System, method and device for detecting granulation end point of traditional Chinese medicine fluidized bed

Abstract

The invention belongs to the technical field of traditional Chinese medicine manufacturing, and particularly relates to a traditional Chinese medicine fluidized bed granulation end point detection system, method and device. The method comprises the following steps: receiving traditional Chinese medicine particle spectrum data collected at at least three preset sites; the spectral data are converted into predicted values of key quality attributes of the traditional Chinese medicine particles at all sites on the basis of a quantitative analysis model, the key quality attributes at least comprise the moisture content and the particle size, and the quantitative analysis model is a partial least square regression model; respectively calculating relative standard deviations of the key quality attributes of all sites; and if the key quality attribute predicted values of all the sites reach a preset target range and the relative standard deviation of each key quality attribute predicted value is lower than a set threshold value, generating a granulation end point signal. The invention provides an intelligent and accurate traditional Chinese medicine fluidized bed granulation end point detection scheme.

Description

Technical Field

[0001] This application belongs to the field of traditional Chinese medicine manufacturing technology, specifically relating to a detection system, method and device for the endpoint of fluidized bed granulation of traditional Chinese medicine. Background Technology

[0002] In the production of solid dosage forms of traditional Chinese medicine, accurate determination of the fluidized bed granulation endpoint is a core step in ensuring granule quality. Traditional experience-based judgment methods rely on operators visually observing through an observation window or manually sampling for offline testing, which has inherent drawbacks such as strong subjectivity, large time lag, and poor batch-to-batch quality consistency.

[0003] To overcome the limitations of traditional methods, near-infrared spectroscopy (NIR) process analysis technology has been introduced into the monitoring of the granulation process. This technology can acquire material spectral information in real time and non-destructively, and can simultaneously predict multiple key quality attributes of particles, such as moisture content and particle size, achieving a technological advancement from "unable to monitor online" to "able to monitor in real time".

[0004] However, existing NIR-based granulation endpoint monitoring technologies either only establish monitoring systems around a single key quality attribute (such as moisture content), failing to comprehensively reflect the complex state of granulation endpoint achievement involving multiple attributes; or, even if some technologies can monitor multiple indicators, they lack effective multi-dimensional information fusion algorithms, making it impossible to perform collaborative analysis and comprehensive judgment of data on key attributes such as moisture content and particle size. This lack of multi-dimensional judgment logic makes it difficult for existing technologies to accurately capture the essential characteristics of the granulation endpoint.

[0005] Furthermore, existing NIR-based granulation endpoint monitoring technologies either only use single-point or non-optimized monitoring locations, which cannot fully reflect the spatial distribution differences of materials within the fluidized bed; or, even if some technologies set up multiple monitoring points, they lack quantitative tools such as relative standard deviation (RSD) and scientific selection methods, making it impossible to establish an effective spatial homogeneity evaluation system.

[0006] The aforementioned lack of fundamental capabilities in multidimensional information fusion and spatial uniformity evaluation prevents existing technologies from providing automatic and reliable decision signals for endpoint determination. Consequently, the technological process is forced to halt at the critical "decision-making" stage, ultimately still requiring manual operation and failing to construct a complete automated closed loop. Summary of the Invention

[0007] This application proposes a detection scheme for the endpoint of fluidized bed granulation of traditional Chinese medicine to solve the problems of lack of multidimensional information fusion, lack of spatial uniformity evaluation, and lack of intelligent decision signal generation capability in existing schemes.

[0008] The first aspect of this application provides a detection system for the endpoint of fluidized bed granulation of traditional Chinese medicine, used in the preparation of traditional Chinese medicine granules, comprising:

[0009] Fluidized bed granulator;

[0010] The near-infrared spectroscopy acquisition system includes an online near-infrared spectrometer and at least three near-infrared spectral probes. The probes are communicatively connected to the online near-infrared spectrometer and are embedded in the chamber wall of the fluidized bed granulator. The selection of the mounting sites is based on the principle of maximizing the relative standard deviation, and the three sites are not all located on the same side of the central axis of the fluidized bed granulator.

[0011] A server, communicatively connected to the online near-infrared spectrometer, is configured with a quantitative analysis model and set as follows:

[0012] Acquire spectral data from the near-infrared spectral acquisition system, and calculate the predicted values ​​of key quality attributes of Chinese medicine granules at each point in real time based on the quantitative analysis model.

[0013] Calculate the relative standard deviation of the predicted values ​​of the key quality attributes for each site;

[0014] When the predicted values ​​of key quality attributes at all sites reach the preset target range, and the relative standard deviation of each predicted value of key quality attribute is lower than a set threshold, an endpoint control signal is generated, wherein the set threshold is 3%.

[0015] The controller is communicatively connected to the server and the fluidized bed granulator, and is used to receive the endpoint control signal and control the fluidized bed granulator to terminate the granulation process.

[0016] In some embodiments of this application, the mounting site of the near-infrared spectral probe is a combination of three points selected from multiple candidate sites in the vertical direction of the fluidized bed granulator chamber based on the following steps:

[0017] During the granulation process, near-infrared spectroscopy was used to obtain spectral data of the Chinese medicine particles at each candidate site, and the measured values ​​of key quality attributes were obtained accordingly.

[0018] For different combinations of three sites among all candidate sites, calculate the relative standard deviation of the measured values ​​of the corresponding key quality attributes.

[0019] From all the three-point combinations, the combination with the largest relative standard deviation is selected as the final monitoring site combination.

[0020] In some embodiments of this application, the three sites in the final monitoring site combination are not all located on the same side of the central axis of the fluidized bed granulator, and the three sites are at different heights, corresponding to the upper, middle and lower parts of the chamber, respectively.

[0021] In some embodiments of this application, the quantitative analysis model is a partial least squares regression model.

[0022] In some embodiments of this application, the partial least squares regression model is established by the following method:

[0023] Obtain samples with known key quality attribute reference values ​​covering different granulation stages, and acquire their full-band near-infrared spectra;

[0024] The full-band near-infrared spectrum was analyzed using a competitive adaptive resampling method to screen out characteristic wavelength variables for establishing a quantitative model;

[0025] Based on the spectral data at the characteristic wavelength variables and their corresponding key quality attribute reference values, the final quantitative prediction model is obtained by training with a partial least squares regression algorithm.

[0026] The process of establishing the partial least squares regression model also includes:

[0027] In some embodiments of this application, the full-band near-infrared spectral data is preprocessed using various methods, and the optimal preprocessing method is determined by comparing the prediction performance of the partial least squares regression model under different preprocessing methods. The preprocessing methods include standard canonical transformation, multivariate scattering correction, first derivative, second derivative, and SG smoothing filter.

[0028] In some embodiments of this application, the optimal preprocessing method is:

[0029] Standard canonical transformation is used for moisture content prediction;

[0030] The original spectrum was used for particle size prediction.

[0031] In some embodiments of this application, the key quality attributes include the moisture content and particle size of the Chinese medicine granules.

[0032] The second aspect of this application provides a method for detecting the endpoint of fluidized bed granulation of traditional Chinese medicine, applied to the system described in the first aspect of this application, including:

[0033] Acquire spectral data of traditional Chinese medicine granules collected at the sites described in the first aspect of the embodiments of this application;

[0034] The spectral data is converted into predicted values ​​of key quality attributes of Chinese medicine granules at each point based on a quantitative analysis model. The key quality attributes include at least moisture content and particle size. The quantitative analysis model is a partial least squares regression model.

[0035] Calculate the relative standard deviation of the key quality attributes for each site;

[0036] If the predicted values ​​of key quality attributes at all sites reach the preset target range, and the relative standard deviation of each predicted value of key quality attribute is lower than a set threshold, then a granulation endpoint signal is generated, wherein the set threshold is 3%.

[0037] A third aspect of this application provides a device for detecting the endpoint of fluidized bed granulation of traditional Chinese medicine, applied to the system described in claim 1, comprising:

[0038] The acquisition module is used to acquire the spectral data of traditional Chinese medicine granules collected from the sites described in the first aspect of the embodiments of this application;

[0039] The prediction module is used to convert the spectral data into predicted values ​​of key quality attributes of Chinese medicine granules at each point based on a quantitative analysis model. The key quality attributes include at least moisture content and particle size. The quantitative analysis model is a partial least squares regression model.

[0040] The calculation module is used to calculate the relative standard deviation of the key quality attributes for each site.

[0041] The endpoint determination module is used to generate a granulation endpoint signal if the predicted values ​​of key quality attributes at all sites reach the preset target range and the relative standard deviation of each predicted value of key quality attributes is lower than a set threshold, wherein the set threshold is 3%.

[0042] In summary, the detection system, method, and apparatus for the endpoint of fluidized bed granulation of traditional Chinese medicine provided in this application achieve parallel monitoring and fusion analysis of key quality attributes such as moisture content and particle size by constructing a multi-attribute synchronous analysis mechanism and leveraging the synergistic effect of a near-infrared spectroscopy real-time acquisition system and a PLSR quantitative prediction model. Furthermore, by establishing a multi-index collaborative judgment algorithm, the traditionally isolated quality attribute evaluation is transformed into a unified decision-making system, thereby accurately identifying the composite compliance status of the granulation endpoint and solving the problem of insufficient multi-dimensional information fusion in existing technologies. By deploying a spatially distributed monitoring network at the feature site with the largest RSD (Real-Size Dispersion), and using the RSD tool to quantitatively analyze the spatial dispersion of materials, a complete spatial uniformity evaluation system is constructed, solving the problem of insufficient spatial uniformity evaluation in existing technologies. Finally, by integrating an automatic decision-making algorithm and a signal generation module, and using dual criteria to perform real-time analysis of monitoring data, control signals are automatically triggered when conditions are met, thereby achieving intelligent judgment of the granulation endpoint and solving the problem of insufficient intelligent decision signal generation capability. Attached Figure Description

[0043] The features and advantages of this application will become clearer with reference to the accompanying drawings, which are illustrative and should not be construed as limiting the application in any way. In the drawings:

[0044] Figure 1 This is a schematic diagram of a fluidized bed granulation endpoint detection system for traditional Chinese medicine provided in an embodiment of this application;

[0045] Figure 2 This is a schematic diagram of the fluidized bed granulator in this application;

[0046] Figure 3 This is a flowchart illustrating a method for detecting the endpoint of fluidized bed granulation of traditional Chinese medicine, according to some embodiments of this application.

[0047] Figure 4 This is a flowchart illustrating the construction process of the quantitative analysis model in this application;

[0048] Figure 5 This is a schematic diagram of the workflow of the fluidized bed granulation endpoint determination method designed in this application;

[0049] Figure 6 This is a schematic diagram of a device for detecting the endpoint of fluidized bed granulation of traditional Chinese medicine, according to some embodiments of this application.

[0050] Figure label:

[0051] 1: Near-infrared online monitoring probe 1 2: Near-infrared online monitoring probe 2

[0052] 3: Near-infrared online monitoring probe; 4: Adhesive nozzle

[0053] 5: Filter bag 6: Control panel

[0054] 7: Hot air 8: Fluidized bed granulator

[0055] 9: Controller; 10: Online near-infrared spectrometer

[0056] 11: Server Detailed Implementation

[0057] In the following detailed description, numerous specific details of this application are illustrated by example to provide a thorough understanding of the relevant disclosure. However, it will be apparent to those skilled in the art that this application can be practiced without these details. It should be understood that the terms “system,” “apparatus,” “unit,” and / or “module” used in this application are one way of distinguishing different parts, elements, sections, or components at different levels in a sequential arrangement. However, these terms may be replaced with other expressions if other expressions can achieve the same purpose.

[0058] It should be understood that when a device, unit, or module is referred to as being "on," "connected to," or "coupled to" another device, unit, or module, it may be directly connected to or coupled to or communicate with other devices, units, or modules, or there may be intermediate devices, units, or modules present, unless the context explicitly indicates otherwise. For example, the term "and / or" as used herein includes any one and all combinations of one or more of the relevant listed items.

[0059] The terminology used in this application is for the purpose of describing specific embodiments only and is not intended to limit the scope of this application. As shown in the specification and claims of this application, unless the context clearly indicates otherwise, words such as "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate that explicitly identified features, integrals, steps, operations, elements, and / or components are included, and such expressions do not constitute an exclusive list, and other features, integrals, steps, operations, elements, and / or components may also be included.

[0060] Referring to the following description and accompanying drawings, these and other features and characteristics, operating methods, functions of related structural elements, combinations of parts, and economics of manufacture of this application can be better understood, wherein the description and drawings form part of the specification. However, it is clearly understood that the drawings are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. It is understood that the drawings are not drawn to scale.

[0061] Various structural diagrams are used in this application to illustrate various variations of the embodiments according to this application. It should be understood that the preceding or following structures are not intended to limit this application. The scope of protection of this application is determined by the claims.

[0062] The manufacturing process of solid dosage forms of traditional Chinese medicine (TCM) involves multiple unit operations, including extraction, concentration, drying, granulation, and tableting. Among these, granulation is a crucial step in ensuring the uniformity of granule quality and the performance of the formulation. Fluidized bed granulation technology, as a highly efficient granulation method, is widely used in TCM production. Its core principle lies in the fact that in a sealed container, hot air rising from the bottom suspends the powder material in a "fluidized state," while a binder solution is sprayed in, causing the powder to aggregate into porous, uniform granules through liquid bridging. Fluidized bed granulation integrates mixing, granulation, and drying into one unit, completing all processes within a single device. This not only improves production efficiency but also reduces contamination and loss caused by material transfer, making it particularly suitable for processing complex components of TCM. TCM raw materials typically have characteristics such as multiple components, hygroscopicity, and variable physicochemical properties, which places higher demands on the precise control of the granulation process. Endpoint determination has become a core challenge affecting batch consistency and product quality.

[0063] Despite the significant advantages of fluidized bed granulation technology, its endpoint determination has long relied on the operator's experience. Traditional methods mainly include two types: one is to visually observe the fluidization state and particle appearance through an observation window, and the other is to manually sample and perform offline testing (such as moisture and particle size analysis) after the machine is shut down. These methods are highly subjective and have a large time lag, failing to reflect the dynamic changes of the material in real time, and are prone to "under-granulation" or "over-granulation," affecting the uniformity and quality of the particles.

[0064] In recent years, near-infrared spectroscopy (NIR) technology has been introduced into the monitoring of granulation processes. The basic principle of NIR technology is based on the absorption and scattering characteristics of near-infrared light by the physicochemical properties of materials (such as the vibrations of hydrogen-containing OH groups). It can capture spectral information in real time and non-destructively, and simultaneously predict multiple key quality attributes, such as moisture content and particle size, through chemometric models. This ability to detect multiple attributes in parallel gives NIR a unique advantage in process analysis, achieving a leap from "unable to monitor online" to "able to monitor in real time."

[0065] However, existing NIR technologies still have significant limitations: First, most technologies focus only on a single indicator or fail to effectively integrate multiple indicators for comprehensive judgment, lacking a multi-attribute fusion decision-making mechanism and unable to fully capture the complex state of synergistic achievement of multiple indicators at the granulation endpoint; second, monitoring points are often limited to single sites or non-optimal locations, failing to assess the spatial uniformity of materials within the fluidized bed; finally, existing methods rely on manual intervention for endpoint determination, failing to achieve a closed-loop operation from data acquisition to automatic control. These problems result in insufficient intelligence and reliability in endpoint determination, making it difficult to meet the high standards required for traditional Chinese medicine manufacturing.

[0066] To address this, this application proposes a method and system for detecting the endpoint of fluidized bed granulation of traditional Chinese medicine, aiming to achieve intelligent judgment of the endpoint of the fluidized bed granulation process, thereby significantly improving production efficiency. This system innovatively transforms the endpoint judgment of granulation from a traditional "experience-dependent" model to a "data-driven" model by introducing spatial multi-point monitoring and dual intelligent criteria. This transformation not only makes the judgment results more accurate and reliable, fundamentally ensuring the uniformity and stability of granule quality, but also automates and intelligentizes the endpoint judgment process, providing a solid foundation for the intelligent upgrading of traditional Chinese medicine manufacturing.

[0067] Figure 1 This is a schematic diagram of a fluidized bed granulation endpoint detection system for traditional Chinese medicine provided in an embodiment of this application, used to achieve online monitoring and automatic control of the granulation process. Figure 1 As shown, the system mainly includes a fluidized bed granulator 8, a near-infrared spectroscopy acquisition system, a server 11, and a controller 9. Wherein:

[0068] The near-infrared spectroscopy acquisition system consists of an online near-infrared spectrometer 10 and at least three near-infrared spectral probes 1, 2, and 3. The near-infrared spectral probes 1, 2, and 3 are communicatively connected to the online near-infrared spectrometer 10 and are used to acquire material spectral information at different spatial locations within the fluidized bed in real time.

[0069] The server 11 is communicatively connected to the online near-infrared spectrometer 10 and is internally configured with a quantitative analysis model (such as a partial least squares regression model). The server 11 is configured to perform the following core functions: receive spectral data streams from the spectrometer 10; calculate in real-time predicted values ​​of key quality attributes (such as moisture content and particle size) of the Chinese medicine granules at each probe site based on the quantitative analysis model; then calculate the relative standard deviation (RSD) of the predicted values ​​of the key quality attributes at all sites; finally, the server 11 automatically generates a granulation endpoint signal only when the dual criteria are met—that is, the predicted values ​​of the key quality attributes at all sites reach a preset target range, and the relative standard deviation of each attribute is lower than a set threshold. This dual criterion ensures that the material not only meets the property standards but also has a highly uniform spatial distribution at the endpoint determination.

[0070] The controller 9 is communicatively connected to both the server 11 and the fluidized bed granulator 8. It receives the granulation endpoint signal from the server 11 and controls the fluidized bed granulator 8 to terminate the granulation process accordingly, thereby achieving closed-loop control from real-time monitoring and intelligent analysis to automatic execution.

[0071] In different embodiments of this application, the hardware deployment methods of the server 11 and the controller 9 can be flexibly selected. In some embodiments, they are functionally independent hardware units, and the server 11 (such as an industrial computer) and the controller 9 (such as a PLC) are connected through a communication interface. In other embodiments, the functions of the server 11 and the controller 9 can be integrated and deployed in the same industrial control computer.

[0072] Figure 2 This is a schematic diagram of the fluidized bed granulator 8 in this application. Fluidized bed granulators are commonly used equipment for solid dosage forms of traditional Chinese medicine; their chambers are typically cylindrical with a conical hopper at the bottom. (Example:...) Figure 2 As shown,

[0073] Hot air 7 is introduced from the bottom of the equipment to fluidize the material. Then, adhesive nozzles 4, installed on the side wall of the chamber, spray adhesive into the fluidized layer, achieving wet agglomeration and granulation of the material. During this process, near-infrared spectral probes 1, 2, and 3, embedded in the upper, middle, and lower parts of the chamber, collect near-infrared spectral information of the material at different spatial locations in real time, providing online monitoring of the granulation process. The dust-laden, humid, and hot air generated during granulation rises to the top of the chamber and is discharged after gas-solid separation by the filter bag 5. The entire process is controlled by a control panel 6 located outside the equipment, which allows for parameter setting and operational monitoring.

[0074] In the fluidized bed granulation process of traditional Chinese medicine, the uneven distribution of key quality attributes (such as moisture content and particle size) within the bed space is a core issue affecting batch-to-batch quality consistency. This application aims to accurately capture this unevenness by optimizing the spatial layout of near-infrared monitoring sites.

[0075] like Figure 1 and Figure 2 As shown, three near-infrared spectral probes 1, 2, and 3 are deployed in the upper, middle, and lower parts of chamber 8 of the fluidized bed granulator, respectively, and are not all located on the same side of the central axis. This combination of probes at these locations allows for optimal monitoring of material spatial inhomogeneity, thereby accurately determining the granulation endpoint.

[0076] In this application, the optimal placement location of the near-infrared spectroscopy acquisition probe is determined based on the principle of maximizing the relative standard deviation (RSD). This principle aims to capture the region with the most uneven spatial distribution of materials within the fluidized bed, thereby achieving accurate determination of the granulation endpoint. It should be noted that the optimal combination of placement locations satisfying the RSD maximization principle is not static; it depends on various factors such as the formulation composition and physical properties (e.g., density, particle size distribution) of the traditional Chinese medicine raw materials, as well as the parameters and operating conditions of the fluidized bed equipment (e.g., airflow, binder flow rate). The "upper, middle, and lower" combination is an optimal embodiment obtained under specific process conditions for "digestive and stomach-strengthening tablets" granules through the following method, used to illustrate the core principle of this application.

[0077] This method first identifies multiple candidate installation sites (e.g., lower 1, lower-middle 2, middle 3, upper-middle 4, and upper 5) in the vertical direction of the fluidized bed chamber. Then, during granulation, offline sampling and detection or online near-infrared spectroscopy monitoring are used to obtain measured values ​​of key quality attributes (such as moisture content and particle size) of the herbal granules at each candidate site. Next, for different combinations of any three sites among all candidate sites, the relative standard deviation (RSD) of the corresponding measured values ​​of the key quality attributes is calculated. Finally, by comparing the RSD values ​​of all three-site combinations, the combination with the largest RSD value is selected as the final monitoring site combination.

[0078] In one embodiment of this application, the moisture content and particle size data of the digestive tablet granules measured at various candidate sites are shown in Table 1:

[0079] Table 1: Key quality data of digestive tablet granules measured at candidate sites:

[0080] Sampling sites Moisture content (%) <![CDATA[Particle size D 50 (μm)]]> lower part(1) 3.61 72.45 middle and lower part(2) 3.37 66.28 Middle (3) 3.35 69.95 Upper middle (4) 3.32 60.11 Upper (5) 3.21 58.33

[0081] The key quality attribute data of all site combinations were analyzed, and the results are shown in Table 2. It can be seen that combinations 1, 3, and 5 (located in the upper, middle, and lower parts of the chamber, respectively, and not all spatially located on the same side of the central axis of the fluidized bed granulator) have the highest RSD values ​​for particle moisture content and particle size. This indicates that the area represented by this combination is the key control area with the most non-uniform quality during fluidized bed granulation. Therefore, this combination (sites 1, 3, and 5) was determined as the final monitoring site, and a near-infrared spectral acquisition probe was embedded in the corresponding chamber wall for online monitoring of the granulation process.

[0082] Table 2 compares the RSD data of the quality attributes of Jianwei Xiaoshi tablets with different site combinations:

[0083]

[0084] In the aforementioned embodiments, candidate sites are preset along the vertical direction of the chamber. It is understood that the initial determination method for candidate sites can be adjusted according to actual circumstances, without deviating from the core principle of maximizing RSD in this application. For example, candidate sites can be initially screened based on simulation calculations or historical data statistical analysis. However, the final determination of monitoring sites still requires evaluation and screening of candidate site combinations based on the principle of maximizing relative standard deviation. In this process, configuring sites so that they are "not all spatially located on the same side of the central axis" is a typical and effective means of achieving the goal of maximizing RSD.

[0085] The server 11 is internally configured with a quantitative analysis model trained based on the partial least squares regression (PLSR) algorithm. This model is used to process the spectral data stream from the near-infrared spectrometer 10 and to calculate in real time the predicted values ​​of key quality attributes (such as moisture content and particle size) of the Chinese medicine granules at positions 1, 2, and 3 of the near-infrared spectral probe.

[0086] The quantitative analysis model is established as follows: First, a training sample set covering different granulation stages is acquired, and the near-infrared spectra of the samples are collected, while reference values ​​of key quality attributes (such as moisture content and particle size) are determined. Next, a spectral preprocessing method (such as Standard Canonical Transform (SNV)) is used to preprocess the original spectra to remove noise and other influencing factors. Then, a feature wavelength selection algorithm (such as Competitive Adaptive Resampling (CARS)) is used to select feature variables across the entire spectrum to optimize model performance. Finally, based on the selected feature wavelength variables and their corresponding reference values ​​of key quality attributes, the final quantitative prediction model is trained using the PLSR algorithm. The model established through this process can achieve accurate and rapid prediction of key quality attributes.

[0087] To quantify model performance, this application uses the coefficient of determination (R²) and performance deviation ratio (RPD) as core evaluation indicators. R² is used to judge the goodness of fit of the model; the closer this indicator is to 1, the better the model fit. RPD is used to evaluate the model's predictive performance; when RPD < 1.4, it indicates that the constructed model is inaccurate, while when RPD > 2.0, it indicates that the model has high reliability and prediction accuracy and can be used for practical process control.

[0088] In addition, during the model building process (such as comparison of spectral preprocessing methods and optimization of feature wavelength extraction), this application also uses root mean square error (RMSE) to measure the deviation between predicted and measured values, thereby optimizing model parameters.

[0089] Figure 3 This is a flowchart illustrating a method for detecting the endpoint of fluidized bed granulation of traditional Chinese medicine according to some embodiments of this application. In some embodiments of this application, the method for detecting the endpoint of fluidized bed granulation of traditional Chinese medicine is... Figure 1 The server execution includes:

[0090] S310, obtain deployment on Figure 1 Or the spectral data of Chinese medicine granules collected by near-infrared spectral probes 1, 2, and 3 at the sites shown in Figure 2.

[0091] Specifically, spectral information from different spatial locations within the fluidized bed granulator 8 is collected in real time and synchronously. As previously described, the locations of probes 1, 2, and 3 are distributed in the upper, middle, and lower parts of the chamber, and are not all located on the same side of the central axis to ensure optimal monitoring of material spatial non-uniformity.

[0092] S320, the spectral data is converted into predicted values ​​of key quality attributes of Chinese medicine granules at each point based on a quantitative analysis model, wherein the key quality attributes include at least moisture content and particle size, and the quantitative analysis model is a partial least squares regression model.

[0093] Specifically, this step inputs the spectral data collected in real time in S301 into a pre-built quantitative analysis model to calculate and output the predicted values ​​of moisture content and particle size of the Chinese medicine granules at each point (probes 1, 2, and 3) in real time. These predicted values ​​provide the data basis for subsequent determination of the granulation endpoint.

[0094] In some embodiments of this application, the quantitative analysis model is a partial least squares regression (PLSR) model. The quantitative analysis model is configured to simultaneously calculate and output predicted values ​​of the key quality attributes of the Chinese medicine particles at each locus, based on near-infrared spectral data streams acquired in real-time and synchronously from at least three of the loci.

[0095] The construction process of the quantitative analysis model is as follows: Figure 4 Specifically, it includes:

[0096] Preparation of training sample set and determination of key quality attribute reference values:

[0097] A training sample set covering different granulation stages was obtained, and this sample set needed to be broadly representative. Taking digestive tablet granules as an example, a batch of samples covering different moisture contents and particle sizes was prepared, and reference values ​​of their key quality attributes were accurately measured to provide basic data for model building. The obtained data are shown in Table 3.

[0098] Table 3 shows particle sample data used to establish a quantitative prediction model for key particle quality attributes using near-infrared spectroscopy. (The data shown in the table uses Jianwei Xiaoshi tablet granules as an example.)

[0099]

[0100]

[0101]

[0102]

[0103]

[0104]

[0105]

[0106]

[0107]

[0108]

[0109] The partitioning of the dataset is crucial for ensuring the stability and accuracy of the prediction model. In one embodiment of this application, a total of 243 samples of digestive tablet granules were used to establish a quantitative prediction model of particle quality attributes using near-infrared spectroscopy. The above sample set was divided into a training set and a test set in a 3:1 ratio using the Kennard-Stone algorithm, whereby the training set contained 183 samples for model building and the test set contained 60 samples for model validation.

[0110] Near-infrared spectral acquisition and preprocessing:

[0111] Full-band near-infrared spectra were collected for the training sample set. The raw spectral data may contain noise from various factors, including spectral band overlap, baseline drift, and light scattering. If not properly processed, these noise sources can affect the accuracy and reliability of subsequent modeling. To reduce the errors caused by external factors, random noise, and baseline drift on the spectral data and improve the model's predictive performance, spectral preprocessing methods were employed to improve spectral quality, thereby enhancing the model's predictive ability. The collected particle near-infrared spectral data were preprocessed using five methods: Standard Normal Transform (SNV), Multivariate Scattering Correction (MSC), First Derivative (1st), Second Derivative (2nd), and SG Smoothing Filter (SG).

[0112] By comparing the performance of PLSR models obtained by different pretreatment methods (results are shown in Table 4), it was determined that the best pretreatment method for moisture content is SNV; for particle size, the original spectrum was used to achieve the best model results.

[0113] Table 4 Results of PLSR prediction models established under different preprocessing methods

[0114] (The data shown in the table uses digestive tablets granules as an example.)

[0115]

[0116] Feature wavelength screening:

[0117] Characteristic wavelengths are wavelengths in a spectral curve that possess special significance or specific information, reflecting the optical properties and other important parameters of a sample. Extracting these characteristic wavelengths reduces redundant information in the spectral data, thereby reducing the computational load on the model and improving computational efficiency and accuracy. Simultaneously, characteristic wavelengths help the model more accurately capture key information about the sample, enhancing its predictive power.

[0118] One embodiment of this application compares the performance of PLSR quantitative prediction models established based on various feature wavelength extraction methods. These methods include Random Frog (RF), Competitive Adaptive Resampling (CARS), and Uninformative Variable Elimination (UVE). Based on the selection of the optimal spectral preprocessing method, this embodiment extracts feature wavelengths from the near-infrared spectral dataset of particle samples and establishes the corresponding final PLSR quantitative prediction model. All modeling results are shown in Table 5 of the disclosure document. Table 5 shows that CARS is the optimal feature wavelength extraction method for both moisture content and particle size.

[0119] Table 5 Results of PLSR prediction models developed under different feature wavelength extraction methods

[0120] (The data shown in the table uses digestive tablets granules as an example.)

[0121]

[0122] The optimal PLSR model has been determined:

[0123] Based on the established optimal preprocessing method (i.e., using SNV for moisture content and the original spectrum for particle size), the predictive performance of various PLSR models obtained after variable screening using different characteristic wavelength extraction methods (RF, CARS, UVE) was systematically compared. The best-performing quantitative prediction model was then selected (its performance evaluation results are shown in Table 6). This application uses this model as a quantitative prediction model between near-infrared spectroscopy and key particle quality attributes to achieve real-time prediction of particle moisture content and particle size at three different sites during fluidized bed granulation.

[0124] Table 6: Prediction results of the optimal quantitative model between near-infrared spectroscopy and particle quality properties established using PLSR

[0125] (The data shown in the table uses digestive tablets granules as an example.)

[0126]

[0127] S330, calculate the relative standard deviation of the key quality attributes for each site.

[0128] S340, if the predicted values ​​of the key quality attributes of all sites reach the preset target range, and the relative standard deviation of each predicted value of the key quality attribute is lower than the set threshold, then a granulation endpoint signal is generated.

[0129] For details on the specific logic flow of endpoint determination and signal generation, please refer to [link / reference]. Figure 5 .like Figure 5 As shown, the following steps are performed in real time during the granulation process:

[0130] Calculate the relative standard deviation (RSD):

[0131] Based on the predicted values ​​of key quality attributes (moisture content, particle size) of Chinese herbal medicine particles obtained from three near-infrared spectral probes (1, 2, and 3) located in the upper, middle, and lower parts of the fluidized bed chamber, obtained from step S302, the relative standard deviation (RSD) of the predicted values ​​of each key quality attribute at all sites was calculated in real time. This RSD value is used to quantitatively characterize the uniformity of particle mass at different spatial sites within the fluidized bed at the same time.

[0132] Endpoint determination and signal generation:

[0133] The system uses dual endpoint determination criteria for simultaneous determination:

[0134] Attribute compliance judgment: The moisture content and particle size prediction values ​​of probes 1, 2, and 3 at each monitoring point must meet the preset process standard range.

[0135] Quality uniformity assessment: The RSD of the predicted values ​​at each point must be lower than the set threshold (usually set to 3%) to ensure uniformity within the batch.

[0136] Once both the "attribute compliance" and "quality uniformity" conditions are met, the system automatically determines that this is the granulation endpoint and immediately generates a granulation endpoint signal.

[0137] The threshold for the relative standard deviation (RSD) is set at 3%, which is based on a full consideration of the non-uniformity of the spatial distribution of materials in the fluidized bed. Preliminary experiments have verified that this threshold can effectively distinguish between the uniform and non-uniform states of particle quality during the granulation process, thereby ensuring the accuracy of the endpoint determination.

[0138] Signal output and process control:

[0139] The generated granulation endpoint signal is immediately sent to the controller 9, which then controls the fluidized bed granulator 8 to terminate the granulation process (such as stopping the binder spraying, turning off the hot air, etc.), thereby realizing a closed-loop control from real-time monitoring and intelligent analysis to automatic execution.

[0140] The fluidized bed granulation endpoint detection system and method constructed in this application are based on a universal "spectral acquisition-quantitative prediction-intelligent decision-making" technical strategy. This strategy is not targeted at a single variety, but rather achieves precise control over the fluidized bed granulation process of traditional Chinese medicine for different prescriptions by establishing a robust quantitative model between near-infrared spectroscopy and key quality attributes of traditional Chinese medicine granules. It is a universal intelligent process quality control solution.

[0141] To demonstrate the universality of this solution, another embodiment of this application, in addition to digestive tablet granules, specifically uses compound herbal lozenges granules, which have significantly different prescription compositions and physicochemical properties, as an example to showcase the establishment results of its near-infrared quantitative prediction model. As shown in Tables 7, 8, and 9, the results indicate that near-infrared spectroscopy also achieved ideal prediction results in the fluidized bed granulation process of compound herbal lozenges. This further proves that the technical solution of this application is a universal process quality control and endpoint intelligent judgment system applicable to the manufacturing of multi-type, multi-component traditional Chinese medicine granules.

[0142] Table 7 Results of PLSR prediction models established under different preprocessing methods

[0143] (The data shown in the table is based on Compound Herba Sarcandrae Lozenges Granules.)

[0144]

[0145] Table 8 Results of PLSR prediction models developed under different feature wavelength extraction methods

[0146] (The data shown in the table is based on Compound Herba Sarcandrae Lozenges Granules.)

[0147]

[0148] Table 9 Results of the optimal quantitative prediction model between near-infrared spectroscopy and particle quality properties established using PLSR

[0149] (The data shown in the table is based on Compound Herba Sarcandrae Lozenges Granules.)

[0150]

[0151] Figure 6 This is a schematic diagram of a device for detecting the endpoint of fluidized bed granulation of traditional Chinese medicine, according to some embodiments of this application. Figure 6 As shown, the device 600 for detecting the endpoint of fluidized bed granulation of traditional Chinese medicine includes an acquisition module 610, a prediction module 620, a calculation module 630, and an endpoint judgment module 640. In some embodiments of this application, the detection function of the endpoint of fluidized bed granulation of traditional Chinese medicine is provided by... Figure 1 The server executes the command. Among them:

[0152] The acquisition module is used to acquire information deployed on... Figure 1 Or the spectral data of Chinese herbal medicine particles collected by near-infrared spectral probes 1, 2, and 3 at the location shown in Figure 2;

[0153] The prediction module is used to convert the spectral data into predicted values ​​of key quality attributes of Chinese medicine granules at each point based on a quantitative analysis model. The key quality attributes include at least moisture content and particle size. The quantitative analysis model is a partial least squares regression model.

[0154] The calculation module is used to calculate the relative standard deviation of the key quality attributes for each site.

[0155] The endpoint determination module is used to generate a granulation endpoint signal if the predicted values ​​of key quality attributes at all sites reach the preset target range and the relative standard deviation of each predicted value of key quality attributes is lower than a set threshold, wherein the set threshold is 3%.

[0156] In summary, the detection system, method, and apparatus for the endpoint of fluidized bed granulation of traditional Chinese medicine provided in this application achieve parallel monitoring and fusion analysis of key quality attributes such as moisture content and particle size by constructing a multi-attribute synchronous analysis mechanism and leveraging the synergistic effect of a near-infrared spectroscopy real-time acquisition system and a PLSR quantitative prediction model. Furthermore, by establishing a multi-index collaborative judgment algorithm, the traditionally isolated quality attribute evaluation is transformed into a unified decision-making system, thereby accurately identifying the composite compliance status of the granulation endpoint and solving the problem of insufficient multi-dimensional information fusion in existing technologies. By deploying a spatially distributed monitoring network at the feature site with the largest RSD (Real-Size Dispersion), and using the RSD tool to quantitatively analyze the spatial dispersion of materials, a complete spatial uniformity evaluation system is constructed, solving the problem of insufficient spatial uniformity evaluation in existing technologies. Finally, by integrating an automatic decision-making algorithm and a signal generation module, and using dual criteria to perform real-time analysis of monitoring data, control signals are automatically triggered when conditions are met, thereby achieving intelligent judgment of the granulation endpoint and solving the problem of insufficient intelligent decision signal generation capability.

[0157] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the devices and modules described above can be referred to the corresponding descriptions in the foregoing device embodiments, and will not be repeated here.

[0158] Although the subject matter described herein is provided in the general context of execution on a computer system in conjunction with an operating system and applications, those skilled in the art will recognize that other implementations can also be executed in conjunction with other types of program modules. Generally, program modules include routines, programs, components, data structures, and other types of structures that perform specific tasks or implement specific abstract data types. Those skilled in the art will understand that the subject matter described herein can be practiced using other computer system configurations, including handheld devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, minicomputers, mainframes, etc., and can also be used in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules may reside on both local and remote memory storage devices.

[0159] Those skilled in the art will recognize that the units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0160] It should be understood that the specific embodiments described above are merely illustrative or explanatory of the principles of this application and do not constitute a limitation thereof. Therefore, any modifications, equivalent substitutions, improvements, etc., made without departing from the spirit and scope of this application should be included within the protection scope of this application. Furthermore, the appended claims are intended to cover all variations and modifications falling within the scope and boundaries of the appended claims, or equivalent forms of such scope and boundaries.

Claims

1. A detection system for the endpoint of fluidized bed granulation of traditional Chinese medicine, used in the preparation of traditional Chinese medicine granules, characterized in that, include: Fluidized bed granulator; The near-infrared spectroscopy acquisition system includes an online near-infrared spectrometer and at least three near-infrared spectral probes. The probes are communicatively connected to the online near-infrared spectrometer and are embedded in the chamber wall of the fluidized bed granulator. The selection of the mounting sites is based on the principle of maximizing the relative standard deviation, and the three sites are not all located on the same side of the central axis of the fluidized bed granulator. A server, communicatively connected to the online near-infrared spectrometer, is configured with a quantitative analysis model and set as follows: Acquire spectral data from the near-infrared spectral acquisition system, and calculate the predicted values ​​of key quality attributes of Chinese medicine granules at each point in real time based on the quantitative analysis model. Calculate the relative standard deviation of the predicted values ​​of the key quality attributes for each site; When the predicted values ​​of key quality attributes at all sites reach the preset target range, and the relative standard deviation of each predicted value of key quality attribute is lower than a set threshold, an endpoint control signal is generated, wherein the set threshold is 3%. The controller is communicatively connected to the server and the fluidized bed granulator, and is used to receive the endpoint control signal and control the fluidized bed granulator to terminate the granulation process.

2. The system according to claim 1, characterized in that, The mounting site for the near-infrared spectroscopy probe is a combination of three points selected from multiple candidate sites in the vertical direction of the fluidized bed granulator chamber based on the following steps: During the granulation process, near-infrared spectroscopy was used to obtain spectral data of the Chinese medicine particles at each candidate site, and the measured values ​​of key quality attributes were obtained accordingly. For different combinations of three sites among all candidate sites, calculate the relative standard deviation of the measured values ​​of the corresponding key quality attributes. From all the three-point combinations, the combination with the largest relative standard deviation is selected as the final monitoring site combination.

3. The system according to claim 2, characterized in that: The three monitoring sites in the final monitoring site combination are not all located on the same side of the central axis of the fluidized bed granulator, and they are at different heights, corresponding to the upper, middle and lower parts of the chamber, respectively.

4. The system according to claim 1, characterized in that: The quantitative analysis model is a partial least squares regression model.

5. The system according to claim 4, characterized in that, The partial least squares regression model is established using the following method: Obtain samples with known key quality attribute reference values ​​covering different granulation stages, and acquire their full-band near-infrared spectra; The full-band near-infrared spectrum was analyzed using a competitive adaptive resampling method to screen out characteristic wavelength variables for establishing a quantitative model; Based on the spectral data at the characteristic wavelength variables and their corresponding key quality attribute reference values, the final quantitative prediction model is obtained by training with a partial least squares regression algorithm.

6. The system according to claim 5, characterized in that, The process of establishing the partial least squares regression model also includes: The full-band near-infrared spectral data is preprocessed using various methods, and the optimal preprocessing method is determined by comparing the prediction performance of the partial least squares regression model under different preprocessing methods. The preprocessing methods include standard canonical transformation, multivariate scattering correction, first derivative, second derivative, and SG smoothing filter.

7. The system according to claim 6, characterized in that, The optimal preprocessing method is: Standard canonical transformation is used for moisture content prediction; Raw spectral data were used for particle size prediction.

8. The system according to claim 1, 2, or 5, characterized in that: The key quality attributes include the moisture content and particle size of the Chinese medicine granules.

9. A method for detecting the endpoint of fluidized bed granulation of traditional Chinese medicine, applied to the system described in claim 1, characterized in that, include: Obtain the spectral data of traditional Chinese medicine granules collected at the site described in claim 1; The spectral data is converted into predicted values ​​of key quality attributes of Chinese medicine granules at each point based on a quantitative analysis model. The key quality attributes include at least moisture content and particle size. The quantitative analysis model is a partial least squares regression model. Calculate the relative standard deviation of the key quality attributes for each site; If the predicted values ​​of key quality attributes at all sites reach the preset target range, and the relative standard deviation of each predicted value of key quality attribute is lower than a set threshold, then a granulation endpoint signal is generated, wherein the set threshold is 3%.

10. A device for detecting the endpoint of fluidized bed granulation of traditional Chinese medicine, applied to the system described in claim 1, characterized in that, include: The acquisition module is used to acquire the spectral data of Chinese herbal medicine particles collected at the site according to claim 1; The prediction module is used to convert the spectral data into predicted values ​​of key quality attributes of Chinese medicine granules at each point based on a quantitative analysis model. The key quality attributes include at least moisture content and particle size. The quantitative analysis model is a partial least squares regression model. The calculation module is used to calculate the relative standard deviation of the key quality attributes for each site. The endpoint determination module is used to generate a granulation endpoint signal if the predicted values ​​of key quality attributes at all sites reach the preset target range and the relative standard deviation of each predicted value of key quality attributes is lower than a set threshold, wherein the set threshold is 3%.

Images