A freeze-dried particle intelligent production control system and method based on an internet of things

Abstract

The application provides a lyophilized particle intelligent production control system and method based on the Internet of Things, and belongs to the lyophilized particle production technical field.The control system comprises a multi-source sensing unit and a data processing unit connected with the multi-source sensing unit;the multi-source sensing unit is arranged on a near-infrared spectrum sensor, an optical fiber temperature sensor, a vacuum degree sensor and a visual camera of a lyophilization device, and is used for collecting material temperature gradient change rate, crystallinity data and vacuum fluctuation amplitude;the data processing unit comprises a mapping module, a decision module and an optimization module;the mapping module converts the collected data into a mass transfer resistance coefficient K, and the conversion process comprises the following steps: step 1: according to the material formula type, the weight factor combination of the temperature gradient change rate, the crystallinity logarithmic transformation value and the vacuum fluctuation amplitude is selected.The application realizes real-time monitoring, dynamic optimization and quality tracing of the lyophilization process, and improves the production efficiency and product consistency.

Description

Technical Field

[0001] This invention relates to the field of freeze-drying technology, specifically to an intelligent production control system and method for freeze-dried granules based on the Internet of Things. Background Technology

[0002] Freeze-drying, also known as lyophilization, is a process used to dry biological materials such as pharmaceuticals, proteins, and enzymes, as well as heat-sensitive materials. Freeze-drying achieves its drying effect by the sublimation of ice crystals into water vapor, directly converting water from a solid to a gaseous state. Freeze-drying is typically performed under vacuum conditions, but can also be carried out at varying atmospheric pressures.

[0003] Chinese Patent Publication No. CN103917842A discloses a processing line for producing freeze-dried granules under closed conditions. The processing line includes at least the following independent devices: a spray chamber for generating droplets and freezing the droplets to form granules, and a bulk freeze dryer for freeze-drying the granules. A conveying section is provided to convey the product from the spray chamber to the freeze dryer. In order to produce granules under end-to-end closed conditions, each of the devices and the conveying section is independently adapted for closed operation, and the spray chamber is adapted to separate the droplets from any cooling circuit.

[0004] Existing freeze-drying control systems rely on fixed process curves and cannot respond to changes in material state in real time, resulting in control lag. They also rely on manual experience to set process parameters, which cannot respond to changes in material state in real time, leading to fluctuations in product quality. At the same time, differences in material characteristics (such as the freeze-drying behavior of proteins and polysaccharides) make traditional methods difficult to apply universally, and process parameter adjustments rely on trial and error, which is time-consuming and costly. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent production control system and method for freeze-dried granules based on the Internet of Things, so as to realize real-time monitoring, dynamic optimization and quality traceability of the freeze-drying process, and improve production efficiency and product consistency.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0007] An IoT-based intelligent production control system for freeze-dried granules includes a multi-source sensing unit and a data processing unit connected to the multi-source sensing unit.

[0008] Among them, the multi-source sensing unit is deployed in the near-infrared spectral sensor, fiber optic temperature sensor, vacuum sensor and vision camera of the freeze-drying equipment to collect material temperature gradient change rate, crystallinity data and vacuum fluctuation amplitude.

[0009] The data processing unit includes a mapping module, a decision module, and an optimization module; the mapping module converts the collected data into a mass transfer resistance coefficient K, and the conversion process includes the following steps:

[0010] Step 1: Select the weighting factor combination of temperature gradient change rate, logarithmic transformation value of crystallinity, and vacuum fluctuation amplitude according to the material formulation type;

[0011] Step 2: Linearly superimpose the weighted temperature gradient change rate, the logarithmic transformation value of crystallinity, and the amplitude of vacuum fluctuation to generate the K value;

[0012] The decision-making module is used to trigger a reinforcement learning algorithm to adjust the freeze-drying process curve when the real-time K value exceeds the dynamic threshold; the optimization module is used to aggregate historical K values ​​to optimize the weight factor library through federated learning and drive digital twin simulation.

[0013] Furthermore, the control system also includes an ice crystal morphology recognition module, which captures the fractal dimension of ice crystals using a high-resolution visual camera and generates a morphology correction coefficient. The mapping module introduces a morphology correction term when calculating the mass transfer resistance coefficient.

[0014] Furthermore, the weighting factor combination method is as follows: the temperature gradient change rate accounts for 40-70% of the weight in the material, crystallinity accounts for 20-50%, and vacuum fluctuation accounts for 5-20%. The dynamic threshold is automatically adjusted within the range of 0.8-1.2 according to the freeze-drying stage.

[0015] Furthermore, the mass transfer resistance coefficient K exhibits a strong negative correlation with the freeze-drying rate, with a correlation coefficient ≥0.95. Its calculation accuracy error is 3-5%, and the edge processing delay is 40-50ms.

[0016] Furthermore, the decision-making module reinforces the temperature control strategy to approximate the target K value as the core optimization direction, while introducing an energy consumption penalty mechanism to adjust the temperature curve.

[0017] Furthermore, the data processing unit also includes a phase adaptation module, which identifies glass transition regions based on the spatial distribution relationship between crystallinity and temperature gradient. When the proportion of glass regions exceeds 15%, the weight of crystallinity data in the K-value calculation is automatically increased.

[0018] This invention also discloses an intelligent production control method for freeze-dried granules based on the Internet of Things (IoT), comprising using the aforementioned intelligent production control system for freeze-dried granules based on the IoT, with the specific steps as follows:

[0019] S101: Real-time acquisition of raw data on material temperature gradient change rate, crystallinity, and vacuum fluctuation amplitude;

[0020] S102: Based on the current material formulation type, match the preset weighting factor, and perform weighted superposition of the three element data to generate the mass transfer resistance coefficient K;

[0021] S103: Compare the real-time K value with the stage dynamic threshold. If the K value continues to exceed the standard, the heating rate is dynamically suppressed or the vacuum degree is compensated through reinforcement learning. If the K value continues to be low, the energy consumption optimization mode is activated.

[0022] S104: Optimize weight factor configuration rules using federated learning based on historical K-value datasets.

[0023] The control logic of S103 is as follows: when the K value exceeds the upper limit of the threshold for 3 consecutive minutes, the cooling amplitude is linearly increased according to the excess ratio; when the K value is below the lower limit of the threshold for 10 minutes, the vacuum compensation program is started and the compensation amount is proportional to the reciprocal of the K value.

[0024] Furthermore, the present invention also includes S105, establishing a method for linking K value and quality, statistically analyzing the variance of K value fluctuation at the end point of batch freeze-drying, automatically determining the batch as unqualified when the variance value is >0.05, and predicting the remaining service life of the vacuum pump based on the K value fluctuation frequency characteristics.

[0025] Furthermore, a K-value benchmark was established during the pre-freezing stage at -38℃. ～ The weighted average of crystallinity calculated at -40℃ is used as the K0 benchmark, and the real-time K value is set to deviate from the K0 benchmark by 15-25% as the process abnormality alarm threshold.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] This invention achieves dynamic adjustment of freeze-drying process parameters through multi-source data fusion and intelligent algorithms, reducing quality fluctuations and effectively improving product consistency. Furthermore, this invention unifies multi-source data dimensions using a K-value to control response latency and introduces energy consumption penalties through reinforcement learning, resulting in overall energy consumption reduction. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0029] Figure 1 This is a block diagram illustrating the overall principle of the control method described in this invention. Detailed Implementation

[0030] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive. The following description is in conjunction with the accompanying drawings. Figure 1 The embodiments of the present invention will be described in detail below.

[0031] Example 1: This example discloses an intelligent production control system for freeze-dried granules based on the Internet of Things, including a multi-source sensing unit and a data processing unit connected to the multi-source sensing unit;

[0032] Among them, the multi-source sensing unit is deployed in the near-infrared spectral sensor, fiber optic temperature sensor, vacuum sensor and vision camera of the freeze-drying equipment to collect material temperature gradient change rate, crystallinity data and vacuum fluctuation amplitude.

[0033] The data processing unit includes a mapping module, a decision module, and an optimization module; the mapping module converts the collected data into a mass transfer resistance coefficient K, and the conversion process includes the following steps:

[0034] Step 1: Select the weighting factor combination of temperature gradient change rate, logarithmic transformation value of crystallinity, and vacuum fluctuation amplitude according to the material formulation type;

[0035] Step 2: Linearly superimpose the weighted temperature gradient change rate, the logarithmic transformation value of crystallinity, and the amplitude of vacuum fluctuation to generate the K value;

[0036] The decision-making module is used to trigger a reinforcement learning algorithm to adjust the freeze-drying process curve when the real-time K value exceeds the dynamic threshold; the optimization module is used to aggregate historical K values ​​to optimize the weight factor library through federated learning and drive digital twin simulation.

[0037] In this embodiment, the multi-source sensing unit transmits data to the mapping module, the decision module outputs control commands to the freeze-drying equipment based on the K value, and the optimization module updates the weight factor library through federated learning.

[0038] The weighting factor combination method is as follows: the temperature gradient change rate accounts for 40-70% of the weight, crystallinity accounts for 20-50%, and vacuum fluctuation accounts for 5-20%. The dynamic threshold is automatically adjusted within the range of 0.8-1.2 according to the freeze-drying stage.

[0039] In this embodiment, the weights of temperature gradient change rate, crystallinity, and vacuum fluctuation in the material are 60%, 30%, and 10%, respectively.

[0040] The mass transfer resistance coefficient K exhibits a strong negative correlation with the freeze-drying rate, with a correlation coefficient ≥ 0.95. Its calculation accuracy error is 3-5%, and the edge processing delay is 40-50ms.

[0041] In this embodiment, the calculation accuracy error is 4% and the edge processing delay is 45ms.

[0042] Among them, the decision-making module uses reinforcement learning to control temperature with the core optimization direction of approximating the target K value, while introducing an energy consumption penalty mechanism to adjust the temperature curve.

[0043] Furthermore, the data processing unit also includes a phase adaptation module, which identifies glass transition regions based on the spatial distribution relationship between crystallinity and temperature gradient. When the proportion of glass regions exceeds 15%, the weight of crystallinity data in the K-value calculation is automatically increased.

[0044] The specific steps are as follows:

[0045] S101: Real-time acquisition of raw data on material temperature gradient change rate, crystallinity, and vacuum fluctuation amplitude;

[0046] S102: Based on the current material formulation type, match the preset weighting factor, and perform weighted superposition of the three element data to generate the mass transfer resistance coefficient K;

[0047] S103: Compare the real-time K value with the stage dynamic threshold. If the K value continues to exceed the standard, the heating rate is dynamically suppressed or the vacuum degree is compensated through reinforcement learning. If the K value continues to be low, the energy consumption optimization mode is activated.

[0048] S104: Optimize weight factor configuration rules using federated learning based on historical K-value datasets.

[0049] The control logic of S103 is as follows: when the K value exceeds the upper limit of the threshold for 3 consecutive minutes, the cooling amplitude is linearly increased according to the excess ratio; when the K value is below the lower limit of the threshold for 10 minutes, the vacuum compensation program is started and the compensation amount is proportional to the reciprocal of the K value.

[0050] Furthermore, this embodiment also includes S105, establishing a method for associating K value with quality, statistically analyzing the variance of the K value fluctuation at the end point of batch freeze-drying, automatically determining the batch as unqualified when the variance value is >0.05, and predicting the remaining vacuum service life based on the K value fluctuation frequency characteristics.

[0051] In the pre-freezing stage, a K-value benchmark was established at -38℃. ～ The weighted average of crystallinity calculated at -40℃ is used as the K0 benchmark, and the real-time K value is set to deviate from the K0 benchmark by 15-25% as the process abnormality alarm threshold.

[0052] In practical applications, for the freeze-drying control of protein materials, a certain monoclonal antibody preparation is equipped with a near-infrared spectral sensor with a wavelength range of 900-1700nm, a fiber optic temperature sensor with an accuracy of ±0.1℃, a vacuum sensor with a resolution of 0.01mbar, and a vision camera with a frame rate of 30fps.

[0053] Weighting factors: temperature gradient change rate 60%, crystallinity 30%, vacuum fluctuation 10%.

[0054] Dynamic thresholds: 0.8 for pre-freezing stage, 1.0 for sublimation stage, and 1.2 for parsing stage.

[0055] K0 benchmark: The weighted average of crystallinity at -38℃ during the pre-freezing stage is used as the benchmark. An alarm is triggered when the real-time K value deviates by 15%.

[0056] The specific control process is as follows:

[0057] S10 1: The sensor collects data in real time. The temperature gradient change rate is calculated by the fiber optic sensor array. The crystallinity is predicted by the near-infrared spectral data through the PLS model. The vacuum fluctuation amplitude is obtained by high-frequency sampling (10Hz) of the vacuum degree sensor.

[0058] S102: The mapping module generates K values ​​by weighting and superimposing them, with the calculation accuracy error controlled within 5% and the edge processing delay 45ms.

[0059] S103: When the K value exceeds 1.05 for 3 consecutive minutes during the sublimation stage, the decision module starts the reinforcement learning algorithm, reduces the heating plate temperature from 25℃ to 20℃, and compensates the vacuum degree to 0.2mbar (originally set to 0.3mbar). The energy consumption penalty mechanism simultaneously reduces the temperature adjustment rate to avoid excessive energy consumption. When the proportion of the glassy region drops below 10%, the original weight factor is restored.

[0060] S104: The optimization module aggregates historical K values ​​through federated learning and finds that the crystallinity weight can be finely adjusted to 35% in the later stage of sublimation, which further improves the freeze-drying rate, with a correlation coefficient ≥0.95.

[0061] In this embodiment, the temperature gradient change rate is 0.15℃ / s, the crystallinity is 0.85, and the vacuum fluctuation amplitude is 12Pa. The vacuum fluctuation amplitude needs to be divided by the reference value of 10Pa for dimensionless conversion.

[0062] Calculate the K value:

[0063]

[0064] in, The temperature gradient change rate is given by log(A), which is the crystallinity, dimensionless, 0-1, and ΔP is the normalized vacuum fluctuation amplitude, ΔP=P1 / P0, where P1 is the original fluctuation amplitude and P0 is the reference value, specifically 10Pa.

[0065] w1 is the weight of the temperature gradient change rate, w2 is the weight of crystallinity, and w3 is the weight of vacuum fluctuation.

[0066] K=0.15×0.6+log(0.85)×0.3+12×0.1=0.09+(-0.07)+1.2=1.22;

[0067] log(0.85) is rounded to two decimal places.

[0068] At this point, K > threshold 1.2, triggering reinforcement learning cooling:

[0069] ΔT=-2℃×(1.22-1.0)=-0.44℃;

[0070] The vacuum fluctuation amplitude needs to be divided by the reference value (e.g., vacuum fluctuation amplitude / 10Pa) to ensure that the K value is dimensionless; after adjustment, the K value returns to 1.05. After testing, the freeze-drying cycle is shortened by 18%, the product moisture content is reduced from 2.5% to 1.8%, the activity retention rate is increased from 92% to 95%, and energy consumption is reduced by 15%.

[0071] Example 2: This example is basically the same as Example 1, except that in this example, the freeze-drying of polysaccharide materials is controlled, specifically a plant polysaccharide extract, and the freeze dryer configuration is the same as in Example 1.

[0072] In this embodiment, the weighting factors are: temperature gradient change rate 30%, crystallinity 50%, and vacuum fluctuation 20%.

[0073] Phase adaptation: The glass transition region identification threshold is set to 15%, and the crystallinity weight is temporarily increased to 60% after triggering.

[0074] The control process is as follows:

[0075] During the pre-freezing stage, a K0 benchmark is established at -40℃. An alarm is triggered when the real-time K value deviates by 20%, and the pre-freezing time is adjusted from 2 hours to 2.5 hours to avoid large ice crystals that could lead to collapse.

[0076] During the sublimation stage, the phase adaptive module identifies that the glassy region accounts for 18% by the spatial distribution of crystallinity and temperature gradient, automatically increases the crystallinity weight to 65%, and reduces the K value from 1.1 to 0.98 to avoid incomplete drying due to excessive mass transfer resistance.

[0077] The calculated K value is: Crystallinity contribution 0.92 × 0.65 + Temperature gradient contribution 0.10 × 0.3 + Vacuum fluctuation contribution 0.8 × 0.05 = 0.628;

[0078] If K value < 0.8, activate vacuum compensation: ΔP = 0.5 / K = 0.8 Pa;

[0079] In step S105, the batch endpoint K value fluctuation variance is 0.03 (pass threshold 0.05), and it is judged to be qualified; at the same time, based on the K value fluctuation frequency (3 times / hour), the remaining life of the vacuum pump is predicted to be 200 hours, and maintenance is arranged in advance.

[0080] Tests showed that the reconstitution time of freeze-dried products was reduced from 120 seconds to 80 seconds, the clarity of the solution after reconstitution was improved, and the batch failure rate decreased from 1.2% to 0.3%.

[0081] Example 3: This example is mainly used to achieve collaborative optimization of multiple devices. In this example, five freeze dryers are used to achieve collaborative optimization of process parameters through federated learning. The specific process is as follows:

[0082] Each device uploads historical K-values, weighting factors, and process parameters to the central server. The data is then aggregated after being anonymized. The FedAWO algorithm is used to optimize the weighting factor library, and differential privacy technology is used to aggregate data from edge devices. The central server only updates the weighting factor model. The heterogeneity of thermal distribution among different devices is identified, and personalized weighting configurations are generated. The temperature gradient weight of edge devices is increased by 5%. A digital twin of the freeze dryer is constructed to simulate freeze-drying curves under different weighting configurations and predict the optimal parameter combination to adjust the sublimation stage pressure from 0.25 mbar to 0.22 mbar.

[0083] Tests showed that the average freeze-drying cycle of multiple devices was shortened by 12%, the standard deviation of energy consumption decreased from 7.3% to 1.8%, and the consistency of product quality was significantly improved.

[0084] The specific data comparison is as follows:

[0085] index Traditional system This embodiment Batch pass rate 88.5％ 99.1％ Average energy consumption 1.2MW·h 0.94MW·h Abnormal response speed 150ms 45ms

[0086] Example 4: This example is a further optimization based on Example 1. In this example, when the mapping module calculates the mass transfer resistance coefficient K, an unsteady-state mass transfer equation is introduced:

[0087]

[0088] Where A represents crystallinity, α is derived from the material porosity by calculating the rate of change in absorbance in the 1200-1400 nm band of near-infrared spectroscopy, and β is the phase transition correction factor. β = 1 indicates no phase transition, while β = 1 indicates a phase transition when the proportion of the glassy region is greater than 15%. Ea R is the mass transfer activation energy, R is the gas constant, and T is the average temperature of the material.

[0089] Because the high viscosity of liposomes (>50 cP) causes freeze-drying collapse, the surface fluctuation relaxation time τ is captured by a visual camera to calculate the viscosity, η = k·ΔT / τ (k is the calibration coefficient), which is obtained by calibrating a standard with known viscosity in the freeze dryer; when η>50 cP, the crystallinity weight is automatically increased to 70%.

[0090] The control process is as follows:

[0091] During the pre-freezing stage, η = 65cP was detected, and the crystallinity weight increased from 50% to 70%.

[0092] K value calculation: K = 0.12 × 0.2 + log(0.75) × 0.7 + 15 × 0.1 = 1.18;

[0093] The decision module failed because K value > threshold 1.1;

[0094] The starting cooling ΔT = -0.5℃ × (1.18 - 1.1) = -0.4℃.

[0095] Tests showed that the encapsulation rate increased from 85% to 93%, and the collapse rate decreased from 8% to 0.5%.

[0096] Furthermore, due to the abnormally high mass transfer resistance at -50℃, a tunneling effect term is added to the K-value calculation:

[0097] K new =K+δ·e -d / λ ;

[0098] Where δ is the barrier height, δ < 0.1, and d is the thickness of the drying layer. h is Planck's constant, m is the mass of a water molecule, and ΔE is the energy barrier.

[0099] Real-time calculation of d = 0.2 mm (near-infrared inversion), ΔE = 0.15 eV;

[0100] Correction amount δ·e -d / λ =2.5, original K=1.3, corrected K new =1.05; The decision module cancels vacuum compensation because the K-value is within the regression threshold. For mRNA vaccines, the mRNA vaccine activity retention rate is increased from 80% to 95%.

[0101] It should be noted that in practical applications, linear superposition is used for conventional materials, while non-steady-state equations are used for special materials, i.e., materials with viscosity η > 50 cP.

[0102] Example 5: This example is a further optimization based on Example 1. Existing freeze-drying systems cannot monitor the microstructure of ice crystals in real time, causing monoclonal antibody drugs to easily form needle-like ice crystals during the pre-freezing stage, resulting in a fractal dimension D. f <1.6, causing protein denaturation and turbidity after reconstitution, resulting in a product failure rate as high as 15%, fractal dimension D f Calculation using grid counting method.

[0103] In this embodiment, a new high-resolution visual camera with a resolution of 5 μm and a frame rate of 60 fps is added. An ice crystal morphology recognition module is also included, which captures the fractal dimension D of the ice crystals using the high-resolution visual camera. f The mapping module generates a shape correction coefficient γ and introduces a shape correction term K when calculating the mass transfer resistance coefficient. new1 :

[0104] K new1 =K+γ·(1 / D f );

[0105]

[0106] The heating rate is forcibly limited to ≤0.1℃ / min.

[0107] The specific process is as follows;

[0108] Step 1: Data Collection

[0109] The fiber optic temperature sensor has a real-time temperature reading of 0.15℃ / s, a crystallinity of 0.88, and an ice crystal fractal dimension D. f It is 1.55.

[0110] Step 2: K value calculation and correction;

[0111] Basic K-value calculation, weighted as follows: temperature 60%, crystallinity 30%, vacuum 10%;

[0112] K=0.15×0.6+kog(0.88)×0.3+(8Pa / 10Pa)×0.1=0.09-0.017+0.08=0.153;

[0113] Shape correction, D f =1.55, γ=0.2;

[0114] K new1 =0.153+0.2×(1 / 1.55)=0.153+0.129=0.282.

[0115] Step 3:

[0116] The temperature rise rate limit was triggered when Df = 1.55 < 1.6 was detected.

[0117] The original heating curve increased the temperature from -40℃ to -20℃ at a rate of 0.3℃ / min; after adjustment, the heating rate was reduced to 0.1℃ / min, and an ice crystal reconstruction program was simultaneously initiated, injecting nitrogen microfluidics to alter the crystallization path. Historical polymerization data revealed that when Df < 1.6, increasing the crystallinity weight to 40% can reduce the risk of denaturation.

[0118] Based on actual testing, the results of this embodiment and the traditional solution are as follows:

[0119] index Traditional solution This embodiment Increase Product defect rate 15％ 0.5％ ↓96.7％ Protein activity retention 85％ 98％ ↑15.3％ Reconstituted Clarity (NTU) 12.5 1.2 ↓90.4％ freeze-drying cycle 48h 45h ↓6.3％

[0120] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0121] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent production control system for freeze-dried granules based on the Internet of Things, characterized in that: It includes a multi-source sensing unit and a data processing unit connected to the multi-source sensing unit; Among them, the multi-source sensing unit is deployed in the near-infrared spectral sensor, fiber optic temperature sensor, vacuum sensor and vision camera of the freeze-drying equipment, and is used to collect the material temperature gradient change rate, crystallinity data predicted by the PLS model through near-infrared spectral data and vacuum fluctuation amplitude. The data processing unit includes a mapping module, a decision module, and an optimization module; the mapping module converts the collected data into a mass transfer resistance coefficient K, and the conversion process includes the following steps: Step 1: Select the weighting factor combination of temperature gradient change rate, logarithmic transformation value of crystallinity, and vacuum fluctuation amplitude according to the material formulation type; Step 2: Linearly superimpose the weighted temperature gradient change rate, the logarithmic transformation value of crystallinity, and the amplitude of vacuum fluctuation to generate the K value; wherein, the mass transfer resistance coefficient K is strongly negatively correlated with the freeze-drying rate, with a correlation coefficient ≥0.95, and its calculation accuracy error is 3-5% and the edge processing delay is 40-50ms; The decision module is used to trigger a reinforcement learning algorithm that adjusts the freeze-drying process curve when the real-time K value exceeds the dynamic threshold. This algorithm focuses on approximating the target K value as the core optimization direction and introduces an energy consumption penalty mechanism. The optimization module is used to aggregate a library of historical K values ​​from multiple devices or batches through federated learning and drive digital twin simulation.

2. The IoT-based intelligent production control system for freeze-dried granules according to claim 1, characterized in that: The weighting factor combination method is as follows: the temperature gradient change rate accounts for 40-70% of the weight in the material, crystallinity accounts for 20-50%, and vacuum fluctuation accounts for 5-20%. The dynamic threshold is automatically adjusted within the range of 0.8-1.2 according to the freeze-drying stage.

3. The IoT-based intelligent production control system for freeze-dried granules according to claim 1, characterized in that: The data processing unit also includes a phase adaptation module, which identifies glass transition regions based on the spatial distribution relationship between crystallinity and temperature gradient. When the proportion of glass regions exceeds 15%, the weight of crystallinity data in the K-value calculation is automatically increased.

4. The IoT-based intelligent production control system for freeze-dried granules according to claim 1, characterized in that: The control system also includes an ice crystal morphology recognition module, which captures the fractal dimension of ice crystals using a high-resolution visual camera. And generate morphology correction coefficients. The mapping module introduces a shape correction term when calculating the mass transfer resistance coefficient. : ； ； The heating rate is forcibly limited to ≤0.1℃ / min.

5. A smart production control method for freeze-dried granules based on the Internet of Things, characterized in that, The system includes the IoT-based intelligent production control system for freeze-dried granules as described in any one of claims 1-4, with the following specific steps: S101: Real-time acquisition of raw data on material temperature gradient change rate, crystallinity, and vacuum fluctuation amplitude; S102: Based on the current material formulation type, match the preset weighting factor, and perform weighted superposition of the three element data to generate the mass transfer resistance coefficient K; S103: Compare the real-time K value with the stage dynamic threshold. If the K value continues to exceed the standard, the heating rate is dynamically suppressed or the vacuum degree is compensated through reinforcement learning. If the K value continues to be low, the energy consumption optimization mode is activated. S104: Optimize weight factor configuration rules using federated learning based on historical K-value datasets.

6. The intelligent production control method for freeze-dried granules based on the Internet of Things according to claim 5, characterized in that, The control logic of S103 is as follows: when the K value exceeds the upper limit of the threshold for 3 consecutive minutes, the cooling amplitude is linearly increased according to the excess ratio; when the K value is below the lower limit of the threshold for 10 minutes, the vacuum compensation program is started and the compensation amount is proportional to the reciprocal of the K value.

7. The method for intelligent production control of freeze-dried granules based on the Internet of Things according to claim 6, characterized in that: It also includes S105, which establishes a method for linking K value and quality, statistically analyzes the variance of K value fluctuation at the end point of batch freeze-drying, automatically determines batches as unqualified when the variance value is >0.05, and predicts the remaining vacuum service life based on the K value fluctuation frequency characteristics.

8. The intelligent production control method for freeze-dried granules based on the Internet of Things according to claim 5, characterized in that: During the pre-freezing stage, a K-value benchmark was established. The weighted average of crystallinity was calculated at -38℃ to -40℃ as the K0 benchmark. A real-time K-value deviation of 15-25% from the K0 benchmark was set as the process abnormality alarm threshold.

Images