffused from the interior to the surface, while maintaining gentle internal heat with extremely low infrared power to prevent the chicken breast from becoming dry due to prolonged heating. The dehumidification strategy involves dehumidifying for 40 seconds every 9 minutes to maintain low humidity within the cavity. The constant temperature color-fixing stage (53℃ / 40 min): the power density is reduced to 0.08 W / cm², the radiation distance is 220 mm, and the temperature is 53℃. Color fixing and final dehydration are completed at a lower temperature to prevent excessive Maillard reaction that could cause the surface to char, while maintaining the tender white texture of the meat.

[0072] Key controls during the drying process: During the rapid dehydration section, infrared thermal imaging monitoring kept the surface temperature of the meat slices stable at 54-56℃, with no hot spots observed (chicken is extremely sensitive to heat; significant denaturation and shrinkage occur above 60℃). Surface humidity sensors showed that the relative humidity was controlled at 62%-72%, indicating that strong airflow and intensified dehumidification effectively reduced surface humidity and inhibited the formation of a denatured protein layer. In the uniform drying section, samples were taken every 20 minutes for observation: the surface of the meat slices was light yellow with no burnt brown spots; it was elastic when lightly pressed by hand, without a dry or hard feel; the cross-section was uniformly moist with no white denatured hard shell. The actual temperature fluctuation throughout the process was ≤±2℃, and the time correction rule was not triggered.

[0073] Drying Results and Quality Evaluation: Total drying time was 212 minutes (approximately 3.53 hours). The final product had a moisture content of 17.5% (wet basis), uniform surface color, and tender, non-dry texture. Moisture uniformity: The standard deviation / mean of moisture content in different locations within the same tray was 3.0%, with no dry exterior and wet interior. Shear force: 57.82 N, hardness: 44.36 N (the suitable shear force for chicken jerky is typically 50-70 N), significantly lower than the 80-105 N of traditional hot air drying, resulting in a tender texture without a woody feel. Chewability: 32.45 N, fine and easy to chew. Protein retention: 94.2% (traditional hot air drying only reaches 55-65%, due to prolonged high temperatures causing significant protein denaturation and loss), thanks to a gentle and mild drying strategy, minimal damage to the protein structure was observed. Nutrient retention (vitamins): 92.8% (heat-sensitive vitamins are excellently preserved under low-temperature, short-time processing). Fat oxidation value (TBARS value): 0.15 mg MDA / kg (far below the limit of 0.5 mg MDA / kg; due to the low-fat nature of chicken combined with temperature control, oxidation is minimal). Cross-sectional observation: The meat is uniformly light white, with clear muscle fibers but no excessive contraction, and no charred or hardened layers, indicating moderate protein denaturation.

[0074] Another control group (conventional hot air drying): constant temperature 60℃, air velocity 2.0 m / s, no infrared radiation. Results: after 7.5 hours of drying, the moisture content was still 21.4% (not up to standard), the surface was obviously hard and dry, the color was dark yellow, the shear force was 88.63 N, the hardness was 72.53 N (too hard, with a noticeable woody feel), and the protein retention rate was 58%. Another control group (using beef drying parameters, i.e., higher power, near-field infrared radiation): using the beef drying parameters in Example 1 (power density of 0.16 W / cm³ in the rapid dehydration section). 2The drying parameters were: radiation distance 200 mm, wavelength 2.90 μm, hot air velocity 2.0 m / s, and drying time halved for chicken. Results: After 35 minutes of drying, multiple charred brown spots appeared on the surface of the meat slices, and infrared thermal imaging showed a local temperature of 68℃. The final product had a moisture content of 18.2%, but the shear force was as high as 106.58 N (severe protein denaturation leading to a hard texture), with a protein retention rate of only 61%. The surface was charred, and the quality was substandard. A control group (using pork drying parameters, i.e., low power, long distance, and high air velocity) was also set up: the pork drying parameters from Example 2 were used (rapid dehydration section power density 0.11 W / cm³). 2 The radiation distance was 350 mm, the wavelength was 2.85 μm, and the hot air velocity was 2.4 m / s. Results: The total drying time was extended to 330 min (5.5 h), the final product had a moisture content of 18.5%, but the surface was slightly hardened (due to excessive drying by hot air under low infrared power), the shear force was 94.36 N, the hardness was 78.92 N, and the protein retention rate was 82%. Although it was better than hot air drying, it was not as tender as the product in Example 3.

[0075] Example 3 successfully achieved "flexible, mild, and short-time" intelligent drying of chicken meat through a four-stage program output by a precise drying kinetic model, characterized by low infrared power density, shallow penetration wavelength, high hot air velocity, and intensified dehumidification. This effectively prevented excessive protein denaturation and surface browning. The final product had a shear force of 57.82 N, a hardness of 44.36 N, a protein retention rate of 94.2%, tender meat, and excellent overall quality.

[0076] Example 4 (Mixed Meat) This product uses a blend of fresh beef hind leg and fresh pork hind leg meat in a 7:3 mass ratio as raw material. The properties of this blend fall between those of beef and pork: it retains the high fiber content and water-holding capacity of beef (prone to surface crusting and difficulty in internal moisture escape), while incorporating the higher fat content of pork (prone to melting upon heating and oxidizing). Traditional drying methods that simply apply parameters specific to either beef or pork will result in quality defects. Applying beef parameters (high power, close distance, low airflow) will cause severe fat oxidation and oil release; applying pork parameters (low power, long distance, high airflow) will fail to provide sufficient heat penetration, resulting in inadequate moisture removal from the beef portion and a product that is dry on the outside but wet on the inside.

[0077] Raw material preparation and physical property parameter acquisition: Fresh beef hind leg and pork hind leg were selected, and visible fascia and fat were removed from each. They were then mixed in a 7:3 mass ratio, minced, and reshaped into 7 mm thick slices (simulating the actual production scenario of meat reorganization or mixed slicing in mixed jerky production). Physical property parameters of the mixed raw meat were determined: the initial moisture content was 74.3% (approximately 2.89 kg water / kg dry matter on a dry basis) determined by gravimetric method; the fat content was 9.7% determined by Soxhlet extraction; the protein content was 21.5% determined by Kjeldahl nitrogen determination; and the fiber content was 16.7% determined by fiber separation method. The target moisture content was set at 18.0% (wet basis). The raw meat variety was designated as mixed meat and coded as D. The beef:pork = 7:3 mixing ratio was also input into the model as an additional physical property parameter.

[0078] The drying kinetics model outputs the drying program: The "mixed meat," the mixing ratio (0.7 beef / 0.3 pork), and the aforementioned six physical property parameters are input into the precise drying kinetics model. The four-stage mid-infrared-hot air synergistic drying program output by the model is shown in Table 4.

[0079] Table 4 Drying parameters for mixed meat (beef:pork = 7:3) Note: The dehumidification strategy in the table uses time-cycle control. The reference settings for humidity control are as follows: Rapid dehydration stage: dehumidification threshold 64%RH, hysteresis 4%RH (corresponding to dehumidification for 32s every 7min); Uniform drying stage: dehumidification threshold 69%RH, hysteresis 5%RH (corresponding to dehumidification for 45s every 15min); Constant temperature color fixing stage: dehumidification threshold 72%RH, hysteresis 7%RH (corresponding to 5min on, 2min off).

[0080] As shown in Table 4, for mixed meat (beef:pork = 7:3): Therefore, Example 4, based on the parameters output by the precise drying kinetic model, implemented a balanced and segmented adaptation strategy: In each drying stage, based on the comprehensive physical properties of the mixed meat, the infrared power density, radiation distance, wavelength, and hot air velocity were optimized through trade-offs. This ensured sufficient moisture release from the beef portion while controlling excessive overheating of the pork fat, achieving uniform drying and good quality of the mixed meat jerky. Specifically: Preheating stage (44℃ / 18 min): Infrared power density 0.11 W / cm³ 2 The radiation distance is 400 mm, and the hot air velocity is 1.8 m / s. Gentle heating is used to avoid sudden heating of the beef surface and premature melting of pork fat. The rapid dehydration section (60℃ / 82 min) uses a 2.88 μm wavelength to balance water molecule absorption (around 2.90 μm) and avoid the strong fat absorption region (3.2~3.5 μm). The power density is 0.14 W / cm³. 2The radiation distance is 280 mm. This compromise parameter ensures deep penetration heating of the muscle fibers in the beef portion (slightly shallower than pure beef, but still effectively driving away internal moisture due to the reduced overall water retention of the mixed meat) without overheating the fat in the pork portion. The hot air velocity is 2.2 m / s to help remove surface moisture and a small amount of exuded oil. The dehumidification frequency is increased to 32 s every 7 minutes to promptly remove high-humidity gases. The uniform drying section (58℃ / 112 min) uses a 2.98 μm wavelength, with the power density reduced to 0.12 W / cm². 2 Hot air velocity 2.0 m / s. The drying time is appropriately extended to allow sufficient time for moisture in the beef portion of the mixed meat to diffuse to the surface, while preventing the surface fat of the pork portion from concentrating due to excessive moisture evaporation. The air velocity is maintained at 2.0 m / s to avoid excessively high air velocities causing the beef surface to dry too quickly. Constant temperature color setting section (56℃ / 52 min): Power density 0.10 W / cm³. 2 The radiation distance was 290 mm, and the temperature was 56℃ (between 58℃ for beef and 55℃ for pork). Color fixing was completed at a relatively low temperature, balancing the color formed by the Maillard reaction in beef and the control of fat oxidation in pork. The dehumidification strategy was intermittent dehumidification.

[0081] Key controls during the drying process: During the rapid dehydration phase, infrared thermal imaging monitoring showed that the surface temperature of the meat slices remained stable at 57-60℃, with no localized hot spots observed. The temperature difference between the beef fiber area and the fat particle area was ≤2℃, indicating uniform infrared energy distribution and no preferential overheating of fat. The cross-section of the meat slices was observed. During the uniform drying phase, samples were taken every 25 minutes for observation: the overall moisture content of the mixed meat decreased uniformly, with no instances of the beef being dry on the outside but wet on the inside, or the pork showing signs of oil exudation. The actual temperature fluctuation throughout the process was ≤±2.5℃, and the time correction rule was not triggered.

[0082] Drying Results and Quality Evaluation: Total drying time: 264 min (approximately 4.4 h). The final product had a moisture content of 17.5% (wet basis), a soft surface without hardening, and a uniform internal texture. Moisture uniformity: The standard deviation / mean of moisture content in meat slices at different locations within the same tray was 3.3%, with no dry exterior and wet interior, and the moisture content difference between the beef and pork components was <2%. Shear force: 105.54 N, hardness: 87.67 N, consistent with the expected texture of the mixed meat jerky. Chewiness: 55.08 N, moderate. Fat oxidation value (TBARS value): 0.24 mg MDA / kg (below the limit of 0.5 mg MDA / kg, indicating that fat oxidation in the pork portion of the mixed meat was effectively controlled, with no rancid taste). Protein retention: 92.0% (approximately 62-70% for traditional hot air drying of mixed meat). Nutrient retention (vitamins): 91.9%. Cross-sectional observation: The interior of the mixed meat is uniformly brownish-red, the beef fibers are well integrated with the pork matrix, there are no fat liquefaction cavities or oil spots, and no white dry heart.

[0083] A control group 1 was set up (using pure beef drying parameters, i.e., the parameters of Example 1): power density of the rapid dehydration section was 0.16 W / cm³. 2 The radiation distance was 200 mm, the wavelength was 2.90 μm, and the hot air velocity was 2.0 m / s. Results: After 50 minutes of drying, the surface of the meat slices showed obvious oiliness, and infrared thermal imaging showed that the local temperature of the pork fat area reached 67℃; the final product had a moisture content of 17.8%, but the fat oxidation value was as high as 0.53 mg MDA / kg (exceeding the standard), with oil spots on the surface and a slight rancid odor, indicating substandard quality. Control 2 (using pure pork drying parameters, i.e., the parameters of Example 2): rapid dehydration section power density 0.11 W / cm³. 2 The radiation distance was 350 mm, the wavelength was 2.85 μm, and the hot air velocity was 2.4 m / s. Results: The total drying time was extended to 320 min (5.3 h), and the final product moisture content was 18.2%. However, the moisture uniformity of the beef portion was poor (coefficient of variation 3.8%), with some fibrous areas of the beef still being too moist (moisture content 21%), while the pork portion met the standard; the protein retention rate was 86%. This indicates that low-power, long-distance infrared radiation has insufficient penetration and cannot effectively remove moisture from the beef portion of the mixed meat.

[0084] Example 4 successfully achieved a balanced and flexible intelligent drying process for a 7:3 beef:pork mixture by using a combination of infrared power density, radiation distance, wavelength, and hot air velocity in a four-segment program output by a precise drying kinetic model. This process balanced deep-penetration dehydration of the beef portion and oil control and anti-oxidation of the pork portion. The results showed excellent overall quality with shear force of 105.54 N, hardness of 87.67 N, chewiness of 55.08 N, fat oxidation value of 0.24 mg MDA / kg, and protein retention of 92.0%.

[0085] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A multi-stage flexible intelligent drying method for dried meat products, characterized in that, include: S1. Construct a drying kinetic model. The drying kinetic model takes the variety of raw meat, the physical properties of raw meat, and the target moisture content as inputs, and outputs a set of parameters for a four-stage mid-infrared-hot air synergistic drying program. The physical properties of raw meat include the thickness, fat percentage, fiber content, protein content, and initial moisture content. The four-stage mid-infrared-hot air synergistic drying program includes a preheating stage, a rapid dehydration stage, a uniform drying stage, and a constant temperature color fixing stage, which are connected sequentially. The parameter set for each stage includes the mid-infrared wavelength and / or mid-infrared power density, mid-infrared heating temperature, mid-infrared radiation distance, hot air velocity, drying time, and dehumidification strategy. S2. Using a digital twin model constructed based on the physical field simulation of mid-infrared-hot air synergistic drying, generate a four-segment mid-infrared-hot air synergistic drying program and its simulated physical performance index corresponding to different varieties, different physical property parameters and different drying target moisture content combinations. Based on the simulated physical performance index, determine the drying program with optimal physical performance and pre-train the initial drying kinetic model. S3. Using real mid-infrared-hot air synergistic drying test data as samples, the optimal drying procedure determined based on the overall comprehensive score of multi-dimensional quality indicators is used to fine-tune the pre-trained drying kinetics model through transfer learning to obtain an accurate drying kinetics model. S4. Obtain the variety of raw meat to be dried, the physical property parameters of the raw meat, and the target moisture content for drying. Execute the four-segment mid-infrared-hot air synergistic drying program parameter set output by the precise drying kinetic model obtained in step S3 until the target moisture content is reached. Among them, the simulated physical performance indicators in step S2 include water activity, moisture uniformity, drying uniformity, hot air efficiency, mid-infrared utilization rate, and total drying time; the multi-dimensional quality indicators in step S3 include shear force, chewiness, protein retention rate, nutrient retention rate, and fat oxidation value.

2. The multi-stage flexible intelligent drying method for dried meat products as described in claim 1, characterized in that, This also includes online evolutionary training of the precise drying kinetics model obtained in step S3, specifically: During the drying process, for each batch of raw meat to be dried, the measured values ​​of physical performance indicators and multi-dimensional quality indicators fed back from the actual drying process are collected to calculate the reward function value. The precise drying kinetic model is updated using a deep deterministic strategy gradient algorithm, so that the set of parameters of the four-segment mid-infrared-hot air synergistic drying program output by the precise drying kinetic model adaptively evolves with the increase of production batches. Among them, the reward function is the weighted sum of the weighted comprehensive score of multi-dimensional quality indicators and the reciprocal of the total drying time.

3. The multi-stage flexible intelligent drying method for dried meat products as described in claim 2, characterized in that, The construction of the digital twin model in step S2 includes: Establish a coupled set of equations for the mid-infrared radiation heat transfer model, the hot air convection mass transfer model, the moisture diffusion model, and the solid mechanics contraction model; Using the raw meat variety and physical property parameters as input, and the four-stage synergistic drying process parameter set as adjustable boundary conditions, the spatiotemporal distribution of the temperature field, humidity field, water activity field and stress field of the raw meat during the drying process is obtained by solving the coupled equation set through finite element method. The spatiotemporal distributions of the temperature field, humidity field, water activity field, and stress field are statistically or integrally calculated to obtain the simulated physical performance indicators, including: water activity obtained from the spatial average of the water activity field; water uniformity obtained from the spatial variation coefficient of the humidity field; drying uniformity obtained from the time difference of different regions reaching the critical moisture content; hot air efficiency obtained from the ratio of surface moisture evaporation to the total heat input of hot air; mid-infrared utilization rate obtained from the ratio of infrared radiation energy absorbed by the raw meat to the total emitted radiation energy; and total drying time obtained from the simulation time taken to reduce the moisture content to the target value.

4. The multi-stage flexible intelligent drying method for dried meat products as described in claim 3, characterized in that, The coupled equation set includes: The mid-infrared radiation heat transfer model is based on Lambert-Beer's law and describes the exponential decay of infrared radiation along the thickness direction inside the raw meat and the heat source term. The hot air convection mass transfer model is based on Newton's cooling formula and Fick's surface diffusion law, which couples hot air with the convective heat transfer and moisture evaporation flux on the meat surface. The moisture diffusion model adopts Fick's second law, with the moisture content gradient as the driving force, and introduces temperature and shrinkage correction terms; The solid mechanics shrinkage model is based on the linear viscoelastic constitutive relation, which equates the volume shrinkage caused by water loss to strain and applies it to the stress field.

5. The multi-stage flexible intelligent drying method for dried meat products as described in claim 4, characterized in that, Step S2, which involves pre-training the initial drying kinetics model, specifically includes: For each set of virtual raw meat varieties and physical property parameter combinations, the parameters of each segment of the four-segment infrared-hot air synergistic drying program are used as decision variables, and the simulated physical performance index is used as the optimization objective. The Pareto front is solved using a multi-objective optimization algorithm, and then the unique drying program with the best physical performance is selected from the Pareto front based on the weighted comprehensive score as the label of the virtual sample. The initial drying kinetic model is pre-trained using a supervised learning approach. The variety and physical properties of the raw meat are used as inputs, and the parameters of each segment of the drying process with optimal physical performance are used as output labels. By minimizing the mean absolute error loss function, the weights and biases of the initial drying kinetic model are iteratively updated until convergence, thus completing the pre-training of the initial drying kinetic model.

6. The multi-stage flexible intelligent drying method for dried meat products as described in claim 5, characterized in that, In step S3, the overall comprehensive score based on multi-dimensional quality indicators is calculated in the following way: Shear force, chewability, protein retention rate, nutrient retention rate, and fat oxidation value were normalized and then weighted and summed according to preset weight coefficients to obtain the overall score. The weight coefficients were determined by the analytic hierarchy process, with protein retention rate having the highest weight and fat oxidation value having the second highest weight.

7. The multi-stage flexible intelligent drying method for dried meat products as described in claim 6, characterized in that, The methods for obtaining real mid-infrared-hot air synergistic drying test data in step S3 include: Raw meat samples of different varieties and different combinations of physical property parameters were selected. On a mid-infrared-hot air co-drying equipment, multiple drying tests with different combinations of drying program parameters were performed on each group of raw meat samples to obtain the measured values ​​of multi-dimensional quality indicators of the final dried products of each test group. The overall score is calculated based on the measured values ​​of the multi-dimensional quality indicators. The drying program corresponding to the group with the highest overall score from multiple tests of the same raw meat sample is selected as the label of the optimal drying program for the raw meat sample.

8. The multi-stage flexible intelligent drying method for dried meat products as described in claim 7, characterized in that, Step S3 involves fine-tuning the pre-trained drying dynamics model using transfer learning, specifically as follows: The drying dynamics model pre-trained in step S2 is used as the source domain model, and the network structure and pre-trained weights of the source domain model remain unchanged. The feature extraction layer of the source domain model is frozen so that its weights are not updated during fine-tuning, in order to preserve the basic mapping relationship between the physical property parameters learned from the digital twin virtual samples and the drying procedure. Using the variety and physical property parameters of raw meat in real drying test data as input, and the parameters of each segment of the corresponding optimal quality drying program as output labels, only the weights of the output layer or the last N fully connected layers of the source domain model are updated, so that the model is adapted from the mapping relationship of optimal physical performance to the mapping relationship of optimal quality, thus obtaining the accurate drying kinetic model. In the transfer learning fine-tuning, a fine-tuned learning rate lower than that in the pre-training stage of step S2 is used for the unfrozen layers, and iterative training is performed by minimizing the mean squared error loss function until the loss converges on the validation set of real experimental data.

9. The multi-stage flexible intelligent drying method for dried meat products as described in claim 8, characterized in that, In step S4, when executing the four-segment mid-infrared-hot air synergistic drying program parameter set output by the precise drying kinetic model, the following process control rules are also included: During any of the preheating, rapid dehydration, uniform drying and constant temperature color fixing sections, the actual temperature inside the cavity or on the surface of the material is collected in real time by a temperature sensor. When the actual temperature in this segment continuously exceeds the upper limit deviation value of the program-set temperature for this segment, and the cumulative duration exceeds the preset duration threshold, the program-set drying time for this segment will be shortened by a preset ratio based on the original value. When the actual temperature in this segment is consistently lower than the lower limit deviation value of the program-set temperature for this segment, and the cumulative duration exceeds the preset duration threshold, the program-set drying time for this segment will be extended by the preset ratio based on the original value. After adjustment, continue operating this section according to the revised drying time; The upper and lower deviation values ​​are 2-3℃, and the preset ratio is 5-8%.

10. A system based on the multi-stage flexible intelligent drying method for dried meat products according to any one of claims 1 to 9, characterized in that, include: The drying chamber is used to hold the raw meat to be dried and to provide a sealed and insulated environment for the drying process. A mid-infrared heating device is installed inside the drying chamber to generate mid-infrared radiation for internal heating of the raw meat. The infrared wavelength and / or mid-infrared power density, mid-infrared heating temperature and mid-infrared radiation distance are all adjustable. A hot air device is installed in the drying chamber to generate circulating hot air for external dehydration of the raw meat by convection, and the hot air speed is adjustable. A dehumidification device, connected to the drying chamber, is used to remove moisture from the chamber according to a dehumidification strategy. Temperature sensors are used to collect the actual temperature inside the drying chamber or on the surface of the raw meat during the drying process in real time; The control system is connected to the mid-infrared heating device, hot air device, dehumidification device, and temperature sensor; the control system has a built-in precise drying kinetic model, which is constructed and trained according to steps S1 to S3 of the method according to any one of claims 1 to 8; The control system is configured as follows: Obtain the variety and physical property parameters of the raw meat to be dried, including the thickness, fat percentage, fiber content, protein content, initial moisture content, and target moisture content of the raw meat; The variety and physical property parameters are input into the precise drying kinetic model, and the precise drying kinetic model outputs a set of parameters for a four-stage mid-infrared-hot air synergistic drying program. The four-stage mid-infrared-hot air synergistic drying program includes a preheating stage, a rapid dehydration stage, a uniform drying stage, and a constant temperature color fixing stage connected in sequence. The parameter set for each stage includes mid-infrared wavelength and / or mid-infrared power density, mid-infrared heating temperature, mid-infrared radiation distance, hot air velocity, drying time, and dehumidification strategy. According to the four-stage mid-infrared-hot air synergistic drying program parameter set, the mid-infrared heating device and the hot air device are controlled to operate in stages, and the dehumidification device is controlled to execute the dehumidification strategy until the raw meat reaches the target moisture content.

Images