Method and device for online monitoring and improving drying uniformity of traditional Chinese medicine granule fluidized bed

Abstract

The invention provides a traditional Chinese medicine granule fluidized bed drying uniformity online monitoring and improving method and device, relates to the field of traditional Chinese medicine preparation production and online quality control, and solves the technical problem of uniformity judgment deviation caused by single-bed data isolated monitoring. The method comprises the following steps: inputting secondary bed outlet particle spectral data into a spectral model to obtain a secondary bed predicted water content M2, and integrating temperature data of each bed, process parameters and the secondary bed predicted water content M2 to construct a database. And an input vector set is generated after preprocessing. And inputting the input vector set into a principal component analysis model, performing PCA dimension reduction on the input vector set, extracting main components, and inputting the main components into a Hotelling T-party monitoring model. And calculating a control limit based on multiple batches of historical qualified data, comparing the T-party statistical magnitude with the control limit to judge a drying uniformity grade, and outputting a corresponding regulation and control strategy according to the drying uniformity grade. The method is used in the production and on-line quality control process of the traditional Chinese medicine preparation.

Description

Technical Field

[0001] This application relates to the field of traditional Chinese medicine preparation production and online quality control, and in particular to a method and apparatus for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules. Background Technology

[0002] Online monitoring of fluidized bed drying uniformity involves real-time, dynamic monitoring of key parameters in the fluidized bed drying process of traditional Chinese medicine granules to ensure consistent and stable product quality. Existing online monitoring methods for fluidized bed drying uniformity of traditional Chinese medicine granules mostly rely on single-bed temperature sensors (reflecting only local temperature) and single-point sampling for rapid moisture content detection (measuring only the moisture content of the granules at the sampling port). Therefore, in multi-bed drying processes, insufficient preheating in one bed cannot be detected by downstream bed monitoring in a timely manner, leading to delayed alarms when overload occurs in another bed, severely impacting production continuity and product quality. Furthermore, relying solely on single-point temperature measurement cannot comprehensively reflect the temperature distribution and material state of the entire bed; while rapid moisture detection can obtain some moisture information, it cannot capture the complex synergistic effects between multiple beds, making isolated monitoring of single-bed data limited and prone to misjudgment when assessing product uniformity, thus hindering accurate quality control. Summary of the Invention

[0003] This application provides a method and apparatus for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules, which solves the technical problem of uniformity judgment deviation caused by isolated monitoring of single-bed data in the prior art.

[0004] To achieve the above objectives, this application adopts the following technical solution: Firstly, a method for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules includes: acquiring temperature data, process parameters, and spectral data of particles exiting the second bed; inputting the spectral data of particles exiting the second bed into a spectral model to obtain the predicted moisture content M2 of the second bed; and integrating the temperature data, process parameters, and predicted moisture content M2 of the second bed to construct a database; reading the temperature, process parameters, and predicted moisture content M2 of the second bed from the database in real time, and generating an input vector set containing 10 variables after preprocessing; inputting the input vector set into a principal component analysis model to perform PCA dimensionality reduction and extract the principal components, and inputting the principal components into a Hotelling T-squared monitoring model; constructing a T-squared statistic based on the principal components, calculating control limits based on multiple batches of historical qualified data, comparing the T-squared statistic with the control limits to determine the level of drying uniformity, and outputting corresponding control strategies according to the level of drying uniformity.

[0005] Based on the above technical solution, in the online monitoring and improvement method for the uniformity of fluidized bed drying of traditional Chinese medicine granules provided in this application, the temperatures (T1, T2, T3) and process parameters (Q1-V3) of the three beds are aligned with the spectral data of the two beds in real time through timestamp synchronization technology. For the first time, a 10-variable data entity covering the entire process of preheating, main drying and shaping is constructed. This solves the problems of blind spots in cross-bed abnormal transmission caused by traditional single-bed monitoring (such as the overload lag alarm of the two beds caused by insufficient preheating of the first bed) and isolated monitoring of single-bed data (such as using only single-point temperature or moisture rapid detection) which cannot capture the synergistic effect of the three beds, resulting in misjudgment of uniformity. At the same time, the spectral model of the two beds outputs the M2 value every 5 seconds (replacing the traditional 30-minute offline detection) and binds it with the process parameters in real time to form a closed-loop link of process parameters (T, Q, V), real-time quality (M2) and control feedback. This allows for rapid tracing of changes in different parameters at the same time point when an output fault occurs.

[0006] In conjunction with the first aspect mentioned above, one possible implementation involves constructing a database by integrating temperature data from each bed, process parameters, and the predicted moisture content M2 of the second bed. This includes updating the temperature data and process parameters of each bed every 5 seconds. The temperature data for each bed includes preheating temperature data T1 for the first bed's preheating stage, main drying temperature data T2 for the second bed's main drying stage, and shaping temperature data T3 for the third bed's shaping stage. The process parameters include air inlet volume Q1 and air inlet velocity V1 for the first bed, air inlet volume Q2 and air inlet velocity V2 for the second bed, and air inlet volume Q3 and air inlet velocity V3 for the third bed. Timestamp synchronization technology is used to time-align the temperature data, process parameters, particle spectral data at the second bed outlet, and the predicted moisture content M2. A database was constructed by integrating preheating temperature data T1, main drying temperature data T2, shaping temperature data T3, air inlet velocity of bed 1 V1, air inlet velocity of bed 2 V2, air inlet velocity of bed 3 V3, air inlet volume of bed 1 Q1, air inlet volume of bed 2 Q2, air inlet volume of bed 3 Q3, and predicted moisture content of bed 2 M2.

[0007] In conjunction with the first aspect mentioned above, in one possible implementation, an input vector set containing 10 variables is generated after preprocessing. This includes: calculating anomaly thresholds based on multiple batches of historical qualified data; removing data exceeding the anomaly threshold from the following data: preheating temperature data T1, main drying temperature data T2, shaping temperature data T3, air inlet velocity V1 for bed 1, air inlet velocity V2 for bed 2, air inlet velocity V3 for bed 3, air inlet volume Q1 for bed 1, air inlet volume Q2 for bed 2, and air inlet volume Q3 for bed 3; smoothing the predicted moisture content M2 for bed 2 using a moving average method with a window size of 5 sets of continuous data; and integrating the processed preheating temperature data T1, main drying temperature data T2, shaping temperature data T3, air inlet velocity V1 for bed 1, air inlet velocity V2 for bed 2, air inlet velocity V3 for bed 1, air inlet volume Q1 for bed 1, air inlet volume Q2 for bed 2, air inlet volume Q3 for bed 3, and predicted moisture content M2 for bed 2 to obtain the input vector set containing 10 variables.

[0008] In conjunction with the first aspect mentioned above, one possible implementation involves inputting the input vector set into a principal component analysis (PCA) model to perform PCA dimensionality reduction and extract the principal components, including: using the covariance matrix formula... Calculate each set of input vectors The covariance matrix C is given by n, where n is the total number of input vector sets. Let X be the mean vector of each set of input vectors. i Let C be the i-th set of input vectors. Perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalues ​​λ1>λ2>…>λ. 10 >0 and the corresponding eigenvectors V1, V 1， V 10 .pass Calculate the cumulative contribution rate LU of the first K principal components, and select the smallest k value that satisfies LU ≥ 90%, where k = 3. The first three main components, PC1, PC2, and PC3, were extracted. PC1 represents the overall drying intensity, and the variables with the largest loading weights are T2, Q2, and M2. PC2 represents inter-bed consistency, and the variables with the largest loading weights are V1, V2, V3, Q1, Q2, and Q3. PC3 represents the final product quality trend, and its loading weight reflects the correlation between M2 and T3 and Q3.

[0009] In conjunction with the first aspect mentioned above, in one possible implementation, the Hotelling T-squared monitoring model constructs a T-squared statistic based on the principal components of the input and calculates control limits based on multiple batches of historical qualified data. The drying uniformity level is determined by comparing the T-squared statistic with the control limits, including: based on the first three principal components PC1, PC2, and PC3 and their corresponding eigenvalues ​​λ1, λ2, and λ3, through... Calculate the real-time T-squared statistic T 2 Based on multiple batches of historical qualified data, through Calculate the T-squared statistic T 2 upper limit of control , of which F α (3,m 3) is the critical value of the F-distribution at a significance level of α=0.05, and m is the number of historical qualified data samples. T 2 With control upper limit By comparing multiple dimensions, the drying uniformity level is output.

[0010] In conjunction with the first aspect mentioned above, in one possible implementation, T is... 2 With control upper limit The drying uniformity level is output by comparing multiple dimensions, including: when T 2 ≤ At that time, the drying uniformity grade is judged to be qualified. <T 2 ≤1.2× If the contribution rate deviation of any one of the principal components PC1, PC2, and PC3 is >5%, it is judged as a mild anomaly. When 1.2× <T 2 ≤1.5× If the contribution rate deviation of any two of the principal components PC1, PC2, and PC3 is >5%, it is judged as moderately abnormal. 2 >1.5× At that time, it was determined to be a severe abnormality.

[0011] In conjunction with the first aspect mentioned above, one possible implementation involves outputting a corresponding control strategy based on the level of drying uniformity. This includes: identifying anomalous principal component types PC1, PC2, or PC3 based on a principal component analysis model, and comparing these with historical anomalous cases in the industrial database to determine the anomalous principal component types corresponding to different levels. When a mild anomaly is identified, the associated anomalous principal component PC3 is invoked, along with the optimal parameter ratio from the historical database. The inlet air velocity V1 of the first bed is adjusted within ±0.1 m / s, and the temperature T1 of the first bed is adjusted within ±0.5℃ to reduce the temperature difference between beds to within the acceptable range. When a moderate anomaly is identified, the associated anomalous principal component PC2 is invoked, and the inlet air volume Q2 of the second bed is adjusted within ±5 m³ / h. The guide vanes of the second bed are opened and their angles adjusted within ±3°. Based on the correlation between the predicted moisture content M2 of the second bed and the temperature of the second bed, the temperature T2 of the second bed is adjusted within ±1℃. When a severe anomaly is detected, the associated principal component of the anomaly, PC1, is invoked. The airflow rate Q2 in the secondary bed is adjusted within ±5 m³ / h, and the secondary bed baffles are opened and their angle adjusted within ±3°. Simultaneously, based on the correlation between M2 and the secondary bed temperature T2 in the industrial database, T2 is adjusted within ±1°C. During the control process, real-time monitoring data is retrieved from the database every 5 seconds to evaluate the control effect until the T² statistic returns to the acceptable range and the moisture deviation meets the standard.

[0012] In conjunction with the first aspect mentioned above, one possible implementation involves using a principal component analysis model to identify abnormal principal component types PC1, PC2, or PC3. By comparing these abnormal cases with historical anomaly cases in the industrial database, the corresponding abnormal principal component types for different levels are determined. These include: if an abnormal principal component PC1 is identified, it is determined to be a deviation in the second-bed main drying parameters or material state. If an abnormal principal component PC2 is identified, it is determined to be a matching problem between the third-bed shaping stage and the second-bed main drying stage. If an abnormal principal component PC3 is identified, it is determined to be a poor synergy between the first and second beds.

[0013] In conjunction with the first aspect mentioned above, one possible implementation, after outputting the corresponding control strategy based on the level of drying uniformity, further includes: based on historical batch data accumulated in the industrial database, after completing N production batches, extracting the historical qualified data of those N batches, and recalculating the control upper limit of the Hotelling T-squared statistic. Where N is the set period, and the updated control limit adaptation error is ≤1.5%. Based on the updated control upper limit... The threshold for determining the drying uniformity level is dynamically adjusted. For different Chinese herbal medicine granule formulations, the historical optimal process parameter ratios are matched from the database to generate a baseline input vector set for that formulation.

[0014] Secondly, a device for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules is provided, comprising: a communication unit and a processing unit; the communication unit is used to acquire temperature data, process parameters, and spectral data of particles exiting the second bed; the processing unit is used to input the spectral data of particles exiting the second bed into a spectral model, output the predicted moisture content M2 of the second bed, and integrate the temperature data, process parameters, and predicted moisture content M2 of the second bed to construct a database. The temperature, process parameters, and predicted moisture content M2 of the second bed are read from the database in real time, and after preprocessing, an input vector set containing 10 variables is generated. The input vector set is input into a principal component analysis model to perform PCA dimensionality reduction and extract the principal components, and the principal components are input into a Hotelling T-squared monitoring model. The Hotelling T-squared monitoring model constructs a T-squared statistic based on the principal components, calculates control limits based on multiple batches of historical qualified data, compares the T-squared statistic with the control limits to determine the drying uniformity level, and outputs corresponding control strategies according to the drying uniformity level.

[0015] Thirdly, this application provides an online monitoring and improvement device for the uniformity of fluidized bed drying of traditional Chinese medicine granules, comprising: a processor and a storage medium; the storage medium includes instructions, and the processor is used to execute the instructions to implement the method described in the first aspect and any possible implementation thereof. This online monitoring and improvement device for the uniformity of fluidized bed drying of traditional Chinese medicine granules can be an electronic device or a chip within an electronic device.

[0016] Fourthly, this application provides a computer-readable storage medium storing instructions that, when executed on the online monitoring and improvement device for the uniformity of fluidized bed drying of traditional Chinese medicine granules, cause the device to perform the method described in the first aspect and any possible implementation thereof.

[0017] Fifthly, this application provides a computer program product containing instructions that, when the computer program product is run on the online monitoring and improvement device for the uniformity of fluidized bed drying of traditional Chinese medicine granules, causes the online monitoring and improvement device for the uniformity of fluidized bed drying of traditional Chinese medicine granules to perform the method as described in the first aspect and any possible implementation thereof.

[0018] This application provides a method and device for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules. It can synchronize the temperatures (T1, T2, T3) and process parameters (Q1-V3) of the three beds with the spectral data of the two beds in real time through timestamp synchronization technology. For the first time, it constructs a 10-variable data entity covering the entire process of preheating, main drying and shaping. It solves the problems of blind spots in cross-bed abnormality transmission caused by traditional single-bed monitoring (such as the overload lag alarm of the second bed caused by insufficient preheating of the first bed) and isolated monitoring of single-bed data (such as using only single-point temperature or moisture rapid detection) which cannot capture the synergistic effect of the three beds, resulting in misjudgment of uniformity. At the same time, the spectral model of the two beds outputs the M2 value every 5 seconds (replacing the traditional 30-minute offline detection) and binds it with the process parameters in real time to form a closed-loop link of process parameters (T / Q / V), real-time quality (M2) and control feedback. This allows for rapid tracing of changes in different parameters at the same time point when an output fault occurs.

[0019] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0020] Figure 1 A system architecture diagram of an online monitoring and improvement system for the uniformity of fluidized bed drying of traditional Chinese medicine granules provided in this application embodiment; Figure 2 A schematic flowchart illustrating an online monitoring and improvement method for the uniformity of fluidized bed drying of traditional Chinese medicine granules provided in this application embodiment; Figure 3 A schematic flowchart of another method for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules provided in this application embodiment; Figure 4 A schematic flowchart of another method for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules provided in this application embodiment; Figure 5 A schematic flowchart of another method for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules provided in this application embodiment; Figure 6 This is a schematic diagram of the structure of an online monitoring and improvement device for the uniformity of fluidized bed drying of traditional Chinese medicine granules provided in this application embodiment. Detailed Implementation

[0021] In the description of this application, unless otherwise stated, "" means "or," for example, A / B can mean A or B. "And / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The words "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0022] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0023] To address the shortcomings of current technologies in monitoring moisture content, which rely heavily on single-bed temperature sensors and single-point sampling for rapid detection, current methods are insufficient. Single-bed temperature sensors only reflect localized temperatures, and single-point sampling only measures particle moisture at the sampling port. They fail to consider the interrelationships between the three drying processes: bed preheating, bed main drying, and bed shaping. For example, insufficient bed preheating can lead to excessive drying load on bed two, and fluctuations in bed two parameters directly affect the final product quality of bed three. Data from a single bed is insufficient to characterize overall drying uniformity. Furthermore, process parameters (such as airflow and velocity in each bed) are not systematically integrated with spectral data, temperature data, and predicted moisture content, making it impossible to establish a comprehensive "input, process, and output" correlation model. This results in delayed early warnings of abnormal drying uniformity. At the control level, due to the lack of three-bed synergistic analysis data, when abnormal drying uniformity occurs, a "uniform adjustment" approach is often adopted, such as overall temperature increase or rate decrease, ignoring the functional differences of each bed. Bed 1 is mainly responsible for preheating, bed 2 for main drying, and bed 3 for shaping. This lack of targeted control method has poor improvement effect. This application provides an online monitoring and improvement method for the drying uniformity of fluidized bed in traditional Chinese medicine granules. This method uses timestamp synchronization technology to real-time align the temperatures (T1, T2, T3) and process parameters (Q1-V3) of the three beds with the spectral data of bed 2, and for the first time constructs a comprehensive... The 10-variable data entity covering the entire process of preheating, main drying, and shaping solves the problems of blind spots in cross-bed anomaly transmission caused by traditional single-bed monitoring (such as overload lag alarm in the second bed caused by insufficient preheating of the first bed) and the inability of isolated single-bed data monitoring (such as using only single-point temperature or moisture rapid detection) to capture the synergistic effect of the three beds, leading to misjudgment of uniformity. At the same time, the second-bed spectral model outputs the M2 value every 5 seconds (replacing the traditional 30-minute offline detection), which is bound to the process parameters in real time to form a closed-loop link of process parameters (T / Q / V), real-time quality (M2), and control feedback. This allows for rapid tracing of changes in different parameters at the same time point when an output fault occurs.

[0024] like Figure 1 As shown in the embodiments of this application, a method for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules includes: Step 101: Obtain temperature data, process parameters, and particle spectral data at the outlet of the second bed for each bed.

[0025] The temperature data for each bed refers to the temperature data collected by temperature sensors in beds one, two, and three during the horizontal fluidized bed drying process of traditional Chinese medicine granules, at a frequency of once per 5 seconds, reflecting the bed temperatures during the preheating, main drying, and shaping stages, respectively. Process parameters refer to the key process parameters for each bed collected by flow meters or anemometers with an accuracy of ±2% and a sampling frequency of once per 5 seconds. The particle spectral data at the outlet of bed two refers to the data collected using a near-infrared diffuse reflectance probe (wavelength 900-1700nm, resolution 8cm²) at the sampling port of bed two. -¹), collect the particle spectrum at the outlet of the second bed (collected when the particles flow through the sampling port, sampling frequency 1 time / 5s).

[0026] Step 102: Input the particle spectral data of the second bed outlet into the spectral model to obtain the predicted moisture content M2 of the second bed, and integrate the temperature data of each bed, process parameters and the predicted moisture content M2 of the second bed to construct a database.

[0027] A database will be constructed by integrating temperature data from each bed, process parameters, and the predicted water content M2 of the second bed, including: The temperature data and process parameters of each bed are updated every 5 seconds. The temperature data of each bed includes the preheating temperature data T1 of the first bed preheating stage, the main drying temperature data T2 of the second bed main drying stage, and the shaping temperature data T3 of the third bed shaping stage. The process parameters include the air inlet volume Q1 and air inlet velocity V1 of the first bed, the air inlet volume Q2 and air inlet velocity V2 of the second bed, and the air inlet volume Q3 and air inlet velocity V3 of the third bed.

[0028] The time stamp synchronization technology was used to align the temperature data, process parameters, particle spectral data at the outlet of the second bed, and the predicted moisture content M2 of each bed with time.

[0029] A database was constructed by integrating preheating temperature data T1, main drying temperature data T2, shaping temperature data T3, air inlet velocity of bed 1 V1, air inlet velocity of bed 2 V2, air inlet velocity of bed 3 V3, air inlet volume of bed 1 Q1, air inlet volume of bed 2 Q2, air inlet volume of bed 3 Q3, and predicted moisture content of bed 2 M2.

[0030] Among them, the spectral model refers to the prediction model established based on the correlation between near-infrared spectroscopy and moisture content. The predicted moisture content M2 of the second bed refers to the percentage value of the moisture content of the particles in the second bed calculated in real time through the spectral model. The database refers to a structured data storage system that integrates temperature, process parameters, and predicted moisture content.

[0031] In some implementations, the real-time collected spectral data of particles at the outlet of the second bed (preprocessed by SG smoothing and multivariate scattering correction) is first input into a pre-trained spectral model, and the model outputs the predicted moisture content M2 of the particles in the second bed. Then, the temperature data (T1, T2, T3) of each bed and the process parameters (Q1, V1, Q2, V2, Q3, V3) updated every 5 seconds are synchronized with the predicted moisture content M2 by timestamp synchronization. Next, M2 is smoothed by a moving average method with a window size of 5 sets of continuous data to eliminate random fluctuations. Finally, the 10 variables, including the preheating temperature T1, the main drying temperature T2, the shaping temperature T3, the inlet air velocity of the first bed V1, the inlet air velocity of the second bed V2, the inlet air velocity of the third bed V3, the inlet air volume of the first bed Q1, the inlet air volume of the second bed Q2, the inlet air volume of the third bed Q3, and the smoothed M2, are integrated and stored in an industrial database to form a standardized input vector set.

[0032] It should be noted that the spectral model needs to be triggered to collect data when the particles flow through the sampling port of the second bed to ensure data validity.

[0033] Step 103: Read the temperature of each bed, process parameters and the predicted water content M2 of the second bed from the database in real time, and generate an input vector set containing 10 variables after preprocessing.

[0034] After preprocessing, an input vector set containing 10 variables is generated, including: Based on multiple batches of historical qualified data, anomaly thresholds were calculated, and data exceeding the anomaly thresholds were removed from the following data: preheating temperature T1, main drying temperature T2, shaping temperature T3, air inlet velocity of bed 1 V1, air inlet velocity of bed 2 V2, air inlet velocity of bed 3 V3, air inlet volume of bed 1 Q1, air inlet volume of bed 2 Q2, and air inlet volume of bed 3 Q3.

[0035] The predicted water content M2 of the second bed was smoothed using a moving average method with a window size of 5 sets of continuous data.

[0036] The preheating temperature data T1, main drying temperature data T2, shaping temperature data T3, first bed air inlet velocity V1, second bed air inlet velocity V2, third bed air inlet velocity V3, first bed air volume Q1, second bed air volume Q2, third bed air volume Q3, and second bed predicted moisture content M2 are integrated to obtain an input vector set containing 10 variables.

[0037] The input vector set refers to a standardized dataset containing 10 variables.

[0038] In some implementations, the temperature of the three beds, the process parameters of the three beds, and the predicted moisture content M2 of the second bed are acquired in real time. Timestamp synchronization technology (error <5ms) is used to align all variables along a unified time axis, ensuring that data at the same point in time represents the same production state. This allows for the automatic removal of sudden outliers exceeding ±3σ (standard deviation) based on dynamic thresholds calculated from multiple batches of historical qualified data (e.g., the qualified range for T2 is 70-75℃). Furthermore, for the highly volatile M2, a moving average method (e.g., window size = 5 sets of continuous data, i.e., a 25-second time interval) is used to eliminate random noise and retain the trend signal, thus fixing the data according to [T1, T2, T3, V1, V2]. The variables [V3,Q1,Q2,Q3,M2] are arranged in sequence to form a 10-dimensional vector X. At this time, all variables are converted into dimensionless values ​​such as temperature / ℃, wind speed / (m / s), air volume / (m³ / h), and moisture / %). This not only integrates isolated parameters (temperature, air volume, wind speed, and moisture) into a unified data entity, breaking the "single-bed data silo" and providing structured input for subsequent multivariate analysis, but also binds the direct quality index M2 with process parameters T, Q, and V, establishing a causal analysis basis from process parameters to drying quality. During use, it can generate a new vector every 5 seconds and transmit it to the PCA model through the industrial bus, supporting streaming data processing.

[0039] Step 104: Input the input vector set into the principal component analysis model to perform PCA dimensionality reduction and extract the principal components, and input the principal components into the Hotelling T-squared monitoring model.

[0040] Principal component analysis (PCA) models refer to mathematical models that achieve dimensionality reduction through covariance matrix calculation and eigenvalue decomposition. PCA dimensionality reduction refers to the process of converting high-dimensional data into low-dimensional principal components. Principal components refer to the top k principal components (k=3) with a cumulative contribution rate ≥90%, including PC1 (characterizing overall drying intensity), PC2 (characterizing inter-bed consistency), and PC3 (characterizing final product quality trends). The Hotelling T-squared monitoring model refers to a statistical model for process monitoring based on the T-squared statistic constructed from principal components.

[0041] In some implementations, the covariance matrix C of the input vector set is calculated using the covariance matrix formula; then, eigenvalue decomposition is performed on the covariance matrix C to obtain eigenvalues ​​and corresponding eigenvectors; next, the cumulative contribution rate is calculated, and the smallest k value (k=3) that satisfies the cumulative contribution rate ≥ 90% is selected; then, the first three principal components are extracted to reduce the dimensionality of the 10-variable input vector set to 3 independent principal components, improving computational efficiency. Among them, PC1 loads the variables with the largest weights T2, Q2, and M2, PC2 loads V1, V2, V3, Q1, Q2, and Q3, and PC3 reflects the correlation between M2 and T3 / Q3. Thus, the principal components with process significance can be clearly identified, improving the interpretability of monitoring complex systems; finally, PC1, PC2, and PC3 are input into the Hotelling T-squared monitoring model.

[0042] Step 105: The Hotling T-square monitoring model constructs the T-square statistic based on the main components and calculates the control limit based on multiple batches of historical qualified data. The T-square statistic and the control limit are compared to determine the drying uniformity level, and the corresponding control strategy is output according to the drying uniformity level.

[0043] Among them, the Hotelling T-squared monitoring model refers to a process monitoring model based on principal component analysis to construct statistics. The T-squared statistic refers to... The calculated composite index (λ1, λ2, and λ3 are the eigenvalues ​​of the first three principal components). Control limits. This refers to the statistical threshold calculated based on historical acceptable data. Drying uniformity is categorized into four levels: acceptable, slightly abnormal, moderately abnormal, and severely abnormal. The control strategy refers to the parameter adjustment instructions output based on the abnormality level.

[0044] In some implementations, the system receives PC1, PC2, PC3 and their corresponding eigenvalues ​​λ1, λ2, λ3 from the principal component analysis model. Then, it calculates the T-squared statistic in real time to quantify the Mahalanobis distance between the current state and the historical acceptable benchmark. At this point, T² integrates the contributions of the three principal components, improving the anomaly sensitivity of multi-parameter coupling. Simultaneously, it calculates the control upper limit based on multiple batches of historical acceptable data (sample size m) and then applies the real-time T²... 2 Values ​​and control limits By conducting multi-dimensional comparisons and determining different agenda levels based on the comparison results, and finally invoking the corresponding control strategy based on the determined anomaly level, the source of the problem can be quickly identified, thus improving the efficiency of problem identification.

[0045] It should be noted that the contribution rate deviation refers to the absolute deviation between the current principal component contribution rate and the historical average, and the control limits need to be recalculated and updated every N production batches.

[0046] Based on the above technical solution, the temperatures (T1, T2, T3) and process parameters (Q1-V3) of the three beds are aligned with the spectral data of the two beds in real time through timestamp synchronization technology. For the first time, a 10-variable data entity covering the entire process of preheating, main drying and shaping is constructed. This solves the problems of blind spots in cross-bed anomaly transmission caused by traditional single-bed monitoring (such as the overload lag alarm of the two beds caused by insufficient preheating of the first bed) and the inability of isolated monitoring of single-bed data (such as using only single-point temperature or moisture rapid detection) to capture the synergistic effect of the three beds, resulting in misjudgment of uniformity. At the same time, the spectral model of the two beds outputs the M2 value every 5 seconds (replacing the traditional 30-minute offline detection) and binds it with the process parameters in real time, forming a closed-loop link of process parameters (T, Q, V), real-time quality (M2) and control feedback. This allows for rapid tracing of changes in different parameters at the same time point when an output fault occurs.

[0047] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 2 As shown, inputting the input vector set into the principal component analysis model to perform PCA dimensionality reduction and extract the principal components can be achieved through the following steps 201 to 204, which are explained in detail below: Step 201: Using the covariance matrix formula Calculate each set of input vectors The covariance matrix C is given by n, where n is the total number of input vector sets. Let X be the mean vector of each set of input vectors. i Let be the i-th set of input vectors.

[0048] The covariance matrix C is a symmetric matrix (10×10 dimensions) that quantifies the linear correlation between variables in the input vector set using mathematical formulas. The mean vector... It is a vector composed of the arithmetic mean of all samples in the input vector set in each variable dimension. Eigenvalue decomposition refers to the process of decomposing the covariance matrix into eigenvalues ​​and their corresponding eigenvectors, used to extract the main directions of change in the data.

[0049] In some implementations, 10 variable data points for the current moment are synchronously retrieved from an industrial database according to their timestamps, forming a single input vector. Next, calculate the total number of sets n for all historical and real-time input vector sets, and solve for the mean vector. (i.e., the average value of each variable across n sets of data); then based on the covariance matrix formula Expanding the computation: For each set of input vectors X i First calculate its relationship with the mean vector. The difference vector (X) i ), and then transpose it with itself (X). i ) T Multiplying them yields a single covariance contribution matrix; finally, summing all n contribution matrices and multiplying by a coefficient. This generates the final 10×10 dimensional covariance matrix C, where the diagonal elements are the variances of each variable and the off-diagonal elements are the covariances between the variables.

[0050] It should be noted that the calculation of the covariance matrix must be based on variable data at the same time with strictly aligned timestamps (time error <5ms) to ensure the comparability of different physical quantities (temperature / airflow / moisture). Furthermore, the calculation must ensure that n≥10 to satisfy the matrix decomposability condition, and the value of n dynamically increases with each production batch.

[0051] Step 202: Perform eigenvalue decomposition on the covariance matrix C to obtain eigenvalues ​​λ1>λ2>…>λ 10 >0 and the corresponding eigenvectors V1, V 1， V 10 .

[0052] Here, the eigenvalue λ is a scalar obtained after decomposing the covariance matrix C, representing the data variance contribution in the corresponding principal component direction (λ1 is the largest, λ...). 10 (Minimum). The eigenvector V is a non-zero vector (dimension 10×1) corresponding to the eigenvalues, indicating the projection direction of the principal components.

[0053] In some implementations, a numerical computation library (such as LAPACK) is first called to perform eigenvalue decomposition on the covariance matrix C; after decomposition, 10 eigenvalues ​​λ1, λ2, ..., λ3 are obtained. 10 Arranged in descending order, satisfying λ1>λ2>…>λ 10>0; simultaneously extract the corresponding 10 unit feature vectors V1, V 1， V 10 (each V) i (All are 10-dimensional column vectors), where V1 corresponds to the direction of the largest eigenvalue λ1, representing the projection axis with the strongest data variation; V2 corresponds to the direction of the second largest eigenvalue λ2 and is orthogonal to V1, and so on up to V... 10 The final output is the sorted sequence of eigenvalues ​​and the standardized eigenvector matrix.

[0054] It should be noted that the eigenvalues ​​must strictly satisfy the positive definite condition (λ). i >0), otherwise data anomalies need to be checked; orthogonality of feature vectors (V i T V j =0, when i≠j) ensures the independence of principal components; in the PCA model, only the first 3 feature vectors (V1,V2,V3) are used for dimensionality reduction.

[0055] Step 203, through Calculate the cumulative contribution rate LU of the first K principal components, select the smallest k value that satisfies the cumulative contribution rate LU≥90%, and satisfy k=3.

[0056] The cumulative contribution rate (LU) refers to the proportion of the sum of the variances of the first k principal components to the total variance, used to measure the amount of information retained after dimensionality reduction. The minimum k value refers to the minimum number of principal components required to satisfy a cumulative contribution rate ≥ 90%.

[0057] In some implementations, the sum of all 10 eigenvalues ​​is calculated first. Then, the first k eigenvalues ​​are accumulated sequentially. (k starts from 1); then through the formula Calculate the cumulative contribution rate of the current k value; cyclically increment the k value (k=1,2,) until the LU is satisfied for the first time. k ≥90%; at this point, record the minimum k value that satisfies the condition; finally, verify whether the k value is equal to 3 (if k<3, then force k=3); output the final determined k value (always 3).

[0058] It should be noted that even if the cumulative contribution rate of the first two principal components is ≥90% (e.g., λ1=6.0, λ2=3.0, the sum of 9.0 accounts for 90% of the total of 10.0), three principal components are still required. This ensures that the key process information of PC3 (final product quality trend) is retained and avoids excessive dimensionality reduction that could result in the loss of bed synergy characteristics.

[0059] Step 204, through The first three main components, PC1, PC2, and PC3, were extracted. PC1 represents the overall drying intensity, and the variables with the largest loading weights are T2, Q2, and M2. PC2 represents inter-bed consistency, and the variables with the largest loading weights are V1, V2, V3, Q1, Q2, and Q3. PC3 represents the final product quality trend, and its loading weight reflects the correlation between M2 and T3 and Q3.

[0060] Among them, the load weight refers to the absolute value of the coefficient of the corresponding variable in the feature vector, which determines the degree of contribution of the variable to the principal component.

[0061] In some implementations, the first three feature vectors V1, V2, and V3 (all 10-dimensional unit vectors) are extracted from the eigenvalue decomposition results; for the real-time standardized input vector X, its relationship with the historical mean vector is calculated. The difference vector (X) Then, matrix operations are performed. Calculate the first principal component (the dot product result is a scalar), where the absolute values ​​of the weighting coefficients corresponding to T2, Q2, and M2 in V1 are the largest, making the PC1 value mainly reflect the drying intensity of the second bed; similarly, through... The second principal component is calculated, and the high weights of V1-V3 and Q1-Q3 in V2 quantify the synergy of the three-bed parameters; finally, through... The third principal component (PPC) is calculated. The weight correlation between M2 and T3 and Q3 in V3 reflects the moisture trend of the final product. The three scalar values ​​PC1, PC2, and PC3 are output for subsequent monitoring.

[0062] It should be noted that PC1 focuses on key parameters (T2, Q2, M2) in the main drying stage of the two beds, and their anomalies are directly related to the overall drying intensity; PC2 evaluates the synergy of the three beds through the wind speed and air volume parameters (V1-V3, Q1-Q3) of each bed, and the load variables cover all bed layers; PC3 uses the weighted correlation between M2 and the three-bed shaping parameters (T3, Q3) to predict the moisture stability of the final product; the load weights of the feature vectors V1, V2, and V3 are fixed and determined by training with historical data.

[0063] Based on the above technical solution, the maximum variables of PC1 load weight are defined as T2, Q2, and M2, representing the overall drying intensity; the maximum variables of PC2 load weight are V1-V3 and Q1-Q3, representing the consistency between beds; and the load weight of PC3 reflects the correlation between M2 and T3 and Q3, representing the final product quality trend. This allows for a strong binding between statistical dimensionality reduction results and the physical meaning of the process (e.g., PC1 focuses on the core parameters of the two-bed main drying process), improving the interpretability of the model. This solves the problem that the statistical PCA only pursues mathematical dimensionality reduction (stopping when the cumulative contribution rate is ≥90%) without associating it with the process logic, resulting in the inability to locate the specific bed responsibility parameters when anomalies occur. Simultaneously, the root causes of anomalies are directly mapped through principal component definition: PC1 anomaly is the main drying problem of the second bed (T2, Q2, M2); PC2 anomaly is the synergistic problem of the three beds (V1-V3, Q1-Q3); PC3 anomaly is the fluctuation of the final product quality (M2 is related to T3, Q3). This can provide a precise basis for subsequent graded regulation (such as adjusting the first bed for mild anomalies, and adjusting the second and third beds for moderate anomalies), thereby avoiding ignoring the functional differences of each bed and improving the targeting of regulation.

[0064] In one possible implementation of this application embodiment, the Hotelling T-squared monitoring model constructs a T-squared statistic based on the input principal components and calculates control limits based on multiple batches of historical qualified data. The determination of the drying uniformity level by comparing the T-squared statistic and the control limits can be achieved through the following steps 301 to 303, which are described in detail below: Step 301: Based on the first three principal components PC1, PC2, and PC3 and their corresponding eigenvalues ​​λ1, λ2, and λ3, through... Calculate the real-time T-squared statistic T 2 .

[0065] Among them, the eigenvalues ​​λ1, λ2, and λ3 correspond to the variance contributions of PC1, PC2, and PC3, respectively (λ1>λ2>λ3>0). The T-squared statistic T² refers to the comprehensive monitoring index calculated by principal components and eigenvalues, used to quantify the degree of deviation between the current state and the historical benchmark.

[0066] In some implementations, the standardized PC1, PC2, and PC3 values ​​are first obtained in real time from the principal component analysis model; simultaneously, λ1, λ2, and λ3 are read from pre-stored eigenvalue results (the eigenvalues ​​λ1, λ2, and λ3 are fixed results from historical data training); then, the formula is used... Item-by-item calculation is performed, that is, the square value of each principal component is divided by its corresponding eigenvalue and then summed. At this time, T² is the Mahalanobis distance between the multi-parameter coupled state and the qualified benchmark, and the contribution weight of the principal components (the inverse of λ) is incorporated, so that the fluctuation of high-variance principal components (such as PC1) has a greater impact on T²; finally, the real-time T² statistic is output.

[0067] Step 302: Based on multiple batches of historical qualified data, through... Calculate the T-squared statistic T 2 upper limit of control , of which F α (3,m 3) is the critical value of the F distribution at a significance level of α=0.05, and m is the number of historical qualified data samples.

[0068] Among them, the historical qualified data for multiple batches refers to the data set corresponding to production batches whose drying uniformity level was judged to be qualified. Control Upper Limit It is the threshold boundary of the T² statistic, calculated using a statistical formula based on historical qualified data, and is used to determine whether the drying uniformity is abnormal.

[0069] In some implementations, multiple batches of historical qualified data samples are extracted from an industrial database, with the sample size recorded as m. Next, the significance level α = 0.05 (i.e., a confidence level of 95%) and the degrees of freedom parameter (3, m-3) are determined. Then, the critical value Fc of the Fdistribution at α = 0.05 is obtained by consulting an Fdistribution table or by calling a statistical calculation library. α (3, m-3). Finally, substitute the critical value into the formula. Complete the calculation. This will be triggered automatically during system initialization or after every N production batches, and the calculation results will be stored in the monitoring model.

[0070] It should be noted that the number of historical samples m > 3; otherwise, the denominator (m-3) will be zero, causing the formula to fail. Furthermore, the calculation needs to be recalculated after every N production batches (e.g., N=20). This ensures that the control limits adapt to process changes. Furthermore, the new control limits must meet the requirement that the adaptation error is ≤1.5%, which is achieved by comparing the deviations between the old and new control limits.

[0071] Step 303, T 2 With control upper limit By comparing multiple dimensions, the drying uniformity level is output.

[0072] T 2 With control upper limit The drying uniformity level is output through multi-dimensional comparison and evaluation, including: When T 2 ≤ When the drying uniformity is determined to be qualified, the drying uniformity level is deemed acceptable.

[0073] when <T 2 ≤1.2× If the contribution rate deviation of any one of the principal components PC1, PC2, and PC3 is greater than 5%, it is judged as a mild anomaly.

[0074] When 1.2× <T 2 ≤1.5× If the contribution rate deviation of any two of the principal components PC1, PC2, and PC3 is greater than 5%, it is judged as a moderate anomaly.

[0075] When T 2 >1.5× At that time, it was determined to be a severe abnormality.

[0076] It should be noted that the process is executed every 5 seconds and feedback is sent to the control unit. Furthermore, the threshold for contribution rate deviation is fixed at >5%, and the real-time contribution rates of the three principal components must be calculated simultaneously during the determination.

[0077] Based on the above technical solution, three principal components with clear process significance (PC1 = drying intensity, PC2 = inter-bed consistency, PC3 = quality trend) are extracted through PCA dimensionality reduction. Combined with their eigenvalues ​​λ1, λ2, and λ3, a weighted statistical formula is constructed, which compresses the 10-dimensional process parameters into a single sensitive indicator. Furthermore, by using inverse weighting of eigenvalues ​​(e.g., if λ1 is minimized, PC1 fluctuations are amplified), high-variance principal components (e.g., parameters of the second bed under PC1 load) dominate the abnormal response, improving the monitoring sensitivity of the core drying stage (the second bed). This solves the problem of existing isolated single-bed monitoring (e.g., using only the temperature of the second bed) failing to capture the coupling effect of cross-bed parameters (e.g., insufficient preheating of the first bed leading to overload of the second bed) and resulting in a high false alarm rate. Simultaneously, the control limit is calculated based on historical qualified batch data (sample size m), and then the correction term suppresses small sample deviations. This allows the control limit to be automatically updated with each production batch (triggered every N batches), adapting to the process differences of different traditional Chinese medicine formulas (e.g., Wenweishu granules → Yangweishu granules). This solves the problem that the traditional fixed threshold method cannot adapt to formula changes and batch fluctuations in traditional Chinese medicine granules, requiring repeated manual calibration. Finally, the dual-condition approach of "threshold comparison + principal component contribution rate deviation verification" can accurately locate the responsible bed layer.

[0078] In one possible implementation of this application embodiment, the control strategy corresponding to the level of drying uniformity can be output through the following steps 401 to 405, which are described in detail below: Step 401: Based on the principal component analysis model, identify the abnormal principal component types PC1, PC2 or PC3, and compare them with historical abnormal cases in the industrial database to determine the abnormal principal component types corresponding to different levels.

[0079] Based on principal component analysis, abnormal principal component types PC1, PC2, or PC3 are identified. By comparing these abnormal cases with historical anomaly cases in the industrial database, the types of abnormal principal components corresponding to different levels are determined, including: If an abnormality is detected in the main component PC1, it is determined to be a deviation in the two-bed main drying parameters or the material state.

[0080] If an abnormality is detected in the principal component PC2, it is determined to be a mismatch between the three-bed shaping stage and the two-bed main drying stage.

[0081] If an abnormality is detected in principal component PC3, it is determined that the synergy between the first and second beds is poor.

[0082] The industrial database is a structured storage system that integrates temperature, process parameters, spectral data, and predicted moisture content, and supports comparison of historical anomaly cases.

[0083] In some implementations, the input vector set (containing 10 variables: T1-T3, V1-V3, Q1-Q3, and M2) is acquired in real time. The covariance matrix is ​​calculated using a principal component analysis model, and eigenvalue decomposition is performed to extract the first three principal components (PC1, PC2, and PC3). Subsequently, anomaly types are identified based on the principal component contribution rate deviation (the absolute deviation between the current contribution rate and the historical mean): if the PC1 contribution rate deviation is >5%, it is determined to be an anomaly in the main drying parameters or material state of the second bed (because PC1 is loaded with T2, Q2, and M2); if the PC2 contribution rate deviation is >5%, it is determined to be an anomaly in the coordination between the third bed's shaping stage and the main drying of the second bed (because PC2 is loaded with the wind speed and air volume of each bed); if the PC3 contribution rate deviation is >5%, it is determined to be an anomaly in the coordination between the first and second beds (because PC3 is associated with M2 and T3 / Q3). Next, by comparing historical anomaly cases in the industrial database, the anomaly type is further verified by matching similar feature value combinations (such as the fluctuation pattern of T2 / Q2 / M2 when PC1 is anomaly). Finally, the anomaly principal component type localization results corresponding to different drying uniformity levels (mild, moderate, and severe) are output.

[0084] Step 402: When a slight anomaly is determined, the associated anomalous principal component PC3 is invoked, the optimal parameter ratio in the historical database is invoked, the air inlet velocity V1 of one bed is adjusted within ±0.1 m / s, and the temperature T1 of one bed is adjusted within ±0.5℃ to reduce the temperature difference between beds to the acceptable range.

[0085] The historical database stores the optimal process parameter ratios (such as the baseline relationship between V1 and T1) from multiple batches of historical qualified data. The inter-bed temperature difference refers to the difference between the preheating temperature of the first bed (T1) and the main drying temperature of the second bed (T2), and the qualified range is set by the historical database (such as |T1-T2|≤3℃).

[0086] In some implementations, when the Hotling T-square monitoring model determines that the drying uniformity level is slightly abnormal and identifies a PC3 contribution rate deviation >5%, it calls the "optimal parameter ratio" (such as V1 baseline value 1.2m / s, T1 baseline value 65℃) that matches the current Chinese medicine granule formula in the industrial database. First, it adjusts the air inlet velocity V1 of the first bed within ±0.1m / s (e.g., from 1.25m / s to 1.15m / s), and simultaneously adjusts the preheating temperature T1 of the first bed within ±0.5℃ (e.g., from 64℃ to 64.5℃). By reducing the temperature difference between the first and second beds (e.g., reducing |T1-T2| from 4.2℃ to 2.8℃), the inter-bed synergy returns to the acceptable range.

[0087] It should be noted that the adjustment range of V1 should not exceed ±0.1 m / s (to avoid sudden changes in wind speed affecting the fluidization state of particles), and the adjustment range of T1 should not exceed ±0.5℃ (to prevent insufficient or excessive preheating). If the T² statistic does not return to the qualified range after two consecutive adjustments, the procedure for handling moderate anomalies should be upgraded.

[0088] Step 403: When the condition is determined to be moderately abnormal, call the associated abnormal principal component PC2, adjust the air intake Q2 of the second bed within the range of ±5m³ / h, and open the guide vane of the second bed and adjust its angle within the range of ±3°. Based on the correlation between the predicted moisture content M2 of the second bed and the temperature of the second bed, adjust the temperature T2 of the second bed within the range of ±1℃.

[0089] The second-bed guide vane is a mechanical device used to optimize particle fluidization, with an angle adjustment range of ±3°. The correlation law refers to the quantitative relationship between M2 and T2 established based on historical data from an industrial database (e.g., increasing M2 requires decreasing T2). Range limitations include Q2 adjustment ±5m³ / h, guide vane angle ±3°, and T2 adjustment ±1℃, ensuring precise control.

[0090] In some implementations, when the Hotelling T-squared monitoring model determines that the drying uniformity level is moderately abnormal and identifies a PC2 contribution rate deviation >5% (or triggered by deviations of other principal components), it calls historical abnormal cases associated with PC2 in the industrial database for comparison to confirm that the root cause of the problem is an abnormality in inter-bed synergy; then, it automatically adjusts the air intake Q2 of the second bed within ±5m³ / h (e.g., increasing it from the baseline value of 50m³ / h to 55m³ / h or decreasing it to 45m³ / h), while simultaneously opening the guide vanes of the second bed and adjusting their angle within ±3° (e.g., ...). To optimize particle fluidization uniformity, the temperature is adjusted from 0° to 3° or from -3°. Then, based on the correlation between the predicted water content M2 and the temperature T2 in the second bed (this correlation is derived from fitting multiple batches of historical data from the database; for example, a 0.5% increase in M2 corresponds to a corresponding decrease in T2 of approximately 0.8°), T2 is fine-tuned within ±1° (e.g., decreasing from 70° to 69° or increasing to 71°) to ensure that the M2 value remains stable within the acceptable range. During the adjustment process, Q2, T2, and M2 data are read from the database every 5 seconds to evaluate the effect. If the T² statistic is ≤ 0.5% for three consecutive monitoring cycles, the effect is considered achieved. If the PC2 contribution rate deviation is ≤5%, the regulation is considered successful; otherwise, the processing procedure is upgraded.

[0091] Step 404: When a severe anomaly is determined, the associated principal component of the anomaly PC1 is invoked, and the air intake Q2 of the second bed is adjusted within ±5 m³ / h. The second bed guide vane is opened and its angle is adjusted within ±3°. At the same time, based on the correlation between M2 and the second bed temperature T2 in the industrial database, T2 is adjusted within ±1°.

[0092] In some implementations, when the Hotling T-squared monitoring model determines that the drying uniformity level is severely abnormal, the associated anomalous principal component PC1 is invoked (because the variables T2, Q2, and M2 with the largest load weights in PC1 are abnormal). First, the airflow rate Q2 of the second bed is adjusted within ±5 m³ / h (e.g., increasing from the baseline value of 55 m³ / h to 60 m³ / h or decreasing to 50 m³ / h). At the same time, the guide vanes of the second bed are opened and their angle is adjusted within ±3° (e.g., from 0° to +3° or -3°) to optimize particle fluidization. Uniformity; then, based on the historical correlation between M2 and T2 in the industrial database (this correlation was derived through fitting multiple batches of data, for example, a 0.5% increase in M2 requires a corresponding decrease in T2 of approximately 0.8℃), the secondary bed temperature T2 is adjusted in real time within ±1℃ (e.g., from 72℃ to 71℃ or from 73℃) to ensure that the M2 value remains stable within the acceptable range; during the adjustment process, real-time Q2, T2, and M2 data are read from the database every 5 seconds to evaluate the T² statistic and moisture deviation. If the T² statistic is ≤ 0.5% for three consecutive monitoring cycles, the adjustment is considered successful. If the moisture deviation meets the standard (e.g., |M2-target value|≤0.3%), the regulation is considered successful; otherwise, it continues to iterate until the standard is met.

[0093] It should be noted that the Q2 adjustment range should not exceed ±5 m³ / h (to prevent sudden changes in airflow from causing particle splashing or uneven drying), the guide vane angle adjustment is limited to ±3° (to avoid excessive angles disrupting the fluidized bed balance structure), and the T2 adjustment must be within ±1℃ (because traditional Chinese medicine granules are highly heat-sensitive, excessive temperature fluctuations can affect the active ingredients). If the T² statistic does not improve within 30 seconds after adjustment (e.g., the data still exceeds the limit after 6 consecutive readings), the system will automatically alarm and switch to manual intervention mode.

[0094] Step 405: During the regulation process, read real-time monitoring data from the database every 5 seconds to evaluate the regulation effect until the T² statistic returns to the qualified range and the moisture deviation meets the standard.

[0095] The control process refers to the stage of parameter adjustment when drying uniformity is abnormal. Real-time monitoring data includes temperature data for each bed (T1, T2, T3), process parameters (Q1-Q3, V1-V3), and predicted moisture content M2 for the second bed. The acceptable range refers to when T² ≤ The drying uniformity level is qualified. Moisture deviation refers to the absolute difference between the predicted moisture content M2 of the second bed and the target moisture value. The condition for compliance is |M2-target|≤0.3% (the threshold is set based on historical qualified data).

[0096] In some implementations, during the execution of the control strategy, the system automatically reads the latest bed temperature data, process parameters, and predicted water content M2 of the second bed from the database every 5 seconds, and ensures data time alignment through timestamp synchronization technology. Then, this data is integrated to generate a standardized input vector set (containing 10 variables: T1-T3, V1-V3, Q1-Q3, and M2). Next, the input vector set is input into a principal component analysis model for PCA dimensionality reduction to extract the principal components PC1, PC2, and PC3. Subsequently, the principal components are input into the Hotelling T-squared monitoring model, and the results are processed using the formula... Calculate the real-time T² statistic while obtaining the current M² value; then, compare the T² statistic with the control upper limit calculated based on multiple batches of historical qualified data. If T²≤ If the moisture deviation |M2-target|≤0.3%, the control effect is deemed to have met the target and the control is terminated; otherwise, the current control strategy is continued and the above steps are repeated until both conditions are met simultaneously.

[0097] Based on the above technical solution, the 10 parameters are reduced to 3 principal components with clear process significance (PC1 drying intensity, PC2 inter-bed synergy, PC3 quality trend) using the PCA model. The statistical model load weights (e.g., PC1 load T2 / Q2 / M2) are bound to physical meaning, enabling anomaly tracing back to specific bed layers and overcoming the blind spots of traditional single-point monitoring. Simultaneously, different principal components (PC3 / PC2 / PC1) are matched according to the anomaly level (light / medium / severe), and the control range is defined. This allows for the identification of the responsible bed layer based on the physical meaning of the principal components (e.g., adjusting bed one for PC3 anomalies, adjusting bed two for PC1 anomalies), avoiding "one-size-fits-all" control that ignores the functional differences between beds, leading to over-control or ineffective responses. Finally, real-time data is read every 5 seconds to dynamically evaluate the T² statistic and moisture deviation, forming a closed loop of monitoring, control, and verification, improving the success rate of control.

[0098] In one possible implementation of this application embodiment, after outputting the corresponding control strategy according to the level of drying uniformity, it can also be implemented through the following steps 501 to 405, which are described in detail below: Step 501: Based on the historical batch data accumulated in the industrial database, after every N production batches are completed, extract the historical qualified data of those N batches and recalculate the upper limit of the control of the Hotelling T-squared statistic. , where N is the set period, and the updated control limit adaptation error is ≤1.5%.

[0099] Here, N production batches refer to a set integer period value, representing the number of production batches completed. Historical qualified data refers to the set of data samples that were judged to be qualified in terms of drying uniformity.

[0100] In some implementations, based on the continuous accumulation of historical batch data in the industrial database, after N production batches are completed (N is a preset integer period value, such as N=20), the system automatically triggers a data extraction process to filter historical data samples from the database whose drying uniformity level has been determined to be qualified from those N batches. Then, based on these selected qualified historical data samples, the formula in the Hotelling T-squared monitoring model is called to recalculate and generate a new control upper limit. Next, it is verified whether the adaptation error of the updated control upper limit meets the requirement of ≤1.5% (e.g., by comparing the percentage deviation between the old and new control upper limit values; if it exceeds the limit, iterative calculation or optimization of the data samples is performed). Finally, the updated control upper limit is dynamically applied to the Hotelling T-squared monitoring model to adjust the judgment threshold for the drying uniformity level (e.g., T). 2 (Minor, moderate, and severe anomaly thresholds) to ensure the model is adapted to the latest production batch data.

[0101] It should be noted that the setting period of N is usually configured according to the production scale and process stability. The default value can be set to 20 batches, but the sample size m must be greater than 3 to avoid the formula from becoming invalid. The adaptation error ≤ 1.5% is achieved through the deviation verification mechanism. If the error exceeds the limit, the data needs to be re-screened or the sample size needs to be adjusted.

[0102] Step 502: Based on the updated control limit The threshold for determining the drying uniformity level is dynamically adjusted.

[0103] In some implementations, after the industrial database completes the control limit update (e.g.) (Updating from the original value of 10.0 to 10.5) immediately triggers the dynamic adjustment process; then, according to the defined drying uniformity level judgment rules, the threshold for mild anomalies is recalculated to 1.2 × 10.5 = 12.6, the threshold for moderate anomalies is recalculated to 1.5 × 10.5 = 15.75, while the threshold for severe anomalies remains at T. 2 >15.75; Then, the system writes the recalculated threshold into the memory parameters of the Hotelling T-squared monitoring model, overwriting the original judgment threshold; finally, in subsequent real-time monitoring, the model directly uses the new threshold for multi-dimensional judgment (e.g., when the real-time T... 2 When the value is 13.0, since 12.6 < 13.0 ≤ 15.75, and it is necessary to check whether the principal component contribution rate deviation meets the condition of > 5%, it is comprehensively judged as mild or moderate abnormality, and the closed-loop adaptation is completed.

[0104] It should be noted that the threshold ratio remains fixed (1.2 times for mild cases and 1.5 times for moderate cases) to ensure consistency in the judgment logic; dynamic adjustments only modify the threshold value and do not change the level definition or principal component analysis rules (e.g., moderate anomalies still need to meet the T-value). 2 >1.2× Or the difference in contribution rates between the two principal components is >5%.

[0105] Step 503: For different Chinese herbal medicine granule formulas, call the database to match the historical optimal process parameter ratios to generate the benchmark input vector set for the formula.

[0106] Among them, Chinese medicine granule formulas refer to different types of Chinese medicine preparations (such as Wenweishu granules and Yangweishu granules).

[0107] In some implementations, when the system detects the start of production of a new traditional Chinese medicine granule formula (such as Yangweishu granules), it automatically calls the historical optimal process parameter ratios that perfectly match the formula stored in the industrial database (e.g., extracting T1=65℃, V1=1.2m / s, Q2=50m from historical qualified batches). 3The optimal combination of parameters such as / h is determined; then, based on the proportions of these parameters, the optimal values ​​of 10 predefined variables are determined in a fixed order ([T1,T2,T3,V1,V2,V3,Q1,Q2,Q3,M2)). T The system integrates and generates a standardized set of benchmark input vectors. Then, the system performs the same preprocessing procedure on the vector set as on the real-time data (including removing outliers exceeding the threshold and smoothing M2 by moving average). Finally, the generated set of benchmark input vectors is written into the memory of the principal component analysis model as a process benchmark for the formulation for comparison in subsequent real-time monitoring.

[0108] Based on the above technical solution, multiple batches of historical qualified data (such as T1, Q2, M2, etc.) are accumulated. Data is extracted during periodic updates (every N batches) to support the recalculation of the control upper limit. The calculation accuracy is ensured through an error verification module (deviation ≤1.5%). At the same time, the optimal parameter ratios of different formulas are stored and quickly called when switching formulas to generate a benchmark vector set. This allows for synchronous adjustment of the control upper limit based on changes in material characteristics (such as humidity fluctuations) between batches of traditional Chinese medicine production, avoiding the possibility of misjudgment caused by static control limits (such as qualified batches being falsely reported as abnormal), and achieving threshold self-correction.

[0109] The above mainly describes the solutions of the embodiments of this application from the perspective of device implementation. It is understood that each device, for example, an online monitoring and improvement device for the uniformity of fluidized bed drying of traditional Chinese medicine granules, includes at least one of the hardware structures and software modules corresponding to each function in order to achieve the above-mentioned functions. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware 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.

[0110] This application embodiment can divide the online monitoring and improvement device for the uniformity of fluidized bed drying of traditional Chinese medicine granules into functional units based on the above method example. For example, each function can be divided into its own functional unit, or two or more functions can be integrated into the same processing unit. The integrated unit can be implemented in hardware or software. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0111] When using integrated units, Figure 6A possible structural schematic diagram of a device for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules (referred to as a device 60 for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules) involved in the above embodiments is shown. The device 60 for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules includes a processing unit 601 and a communication unit 602, and may also include a storage unit 603. Figure 6 The schematic diagram shown can be used to illustrate the structure of the online monitoring and improvement device for the uniformity of fluidized bed drying of traditional Chinese medicine granules involved in the above embodiments.

[0112] when Figure 6 The schematic diagram shown illustrates the structure of the online monitoring and improvement device for the uniformity of fluidized bed drying of traditional Chinese medicine granules involved in the above embodiments. The processing unit 601 is used to control and manage the operation of the online monitoring and improvement device for the uniformity of fluidized bed drying of traditional Chinese medicine granules. The communication unit 602 is used for the online monitoring and improvement device for the uniformity of fluidized bed drying of traditional Chinese medicine granules to communicate with other devices. The storage unit 603 is used to store the program code and data of the online monitoring and improvement device for the uniformity of fluidized bed drying of traditional Chinese medicine granules.

[0113] For example, communication unit 602 is used to acquire temperature data of each bed, process parameters and particle spectrum data of the second bed outlet; Processing unit 601 is used to input the particle spectral data from the second bed outlet into the spectral model, output the predicted moisture content M2 of the second bed, and integrate the temperature data, process parameters, and predicted moisture content M2 of each bed to construct a database. It reads the temperature, process parameters, and predicted moisture content M2 of each bed in real time from the database, and after preprocessing, generates an input vector set containing 10 variables. The input vector set is then input into a principal component analysis (PCA) model to perform dimensionality reduction and extract the principal components. These principal components are then input into a Hotelling T-squared monitoring model. The Hotelling T-squared monitoring model constructs a T-squared statistic based on the principal components and calculates control limits based on multiple batches of historical qualified data. It compares the T-squared statistic with the control limits to determine the drying uniformity level and outputs the corresponding control strategy based on the drying uniformity level.

[0114] In one possible implementation, the processing unit 601 is further configured to construct a database integrating temperature data from each bed, process parameters, and the predicted moisture content M2 of the second bed. This includes updating the temperature data and process parameters of each bed every 6 seconds. The temperature data for each bed includes preheating temperature data T1 for the first bed preheating stage, main drying temperature data T2 for the second bed main drying stage, and shaping temperature data T3 for the third bed shaping stage. The process parameters include inlet air volume Q1 and inlet air velocity V1 for the first bed, inlet air volume Q2 and inlet air velocity V2 for the second bed, and inlet air volume Q3 and inlet air velocity V3 for the third bed. Timestamp synchronization technology is used to time-align the temperature data, process parameters, particle spectral data at the outlet of the second bed, and the predicted moisture content M2. A database was constructed by integrating preheating temperature data T1, main drying temperature data T2, shaping temperature data T3, air inlet velocity of bed 1 V1, air inlet velocity of bed 2 V2, air inlet velocity of bed 3 V3, air inlet volume of bed 1 Q1, air inlet volume of bed 2 Q2, air inlet volume of bed 3 Q3, and predicted moisture content of bed 2 M2.

[0115] In one possible implementation, the processing unit 601 is further configured to generate an input vector set containing 10 variables after preprocessing, including: calculating anomaly thresholds based on multiple batches of historical qualified data, and removing data exceeding the anomaly thresholds from the preheating temperature data T1, main drying temperature data T2, shaping temperature data T3, first bed air inlet velocity V1, second bed air inlet velocity V2, third bed air inlet velocity V3, first bed air volume Q1, second bed air volume Q2, and third bed air volume Q3. The predicted moisture content M2 of the second bed is smoothed using a moving average method with a window size of 6 sets of continuous data. The preheating temperature data T1, main drying temperature data T2, shaping temperature data T3, first bed air inlet velocity V1, second bed air inlet velocity V2, third bed air inlet velocity V3, first bed air volume Q1, second bed air volume Q2, third bed air volume Q3, and the predicted moisture content M2 of the second bed are then integrated to obtain the input vector set containing 10 variables.

[0116] In one possible implementation, the processing unit 601 is further configured to input the input vector set into the principal component analysis model to perform PCA dimensionality reduction and extract the principal components, including: using the covariance matrix formula Calculate each set of input vectors The covariance matrix C is given by n, where n is the total number of input vector sets. Let X be the mean vector of each set of input vectors. i Let C be the i-th set of input vectors. Perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalues ​​λ1>λ2>…>λ. 10 >0 and the corresponding eigenvectors V1, V 1， V 10 .pass Calculate the cumulative contribution rate LU of the first K principal components, and select the smallest k value that satisfies LU ≥ 90%, where k = 3. The first three main components, PC1, PC2, and PC3, were extracted. PC1 represents the overall drying intensity, and the variables with the largest loading weights are T2, Q2, and M2. PC2 represents inter-bed consistency, and the variables with the largest loading weights are V1, V2, V3, Q1, Q2, and Q3. PC3 represents the final product quality trend, and its loading weight reflects the correlation between M2 and T3 and Q3.

[0117] In one possible implementation, the processing unit 601 is further configured to construct a T-square statistic based on the principal components of the Hotelling T-square monitoring model, calculate control limits based on multiple batches of historical qualified data, and determine the drying uniformity level by comparing the T-square statistic with the control limits, including: based on the first three principal components PC1, PC2, and PC3 and their corresponding eigenvalues ​​λ1, λ2, and λ3, through... Calculate the real-time T-squared statistic T 2 Based on multiple batches of historical qualified data, through Calculate the T-squared statistic T 2 upper limit of control , of which F α (3,m 3) is the critical value of the F-distribution at a significance level of α=0.06, and m is the number of historical qualified data samples. T 2 With control upper limit By comparing multiple dimensions, the drying uniformity level is output.

[0118] In one possible implementation, the processing unit 601 is further configured to process T 2 With control upper limit The drying uniformity level is output by comparing multiple dimensions, including: when T 2 ≤ At that time, the drying uniformity grade is judged to be qualified. <T 2 ≤1.2× If the contribution rate deviation of any one of the principal components PC1, PC2, and PC3 is >6%, it is judged as a mild anomaly. When 1.2× <T 2 ≤1.6× If the contribution rate deviation of any two of the principal components PC1, PC2, and PC3 is >6%, it is judged as moderately abnormal. 2 >1.6× At that time, it was determined to be a severe abnormality.

[0119] In one possible implementation, the processing unit 601 is further configured to output corresponding control strategies based on the level of drying uniformity, including: identifying abnormal principal component types PC1, PC2, or PC3 based on a principal component analysis model, and determining the abnormal principal component types corresponding to different levels by comparing historical abnormal cases in the industrial database. When a mild abnormality is determined, the associated abnormal principal component PC3 is invoked, the optimal parameter ratio in the historical database is invoked, the inlet air velocity V1 of the first bed is adjusted within ±0.1 m / s, and the temperature T1 of the first bed is adjusted within ±0.6℃ to reduce the temperature difference between beds to the acceptable range. When a moderate abnormality is determined, the associated abnormal principal component PC2 is invoked, the inlet air volume Q2 of the second bed is adjusted within ±6 m³ / h, the guide vanes of the second bed are opened and their angles are adjusted within ±3°, and the temperature T2 of the second bed is adjusted within ±1℃ based on the correlation between the predicted moisture content M2 of the second bed and the temperature of the second bed. When a severe anomaly is detected, the associated principal component of the anomaly, PC1, is invoked. The airflow rate Q2 in the secondary bed is adjusted within ±6 m³ / h, and the secondary bed baffles are opened and their angle adjusted within ±3°. Simultaneously, based on the correlation between M2 and the secondary bed temperature T2 in the industrial database, T2 is adjusted within ±1°C. During the control process, real-time monitoring data is retrieved from the database every 6 seconds to evaluate the control effect until the T² statistic returns to the acceptable range and the moisture deviation meets the standard.

[0120] In one possible implementation, the processing unit 601 is further configured to identify abnormal principal component types PC1, PC2, or PC3 based on a principal component analysis model, and, by comparing with historical abnormal cases in the industrial database, determine the abnormal principal component types corresponding to different levels of location. This includes: if an abnormal principal component PC1 is identified, it is determined to be a deviation in the secondary bed main drying parameters or material state. If an abnormal principal component PC2 is identified, it is determined to be a matching problem between the secondary bed and the tertiary bed main drying stage. If an abnormal principal component PC3 is identified, it is determined to be a poor coordination between the primary and secondary beds.

[0121] In one possible implementation, the processing unit 601 is further configured to output the corresponding control strategy according to the level of drying uniformity, and then further configured to: based on historical batch data accumulated in the industrial database, after completing N production batches, extract the historical qualified data of those N batches and recalculate the control upper limit of the Hotelling T-squared statistic. Where N is the set period, and the updated control limit adaptation error is ≤1.6%. Based on the updated control upper limit... The threshold for determining the drying uniformity level is dynamically adjusted. For different Chinese herbal medicine granule formulations, the historical optimal process parameter ratios are matched from the database to generate a baseline input vector set for that formulation.

[0122] The processing unit 601 can be a processor or a controller, and the communication unit 602 can be a communication interface, transceiver, transceiver circuit, transceiver device, etc. The term "communication interface" is a general term and may include one or more interfaces. The storage unit 603 can be a memory. When the online monitoring and improvement device 60 for the uniformity of fluidized bed drying of traditional Chinese medicine granules is a chip, the processing unit 601 can be a processor or a controller, and the communication unit 602 can be an input interface and / or an output interface, pins, or circuits, etc. The storage unit 603 can be a storage unit within the chip (e.g., a register, cache, etc.) or a storage unit located outside the chip (e.g., read-only memory (ROM), random access memory (RAM, etc.)).

[0123] The communication unit can also be referred to as a transceiver unit. The antenna and control circuit with transceiver functions in the online monitoring and improvement device 60 for the uniformity of fluidized bed drying of traditional Chinese medicine granules can be considered as the communication unit 602 of the online monitoring and improvement device 60 for the uniformity of fluidized bed drying of traditional Chinese medicine granules, and the processor with processing functions can be considered as the processing unit 601 of the online monitoring and improvement device 60 for the uniformity of fluidized bed drying of traditional Chinese medicine granules. Optionally, the device in the communication unit 602 used to implement the receiving function can be considered as a communication unit, which is used to execute the receiving steps in the embodiments of this application. The communication unit can be a receiver, a receiver circuit, etc. The device in the communication unit 602 used to implement the transmitting function can be considered as a transmitting unit, which is used to execute the transmitting steps in the embodiments of this application. The transmitting unit can be a transmitter, a transmitter, a transmitting circuit, etc.

[0124] Figure 6 If the integrated units in the process are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. Storage media for storing computer software products include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0125] In implementation, each step of the method provided in this embodiment can be completed by integrated logic circuits in the processor or by instructions in software form. The steps of the method disclosed in the embodiments of this application can be directly manifested as being executed by a hardware processor, or being executed by a combination of hardware and software modules in the processor.

[0126] The processor in this application may include, but is not limited to, at least one of the following: a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), a microcontroller unit (MCU), or an artificial intelligence processor, etc., and other computing devices that run software. Each computing device may include one or more cores for executing software instructions to perform calculations or processing. The processor may be a standalone semiconductor chip or integrated with other circuits into a single semiconductor chip. For example, it may form a System-on-a-Chip (SoC) with other circuits (such as encoding / decoding circuits, hardware acceleration circuits, or various bus and interface circuits), or it may be integrated as a built-in processor within an ASIC. The ASIC of this integrated processor may be packaged separately or together with other circuits. In addition to the cores for executing software instructions to perform calculations or processing, the processor may further include necessary hardware accelerators, such as field-programmable gate arrays (FPGAs), programmable logic devices (PLDs), or logic circuits that implement dedicated logic operations.

[0127] The memory in the embodiments of this application may include at least one of the following types: read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions; random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions; or electrically erasable programmable-only memory (EEPROM). In some scenarios, the memory may also be compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0128] This application also provides a computer-readable storage medium including instructions that, when run on a computer, cause the computer to perform any of the methods described above.

[0129] This application also provides a computer program product containing instructions that, when run on a computer, cause the computer to perform any of the methods described above.

[0130] This application also provides a chip including a processor and an interface circuit. The interface circuit is coupled to the processor. The processor is used to run computer programs or instructions to implement the above-described method. The interface circuit is used to communicate with other modules outside the chip.

[0131] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software programs, implementation can be, in whole or in part, in the form of a computer program product. This computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device containing one or more servers, data centers, etc., that can be integrated with the medium. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0132] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, the disclosure, and the appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple components. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0133] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely illustrative descriptions of the application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and variations.

Claims

1. A method for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules, characterized in that, include: Acquire temperature data, process parameters, and particle spectral data at the outlet of the second bed for each bed; The particle spectral data of the two beds outlet are input into the spectral model to obtain the predicted moisture content M2 of the two beds. The temperature data of each bed, process parameters and the predicted moisture content M2 of the two beds are integrated to construct a database. The temperature, process parameters, and predicted water content M2 of each bed are read in real time from the database and then preprocessed to generate an input vector set containing 10 variables. The input vector set is input into the principal component analysis model to perform PCA dimensionality reduction and extract the principal components, and the principal components are then input into the Hotling T-squared monitoring model. The Hotling T-square monitoring model constructs a T-square statistic based on the input main components, calculates control limits based on multiple batches of historical qualified data, compares the T-square statistic with the control limits to determine the drying uniformity level, and outputs the corresponding control strategy according to the drying uniformity level.

2. The method for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules according to claim 1, characterized in that, A database will be constructed by integrating temperature data from each bed, process parameters, and the predicted water content M2 of the second bed, including: The temperature data and process parameters of each bed are updated every 5 seconds. The temperature data of each bed includes the preheating temperature data T1 of the first bed preheating stage, the main drying temperature data T2 of the second bed main drying stage, and the shaping temperature data T3 of the third bed shaping stage. The process parameters include the air intake volume Q1 and air intake velocity V1 of the first bed, the air intake volume Q2 and air intake velocity V2 of the second bed, and the air intake volume Q3 and air intake velocity V3 of the third bed. The temperature data, process parameters, particle spectral data at the outlet of the second bed, and the predicted moisture content M2 of each bed are time-aligned using timestamp synchronization technology. A database is constructed by integrating the preheating temperature data T1, main drying temperature data T2, shaping temperature data T3, air inlet velocity V1 of bed 1, air inlet velocity V2 of bed 2, air inlet velocity V3 of bed 3, air inlet volume Q1 of bed 1, air inlet volume Q2 of bed 2, air inlet volume Q3 of bed 3, and predicted moisture content M2 of bed 2.

3. The method for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules according to claim 2, characterized in that, The preprocessed input vector set containing 10 variables includes: Based on multiple batches of historical qualified data, an abnormal threshold is calculated, and data exceeding the abnormal threshold in the following categories are removed: preheating temperature data T1, main drying temperature data T2, shaping temperature data T3, first bed air inlet velocity V1, second bed air inlet velocity V2, third bed air inlet velocity V3, first bed air inlet volume Q1, second bed air inlet volume Q2, and third bed air inlet volume Q3. The predicted water content M2 of the second bed was smoothed using a moving average method with a window size of 5 sets of continuous data; The preheating temperature data T1, main drying temperature data T2, shaping temperature data T3, first bed air inlet velocity V1, second bed air inlet velocity V2, third bed air inlet velocity V3, first bed air inlet volume Q1, second bed air inlet volume Q2, third bed air inlet volume Q3, and second bed predicted moisture content M2 are integrated to obtain an input vector set containing 10 variables.

4. The method for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules according to claim 3, characterized in that, The step of inputting the input vector set into the principal component analysis model to perform PCA dimensionality reduction and extract principal components includes: Using the covariance matrix formula Calculate each set of input vectors The covariance matrix C, where n is the total number of input vector sets. Let X be the mean vector of each set of input vectors. i Let i be the set of input vectors for the i-th group; Perform eigenvalue decomposition on the covariance matrix C to obtain eigenvalues ​​λ1>λ2>…>λ 10 >0 and the corresponding eigenvectors V1, V 1， V 10 ; pass Calculate the cumulative contribution rate LU of the first K principal components, select the smallest k value that satisfies the cumulative contribution rate LU≥90%, and k=3; pass The first three main components, PC1, PC2 and PC3, were extracted. PC1 represents the overall drying intensity, and the variables with the largest loading weight are T2, Q2 and M2. PC2 represents the inter-bed consistency, and the variables with the largest loading weight are V1, V2, V3, Q1, Q2 and Q3. PC3 represents the final product quality trend, and its loading weight reflects the correlation between M2 and T3 and Q3.

5. The method for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules according to claim 4, characterized in that, The Hotling T-square monitoring model constructs a T-square statistic based on the input major components and calculates control limits based on multiple batches of historical qualified data. The drying uniformity level is determined by comparing the T-square statistic with the control limits, including: Based on the first three principal components PC1, PC2, and PC3 and their corresponding eigenvalues ​​λ1, λ2, and λ3, through... Calculate the real-time T-squared statistic T 2 ; Based on multiple batches of historical qualified data, through Calculate the T-squared statistic T 2 upper limit of control , of which F α (3,m 3) is the critical value of the F distribution at a significance level of α=0.05, and m is the number of historical qualified data samples; T 2 With control upper limit By comparing multiple dimensions, the drying uniformity level is output.

6. The method for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules according to claim 5, characterized in that, The T 2 With control upper limit The drying uniformity level is output through multi-dimensional comparison and evaluation, including: When T 2 ≤ When the drying uniformity is determined to be qualified; when <T 2 ≤1.2× If the contribution rate deviation of any one of the principal components PC1, PC2, and PC3 is >5%, it is judged as a mild anomaly. When 1.2× <T 2 ≤1.5× If the contribution rate deviation of any two of the principal components PC1, PC2, and PC3 is >5%, it is judged as moderately abnormal. When T 2 >1.5× At that time, it was determined to be a severe abnormality.

7. The method for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules according to claim 6, characterized in that, The control strategy corresponding to the level of drying uniformity output includes: Based on the principal component analysis model, the abnormal principal component types PC1, PC2 or PC3 are identified, and by comparing with historical abnormal cases in the industrial database, the abnormal principal component types corresponding to different levels are determined. When a mild anomaly is identified, the associated anomalous principal component PC3 is invoked, the optimal parameter ratio in the historical database is invoked, the air inlet velocity V1 of one bed is adjusted within ±0.1 m / s, and the temperature T1 of one bed is adjusted within ±0.5℃ to reduce the temperature difference between beds to the acceptable range. When the condition is determined to be moderately abnormal, the associated principal component of the abnormality PC2 is invoked, the air intake Q2 of the second bed is adjusted within ±5m³ / h, and the guide vane of the second bed is opened and its angle is adjusted within ±3°. Based on the correlation between the predicted moisture content M2 of the second bed and the temperature of the second bed, the temperature T2 of the second bed is adjusted within ±1℃. When a severe anomaly is identified, the associated principal component of the anomaly, PC1, is invoked. The air intake volume Q2 of the second bed is adjusted within ±5 m³ / h. The guide vane of the second bed is opened and its angle is adjusted within ±3°. At the same time, based on the correlation between M2 and the temperature T2 of the second bed in the industrial database, T2 is adjusted within ±1°. During the regulation process, real-time monitoring data is read from the database every 5 seconds to assess the regulation effect until the T² statistic returns to the qualified range and the moisture deviation meets the standard.

8. The method for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules according to claim 7, characterized in that, The principal component analysis model identifies anomalous principal component types PC1, PC2, or PC3, and, by comparing with historical anomaly cases in the industrial database, determines the anomalous principal component types corresponding to different levels, including: If an abnormality is detected in the main component PC1, it is determined to be a deviation in the two-bed main drying parameters or a deviation in the material state. If an abnormality is detected in the principal component PC2, it is determined to be a mismatch between the three-bed shaping stage and the two-bed main drying stage. If an abnormality is detected in principal component PC3, it is determined that the synergy between the first and second beds is poor.

9. The method for online monitoring and improvement of the uniformity of fluidized bed drying of traditional Chinese medicine granules according to claim 8, characterized in that, Following the description of outputting the corresponding control strategy based on the level of drying uniformity, the following is also included: Based on historical batch data accumulated in the industrial database, after every N production batches are completed, the historical qualified data of those N batches are extracted, and the upper limit of the control of the Hotelling T-squared statistic is recalculated. Wherein N is the set period, and the updated control limit adaptation error is ≤1.5%; According to the updated control limits The threshold for determining the drying uniformity level is dynamically adjusted. For different Chinese herbal granule formulations, the database is used to match the historical optimal process parameter ratios to generate a baseline input vector set for that formulation.

10. An online monitoring and improvement device for the uniformity of fluidized bed drying of traditional Chinese medicine granules, characterized in that, The device includes: a communication unit and a processing unit; The communication unit is used to acquire temperature data of each bed, process parameters, and particle spectral data at the outlet of the second bed; The processing unit is used to input the particle spectral data of the second bed outlet into the spectral model, output the predicted moisture content M2 of the second bed, and integrate the temperature data of each bed, process parameters and the predicted moisture content M2 of the second bed to construct a database; The temperature, process parameters, and predicted water content M2 of each bed are read in real time from the database and then preprocessed to generate an input vector set containing 10 variables. The input vector set is input into the principal component analysis model to perform PCA dimensionality reduction and extract the principal components, and the principal components are then input into the Hotling T-squared monitoring model. The Hotling T-square monitoring model constructs a T-square statistic based on the main components and calculates control limits based on multiple batches of historical qualified data. It compares the T-square statistic with the control limits to determine the drying uniformity level and outputs the corresponding control strategy according to the drying uniformity level.

Images