Drug production monitoring system and method based on big data

Abstract

The application relates to the technical field of quality control of a medicament production process, and discloses a medicine production monitoring system and method based on big data; the system obtains reference state characteristic data, fluidity evaluation index data and electrostatic evaluation index data of powder medicaments in a hopper conveying stage, analyzes the fluidity recession degree of the powder medicaments under the influence of environmental moisture absorption and the electrostatic adsorption degree in the conveying process, outputs production process regulation suggestions based on the production risk under the joint action of the fluidity recession degree and the electrostatic adsorption degree, and realizes quantitative evaluation and accurate regulation of the production risk of the powder medicaments, thereby effectively improving the production process quality control level.

Description

Technical Field

[0001] This application relates to the field of quality control technology in pharmaceutical manufacturing processes, and in particular to a pharmaceutical manufacturing monitoring system and method based on big data. Background Technology

[0002] During the production process of powdered agents in the hopper conveying stage, phenomena such as sticking to the wall, arching, and poor material flow often occur, leading to a decrease in batch-to-batch quality consistency, inaccurate measurement, and reduced production efficiency, and in severe cases, even production interruption. At the same time, static electricity accumulation caused by friction between powder particles and the hopper wall surface can cause powder agglomeration and residue sticking to the wall, which not only affects the stability of material flow, but may also cause safety hazards such as dust explosions due to electrostatic discharge.

[0003] In existing technologies, risk assessments for these two types of problems often employ single-variable threshold alarm methods. For example, online monitoring of the mass flow rate of powdered agents during hopper transport can be used to determine whether the flowability is abnormal, or monitoring of the electrostatic potential value of powdered agents during hopper transport can be used to determine whether the electrostatic risk exceeds the safety limit. However, in actual production, flowability degradation is affected by the coupling of multiple factors such as the initial moisture content of the powder, the inlet moisture content, the median particle size, and the set feed rate. There may be complex interactions between different factors. Electrostatic adsorption is also closely related to multiple factors such as ambient humidity, ambient temperature, conveying speed, and powder particle size distribution. Existing single-variable threshold alarm methods can only reflect the abnormality of a single indicator and cannot capture the complex abnormal state under the combined effect of multiple factors. Therefore, it is difficult to quantify the degree of production risk, resulting in a high false alarm rate and a high missed alarm rate.

[0004] Furthermore, when anomalies occur, existing technologies typically only provide simple alarm prompts and lack targeted process control suggestions. This leads operators to rely on personal experience to handle the situation, making it difficult to guarantee the accuracy and consistency of control measures. Therefore, how to construct a production risk control system that can comprehensively consider the influence of multiple factors, quantitatively assess the risks of powder agent flowability decline and electrostatic adsorption, and output differentiated control suggestions based on risk levels has become an urgent technical problem to be solved in this field. Summary of the Invention

[0005] To overcome the above-mentioned drawbacks, this application provides a drug production monitoring system and method based on big data, aiming to solve the technical problems in the prior art, such as the difficulty in quantifying and assessing the risks of powder drug production, the difficulty in identifying anomalies caused by the coupling of multiple factors, and the lack of targeted and extensive control measures.

[0006] To achieve the above objectives, this application adopts the following technical solution:

[0007] Firstly, this application provides a drug production monitoring system based on big data, including:

[0008] The data acquisition module is used to acquire baseline state characteristic data, flowability assessment index data, and electrostatic assessment index data of powdered agents during the hopper conveying stage.

[0009] The rheological analysis module is used to correlate and process the baseline state characteristic data and flowability assessment index data of powdered agents to construct an assessment model for the degree of flowability degradation of powdered agents.

[0010] The electrostatic analysis module is used to jointly analyze the baseline state characteristic data and electrostatic assessment index data of powdered agents during the hopper conveying stage, and construct a risk identification model for electrostatic adsorption of powdered agents.

[0011] The risk diagnosis module is used to collect baseline state characteristic data, flowability assessment index data and electrostatic assessment index data of the powder agent to be tested in real time during the hopper conveying stage. Combined with the powder agent flowability decay assessment model and the powder agent electrostatic adsorption risk identification model, the production risk of the powder agent to be tested is analyzed.

[0012] The control output module is used to output control suggestions for the production process of the powder reagent to be tested based on the production risks of the powder reagent to be tested.

[0013] According to the above technical solution, the steps for obtaining the baseline state characteristic data, flowability assessment index data, and electrostatic assessment index data of powdered agents during the hopper conveying stage include:

[0014] Step S11: Obtain the baseline state characteristic data of the powder agent: Obtain the initial moisture value of each batch of powder agent measured before feeding from the production records of multiple batches of the same powder agent; Collect the inlet moisture value of multiple batches of the same powder agent during the feeding process from the near-infrared moisture meter installed on the conveying pipe at the hopper inlet.

[0015] Step S12: Obtain flowability assessment index data of powdered agents: Collect the median particle size value of multiple batches of the same powdered agent during the feeding process from the online particle size analyzer installed on the conveying pipe at the hopper inlet; obtain the feeding speed set value of each batch of powdered agent from the hopper production process parameter records of multiple batches of the same powdered agent; collect the mass flow time series data of each batch of powdered agent during the feeding process from the flow meter installed on the conveying pipe at the hopper inlet; divide the inner wall of the hopper into multiple monitoring areas according to the principle of equal area; before feeding multiple batches of the same powdered agent, use a surface roughness meter to measure the surface roughness of the inner wall of the hopper in each monitoring area, and calculate its arithmetic mean as the average surface roughness value of the hopper corresponding to each batch of powdered agent;

[0016] Step S13: Obtain electrostatic evaluation index data of powdered agents: Collect d10 and d90 particle size values ​​of multiple batches of the same powdered agent during the feeding process using an online particle size analyzer installed on the conveying pipe at the hopper inlet; collect inlet flow velocity values ​​of multiple batches of the same powdered agent during the feeding process using a flow velocity sensor installed on the conveying pipe at the hopper inlet; divide the inner wall of the hopper into multiple monitoring areas according to the principle of equal area, and arrange a non-contact electrometer in each area. During the feeding process of multiple batches of the same powdered agent, collect the wall potential values ​​of all monitoring points, and calculate their arithmetic mean as the average wall potential value of the hopper corresponding to each batch of powdered agent; according to the same equal area division, before feeding multiple batches of the same powdered agent, use a surface resistance tester to measure the surface resistivity of the inner wall of the hopper in each monitoring area, and calculate its arithmetic mean as the average surface resistivity value of the hopper corresponding to each batch of powdered agent.

[0017] According to the above technical solution, the steps for constructing an assessment model for the degree of flowability degradation of powdered pharmaceuticals include: (1) Correlation processing of the baseline state characteristic data and flowability assessment index data of the powdered pharmaceuticals.

[0018] Step S21: Extract the mass flow time series data of each batch of powdered agent during the feeding process from the baseline state characteristic data and flowability assessment index data of the powdered agent. By analyzing the fluctuation of mass flow, identify and extract the initial moisture value, inlet moisture value, median particle size value, feeding speed set value and the average surface roughness value of the corresponding hopper for the powdered agent with normal flow rate. At the same time, normalize these five data values ​​for each batch of powdered agent with normal flow rate to generate the training dataset of the powdered agent flowability decay assessment model.

[0019] Step S22: Based on the training dataset of the powder agent flowability degradation assessment model, construct a subsample set for each isolated tree. On each subsample set, recursively construct isolated trees by randomly selecting segmentation features and randomly generating segmentation values. Combine the generated isolated trees to form an isolated forest, thereby completing the powder agent flowability degradation assessment model.

[0020] Based on the above technical solution, the steps for constructing a risk identification model for electrostatic adsorption of powdered agents by jointly analyzing the baseline state characteristic data and electrostatic assessment index data of powdered agents during the hopper conveying stage include:

[0021] Step S31: Extract the average wall potential value of the corresponding hopper during the feeding process of each batch of powdered agent from the baseline state characteristic data and electrostatic evaluation index data of the powdered agent. In this way, identify and extract the initial moisture value, inlet moisture value, d10 particle size value, d90 particle size value, inlet flow rate value and average surface resistivity value of the powdered agent in the electrostatic normal batch. At the same time, normalize these six data values ​​of each powdered agent in the electrostatic normal batch to generate the training dataset of the powdered agent electrostatic adsorption risk identification model.

[0022] Step S32: Based on the training dataset of the powder agent electrostatic adsorption risk identification model, the single-class support vector machine algorithm is used to fit the electrostatic normal samples in the kernel space. The support vectors and their corresponding weight coefficients and decision thresholds are determined by solving the quadratic programming problem, thereby completing the powder agent electrostatic adsorption risk identification model.

[0023] According to the above technical solution, the steps for analyzing the production risks of the powdered agent under test by real-time acquisition of baseline state characteristic data, flowability assessment index data, and electrostatic assessment index data during the hopper conveying stage, combined with the powdered agent flowability degradation assessment model and the powdered agent electrostatic adsorption risk identification model, include:

[0024] Step S41: Read the real-time production data of the current batch of powder reagent to be tested during the hopper conveying stage from the production process control system, and extract the 5-dimensional feature vector required for assessing the degree of flowability degradation, denoted as... ,in This is the initial moisture value. This is the inlet moisture value. This is the median particle size value. Set the feeding speed value. This represents the average surface roughness value of the corresponding hopper; simultaneously, the 6-dimensional feature vector required for electrostatic adsorption risk identification is extracted, denoted as... ,in This is the initial moisture value. This is the inlet moisture value. The d10 particle size value The d90 particle size value This is the inlet flow rate value. This represents the average surface resistivity value of the corresponding hopper;

[0025] Step S42: Transfer the liquidity feature vector Input the powder agent flowability degradation assessment model to obtain the flowability degradation value of the current batch of powder agent to be tested; and input the electrostatic feature vector. Input the electrostatic adsorption risk identification model for powdered reagents to obtain the electrostatic adsorption risk value of the current batch of powdered reagents to be tested;

[0026] Step S43: Based on the fluidity degradation value and electrostatic adsorption risk value of the current batch of powder reagent to be tested, the entropy weight method is used to determine the weight coefficients of the fluidity degradation value and electrostatic adsorption risk value, and then a weighted sum is performed to calculate the comprehensive production risk value of the batch of powder reagent to be tested.

[0027] Based on the above technical solution, and considering the production risks of the powder reagent to be tested, the steps for providing recommendations on the production process control of the powder reagent to be tested include:

[0028] Step S51: Based on the distribution of the comprehensive production risk value of all normal batches of powdered reagents, set a risk level classification threshold, compare the comprehensive production risk value of the batch of powdered reagents to be tested with the risk level classification threshold, and determine the production risk level of the batch of powdered reagents to be tested.

[0029] Step S52: Based on the production risk level of the batch of powder reagent to be tested, generate differentiated production process control suggestions for the powder reagent to be tested, specifically including setting corresponding control operations, monitoring frequencies and handling measures for different production risk levels.

[0030] Secondly, this application also provides a drug production monitoring method based on big data, including the following steps:

[0031] S1. Obtain baseline state characteristic data, flowability assessment index data, and electrostatic assessment index data of powdered agents during the hopper conveying stage.

[0032] S2. Correlate the baseline state characteristic data and flowability assessment index data of powdered agents to construct an assessment model for the degree of flowability degradation of powdered agents.

[0033] S3. Combine the baseline state characteristic data and electrostatic assessment index data of powdered agents during the hopper conveying stage to construct a risk identification model for electrostatic adsorption of powdered agents.

[0034] S4. Real-time acquisition of baseline state characteristic data, flowability assessment index data, and electrostatic assessment index data of the powder agent to be tested during the hopper conveying stage. Combined with the powder agent flowability decay assessment model and the powder agent electrostatic adsorption risk identification model, the production risk of the powder agent to be tested is analyzed.

[0035] S5. Based on the production risks of the powder reagent to be tested, provide suggestions for the control of the production process of the powder reagent to be tested.

[0036] Thirdly, this application provides an electronic device including a processor and a memory, wherein the memory stores a computer program that can be called by the processor, and the processor executes a big data-based drug production monitoring method by calling the computer program stored in the memory.

[0037] Fourthly, this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform a big data-based drug production monitoring method.

[0038] Compared with the prior art, this application has the following advantages and beneficial effects:

[0039] This application constructs a liquidity degradation assessment model based on normal batch data and an electrostatic adsorption risk identification model, which enables accurate identification and quantitative assessment of multi-factor coupling anomalies in the production process of powdered reagents. Based on this, an entropy weight method is used to objectively weight and fuse the data to obtain a comprehensive production risk value. Based on the statistical distribution of normal batches, a risk threshold is set to output differentiated control suggestions, forming a closed-loop management system from risk identification and quantitative assessment to precise control. This effectively solves the problems of difficulty in quantifying risks, difficulty in identifying anomalies, and coarse control measures in existing technologies, and significantly improves the quality control level of the powdered reagent production process. Attached Figure Description

[0040] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0041] Figure 1 This is an overall flowchart of the big data-based drug production monitoring system provided in the embodiments of this application;

[0042] Figure 2 This is a data acquisition flowchart provided in an embodiment of this application;

[0043] Figure 3 This is a flowchart illustrating the construction of a powder agent flowability degradation assessment model provided in this application embodiment;

[0044] Figure 4 This is a flowchart illustrating the construction of a risk identification model for electrostatic adsorption of powdered pharmaceuticals provided in this application embodiment;

[0045] Figure 5 This is a flowchart of risk analysis for powder pharmaceutical production provided in the embodiments of this application;

[0046] Figure 6 This is a flowchart showing the output of the production process control suggestions for the powdered reagent to be tested provided in the embodiments of this application. Detailed Implementation

[0047] The technical solution of this application will be further described in detail below with reference to the accompanying drawings and specific embodiments, so as to facilitate understanding and implementation by those skilled in the art. It should be understood that the specific embodiments described herein are only for explaining this application and are not intended to limit this application.

[0048] Please see Figure 1 , Figure 1 This is an overall flowchart of the big data-based drug production monitoring system provided in this application embodiment, which specifically includes the following modules:

[0049] The data acquisition module is used to acquire baseline state characteristic data, flowability assessment index data, and electrostatic assessment index data of powdered agents during the hopper conveying stage.

[0050] Please see Figure 2 , Figure 2 The complete technical process for data acquisition in the embodiments of this application is illustrated, and the specific steps are as follows:

[0051] Step S11: Obtain the baseline state characteristic data of the powder agent: Obtain the initial moisture value of each batch of powder agent measured before feeding from the production records of multiple batches of the same powder agent; Collect the inlet moisture value of multiple batches of the same powder agent during the feeding process from the near-infrared moisture meter installed on the conveying pipe at the hopper inlet.

[0052] Step S12: Obtain flowability assessment index data of powdered agents: Collect the median particle size value of multiple batches of the same powdered agent during the feeding process from the online particle size analyzer installed on the conveying pipe at the hopper inlet; obtain the feeding speed set value of each batch of powdered agent from the hopper production process parameter records of multiple batches of the same powdered agent; collect the mass flow time series data of each batch of powdered agent during the feeding process from the flow meter installed on the conveying pipe at the hopper inlet; divide the inner wall of the hopper into multiple monitoring areas according to the principle of equal area; before feeding multiple batches of the same powdered agent, use a surface roughness meter to measure the surface roughness of the inner wall of the hopper in each monitoring area, and calculate its arithmetic mean as the average surface roughness value of the hopper corresponding to each batch of powdered agent;

[0053] Step S13: Obtain electrostatic evaluation index data of powdered agents: Collect d10 and d90 particle size values ​​of multiple batches of the same powdered agent during the feeding process using an online particle size analyzer installed on the conveying pipe at the hopper inlet; collect inlet flow velocity values ​​of multiple batches of the same powdered agent during the feeding process using a flow velocity sensor installed on the conveying pipe at the hopper inlet; divide the inner wall of the hopper into multiple monitoring areas according to the principle of equal area, and arrange a non-contact electrometer in each area. During the feeding process of multiple batches of the same powdered agent, collect the wall potential values ​​of all monitoring points and calculate their arithmetic mean as the average wall potential value of the hopper corresponding to each batch of powdered agent; according to the same equal area division, before feeding multiple batches of the same powdered agent, use a surface resistance tester to measure the surface resistivity of the inner wall of the hopper in each monitoring area, and calculate its arithmetic mean as the average surface resistivity value of the hopper corresponding to each batch of powdered agent.

[0054] The data acquisition process in this embodiment collects baseline state characteristic data, flowability assessment index data, and electrostatic assessment index data of powdered agents during the hopper conveying stage. This lays a comprehensive and accurate data foundation for subsequent multi-dimensional risk assessment and provides reliable data support for building an industrially feasible risk assessment solution.

[0055] The rheological analysis module is used to correlate and process the baseline state characteristic data and flowability assessment index data of powdered agents to construct an assessment model for the degree of flowability degradation of powdered agents.

[0056] Please see Figure 3 , Figure 3 This is a flowchart illustrating the construction of a powder agent flowability degradation assessment model provided in this application embodiment. The specific steps are as follows:

[0057] Step S21: Extract the time-series mass flow rate data of each batch of powdered agent during the feeding process from the baseline state characteristic data and flowability assessment index data of the powdered agent. By analyzing the fluctuation of mass flow rate, identify and extract the initial moisture value, inlet moisture value, median particle size value, feed rate set value, and average surface roughness value of the corresponding hopper for the powdered agent with normal flow rate. At the same time, normalize these five data values ​​for each batch of powdered agent with normal flow rate to generate the training dataset for the powdered agent flowability degradation assessment model; the specific steps are as follows:

[0058] First, from the production records of multiple batches of the same powder agent, batches that did not trigger any flowability alarms and did not involve manual tapping of the hopper were selected as the initial normal flowability batches.

[0059] Secondly, from the flowability assessment index data of powder agents, the time series data of mass flow rate during the feeding process of each initially selected normal flow batch are obtained. The standard deviation of mass flow rate is divided by the average mass flow rate to obtain the mass flow rate variation coefficient of each initially selected normal flow batch. This value reflects the degree of fluctuation of mass flow rate during the feeding process of the batch. The smaller the variation coefficient, the more stable the flowability.

[0060] Next, considering that the mass flow rate of a batch with normal flowability should be stable and continuous during the material feeding process, and its fluctuation should be controlled within a certain range, the mass flow rate variation coefficients of all the initially selected batches with normal flowability are arranged from smallest to largest, and the 30th quantile is taken as the screening threshold. The initially selected batches with mass flow rate variation coefficients less than or equal to this threshold are defined as the final batches with normal flowability.

[0061] Then, for each sample marked as a normal flow batch, the corresponding initial moisture value, inlet moisture value, median particle size value, feed rate set value, and average surface roughness value of the corresponding hopper are extracted; the original data of all normal batches are organized into an initial feature matrix of shape (N, 5), where N is the total number of normal batches and 5 is the number of feature dimensions.

[0062] Then, the initial feature matrix of all normal batches is subjected to min-max normalization, that is, the feature value of each dimension is subtracted from the minimum value of the feature in all normal batches, and then divided by the difference between the maximum and minimum values ​​of the feature in all normal batches.

[0063] Finally, all the normalized data are combined to obtain a feature matrix of shape (N, 5), which is the training dataset used to train the powder agent flowability degradation assessment model.

[0064] Step S22: Based on the training dataset of the powder agent flowability degradation assessment model, construct a subsample set for each isolated tree. On each subsample set, recursively construct isolated trees by randomly selecting segmentation features and randomly generating segmentation values. Combine the generated isolated trees to form an isolated forest, thus completing the powder agent flowability degradation assessment model. The specific steps are as follows:

[0065] First, given that the phenomenon of fluidity degradation in powdered pharmaceuticals is characterized by low frequency, scarcity of degradation samples, and difficulty in pre-labeling, the isolated forest algorithm, as an unsupervised anomaly detection method, has the advantages of not relying on degradation samples for training, building models based solely on normal batch data, and effectively identifying outliers that deviate from the normal distribution by randomly segmenting the feature space. Thus, it can accurately identify whether fluidity degradation has occurred in the current production batch. Therefore, the isolated forest algorithm is used to construct a model for assessing the degree of fluidity degradation in powdered pharmaceuticals.

[0066] Secondly, determine the basic parameter configuration of the isolated forest model: the size of the subsampling set. Following the recommended value in the paper on the Isolation Forest algorithm, the value is set to 256, which strikes a good balance between training efficiency and anomaly detection performance. If the total number of powder agents in normal flowability batches is less than 256, the size of the subsampling set is directly set to the total number of powder agents in normal flowability batches, i.e., all samples are used for subsampling. The number of isolation trees is set to 100 to ensure that the model has sufficient ensemble size to stabilize anomaly scores, while avoiding the waste of computational resources due to an excessive number of trees. Therefore, the maximum height of each isolation tree is limited. Based on the principles of the Isolation Forest algorithm, it is set as follows: ,when hour, That is, the growth depth of each isolated tree does not exceed 8 layers;

[0067] Then, randomly sample from the training dataset of the powder agent flowability degradation assessment model. Each sample can only be drawn once. Each sample constitutes a training subset for the first isolated tree; the above process is repeated to construct the respective subsets for the second to the 100th isolated trees in order to ensure diversity between trees;

[0068] Next, for each isolated tree and its corresponding... A subset of samples, starting from the root node (which contains all samples), is a subset of samples. The recursive partitioning begins with the initial data set of 100 samples.

[0069] A1. Randomly select one feature from the five feature dimensions as the segmentation feature, denoted as feature A1. These five feature dimensions correspond to the initial moisture value, inlet moisture value, median particle size value, feed rate setting value, and average surface roughness value required for assessing the degree of flowability degradation of powder agents; simultaneously, the feature dimensions of all samples in the current node are obtained. minimum value on and maximum value and in the interval A random segment value is generated within. ;

[0070] A2. Next, all nodes in the current node that satisfy the condition... The samples are divided into left child nodes, and all those that satisfy the condition are divided into left child nodes. The samples are divided into right child nodes;

[0071] A3. Repeat the above A1 and A2 processes for the left and right child nodes respectively, and continue to divide downwards until any one of the following three conditions is met: the current node reaches the 8th layer depth, the current node contains only one sample, and all samples in the current node have the same value on all 5 features. Then the training of the current isolated tree is completed.

[0072] Taking the construction process of an isolated tree as an example, the sub-sampling set of this isolated tree contains 5 samples of powdered reagents from normal flow batches, whose average surface roughness values ​​are respectively During the first segmentation at the root node, the average surface roughness value is randomly selected as the segmentation feature, and the minimum value of this feature is... The maximum value is In the interval A random segment value is generated within. ; The average surface roughness value is less than Three samples To be assigned to the left child node, the node will be greater than or equal to Two samples The sample is assigned to the right child node. During the second segmentation in the left child node, an initial moisture value is randomly selected as the segmentation feature. The initial moisture values ​​of these three samples are 2.3%, 2.5%, and 2.4%, with a minimum of 2.3% and a maximum of 2.5%. A random segmentation value p = 2.45% is generated. The two samples with initial moisture values ​​less than 2.45% (2.3% and 2.4%) are assigned to the left grandchild node, and the sample with an initial moisture value greater than or equal to 2.45% (2.5%) is assigned to the right grandchild node. At this point, the left grandchild node contains two samples, and the segmentation continues downwards. The right grandchild node contains only one sample, reaching the termination condition, and the segmentation stops. This process is repeated recursively until all branches meet the termination condition, completing the training of the isolated tree.

[0073] Repeat steps A1, A2, and A3 above to complete the training of the 1st to 100th isolated trees in sequence;

[0074] Finally, the 100 isolated trees obtained from the training, along with the minimum and maximum values ​​of each feature in the training dataset of the powder agent flowability degradation assessment model, are saved together to complete the powder agent flowability degradation assessment model; for any batch of powder agent to be tested, its corresponding 5-dimensional feature vector is extracted. First, the minimum and maximum values ​​of each feature are stored and subjected to min-max normalization to obtain the normalized feature vector. Then Given each isolated tree, recursively traverse downwards from the root node based on the segmentation features and segmentation values ​​of each node until a leaf node is reached, recording its path length on the k-th isolated tree. Calculate the average path length of this sample across all 100 isolated trees. ;

[0075] After obtaining the average path length, it needs to be converted into a comparable anomaly score. Since the scale of path length differs with different subset sizes, directly using the average path length cannot uniformly measure the degree of anomaly. Therefore, a normalization factor needs to be introduced. Perform the conversion; The calculation requires first finding the harmonic number. = Then substitute into the formula , so when When, substituting into the formula, we can obtain ;

[0076] Average path length With normalization factor Substituting the anomaly score formula from the Isolation Forest algorithm, we calculate the degree of flowability degradation of the batch of powdered reagent to be tested:

[0077]

[0078] In the formula, The value represents the degree of flowability degradation of the batch of powdered reagent to be tested, characterizing the degree of deviation of the flowability of this batch of powdered reagent from that of a batch with normal flowability, and the value ranges from (0,1]. The closer the value is to 1, the shorter the average path length required for a sample to be isolated on the isolation tree, meaning the sample is more easily isolated. This indicates a greater deviation of the powder reagent batch from its normal distribution in terms of flowability and a greater degree of flowability degradation. The closer it is to 0, the longer the average path length required for a sample to be isolated on the isolation tree, meaning the less likely the sample is to be isolated. This indicates that the flowability of the powder agent batch is closer to the normal distribution and the degree of flowability degradation is lower.

[0079] The process of constructing the powder agent flowability degradation assessment model in this embodiment provides a unified and comparable benchmark for the subsequent quantitative assessment of flowability degradation, thereby enabling early warning and accurate identification of powder agent flowability degradation and effectively improving the quality control level of the production process.

[0080] The electrostatic analysis module is used to jointly analyze the baseline state characteristic data and electrostatic assessment index data of powdered agents during the hopper conveying stage, and construct a risk identification model for electrostatic adsorption of powdered agents.

[0081] Please see Figure 4 , Figure 4This is a flowchart illustrating the construction of a risk identification model for electrostatic adsorption of powdered pharmaceuticals provided in this application embodiment. The specific steps are as follows:

[0082] Step S31: Extract the average wall potential value of the corresponding hopper during the feeding process of each batch of powdered agent from the baseline state characteristic data and electrostatic assessment index data of the powdered agent. This identifies and extracts the initial moisture value, inlet moisture value, d10 particle size value, d90 particle size value, inlet flow rate value, and average surface resistivity value of the electrostatically normal batch of powdered agent. Simultaneously, normalize these six data values ​​for each electrostatically normal batch of powdered agent to generate the training dataset for the electrostatic adsorption risk identification model of the powdered agent. The specific steps are as follows:

[0083] First, from the production records of multiple batches of the same powder agent, batches that did not trigger any electrostatic adsorption alarms and did not have any abnormal shutdown interventions during the production process were selected as the initial electrostatic normal batches.

[0084] Secondly, considering that even without triggering an alarm, the wall potential values ​​of some batches may already be close to the critical level of electrostatic adsorption, including these critically state batches in the training set would lead to an overly broad definition of normal electrostatics in the model, reducing its sensitivity to early electrostatic risks. Therefore, it is necessary to further screen the batches with lower wall potential values ​​and the slightest degree of electrostatic accumulation from the initially selected batches with normal electrostatics as the final batches with normal electrostatics. Specifically, the average wall potential value of all the initially selected batches with normal electrostatics is extracted from the electrostatic assessment index data of powdered agents and arranged in ascending order. The corresponding 20th quantile is used as the threshold for judging batches with normal electrostatics. The batches with average wall potential values ​​lower than this threshold are further screened from the initially selected batches with normal electrostatics and defined as the final batches with normal electrostatics.

[0085] Then, for each powder reagent sample marked as an electrostatically normal batch, the corresponding initial moisture value, inlet moisture value, d10 particle size value, d90 particle size value, inlet flow rate value, and average surface resistivity value of the corresponding hopper are extracted from the baseline state characteristic data and electrostatic evaluation index data of the powder reagent. The original data of all electrostatically normal batch powder reagent samples are organized into an initial feature matrix of shape (M,6).

[0086] Finally, the initial feature matrix of all electrostatic normal batches is subjected to min-max normalization, that is, the feature value of each dimension is subtracted from the minimum value of the feature in all electrostatic normal batches, and then divided by the difference between the maximum and minimum values ​​of the feature in all electrostatic normal batches; all normalized data are combined to obtain a feature matrix of shape (M,6), which is the training dataset used to train the powder agent electrostatic adsorption risk identification model.

[0087] Step S32: Based on the training dataset of the powder agent electrostatic adsorption risk identification model, a single-class support vector machine algorithm is used to fit the electrostatic normal samples in the kernel space. The support vectors and their corresponding weight coefficients and decision thresholds are determined by solving a quadratic programming problem, thereby completing the powder agent electrostatic adsorption risk identification model; the specific steps are as follows:

[0088] First, because the electrostatic adsorption risk of powdered agents is affected by multiple factors such as initial moisture value, inlet moisture value, d10 particle size value, d90 particle size value, inlet flow rate value, and average surface resistivity value, and there may be complex nonlinear coupling relationships between these factors, and because single-class support vector machine, as an unsupervised anomaly detection method, has the characteristics of learning decision boundaries based only on electrostatically normal samples, effectively handling nonlinear relationships between features through kernel function mapping, and using the distance from the output sample to the hyperplane as a risk quantification indicator, it can accurately quantify the degree of deviation between the current production batch and the electrostatically normal benchmark. Therefore, the single-class support vector machine algorithm is used to construct the electrostatic adsorption risk identification model for powdered agents.

[0089] Secondly, the kernel function and related parameter configuration of the single-class support vector machine model are determined: Given that the radial basis function can map the original features to a high-dimensional space to handle nonlinear relationships, and has the characteristics of few parameters and strong adaptability, the radial basis function is selected as the kernel function of the model in this embodiment; the kernel function width parameter σ is set to the reciprocal of the feature dimension according to the empirical rule, that is, σ=1 / 6≈0.1667; the model hyperparameter μ is used to control the upper limit of the proportion of training samples judged as abnormal, and is set to μ=0.05, that is, a maximum of 5% of the normal training samples are allowed to fall outside the decision boundary, so as to improve the generalization ability of the model;

[0090] Then, from the training dataset of the electrostatic adsorption risk identification model for powdered agents, a normalized feature matrix of shape (M,6) is extracted as input. This matrix contains six feature dimensions: initial moisture content, inlet moisture content, d10 particle size, d90 particle size, inlet flow rate, and average surface resistivity of M normal batches of electrostatically adsorbed powdered agents. A single-class support vector machine algorithm is used for model training. The core of the training is solving a quadratic programming problem with the objective function in the form of:

[0091]

[0092] In the formula, and The first The and the first The Lagrange multipliers corresponding to each electrostatic normal batch of powder reagent sample are used to measure the contribution weight of each sample to the electrostatic adsorption risk decision boundary. Here is the radial basis function kernel, used to calculate the first... The and the first Similarity of electrostatic normal batch powder reagent samples in high-dimensional feature space; This represents the total number of powder reagent samples from normal batches with static electricity. ; This means finding the Lagrange multipliers that minimize the objective function using an optimization algorithm. The value of ;

[0093] The objective function in this embodiment minimizes the weighted similarity sum among all pairs of electrostatically normal batch powder reagent samples, ensuring that the decision boundary for electrostatic adsorption risk most compactly encompasses the distribution region of the electrostatically normal batch powder reagent samples. The constant coefficients are used to simplify subsequent differentiation calculations; double summation symbols. Then, iterate through all pairs of electrostatic normal batch powder reagent samples, calculate the weighted similarity of each pair of samples, and sum them up. The product form ensures that only electrostatic normal batch powder reagent sample pairs with relatively large weights will have a significant impact on the objective function;

[0094] The constraints are:

[0095]

[0096] In the formula, The upper limit parameter for the pre-set anomaly ratio is set to 0.05, which is used to control the model's tolerance for samples with electrostatic adsorption risk.

[0097] First constraint This is to limit the maximum influence weight of each electrostatically normal batch of powder reagent sample on the electrostatic adsorption risk decision boundary, and to prevent a single abnormal powder reagent sample from excessively dominating the position of the decision boundary. The lower limit of 0 indicates that electrostatically normal batches of powder reagent samples can completely exclude the construction of the decision boundary, and the upper limit... It ensures that at least one Each electrostatically normal batch of powder reagent samples can obtain a non-zero weight; the second constraint. This is to ensure that the sum of all Lagrange multipliers is 1, so that the Lagrange multipliers form a probability distribution, which facilitates the calculation and interpretation of subsequent decision thresholds;

[0098] Radial basis kernel function The calculation formula is:

[0099]

[0100] In the formula, The kernel function width parameter, with a value of 0.1667, is used to control the influence range of a single normal batch of powder reagent sample in electrostatic adsorption risk identification. For the first The and the first The Euclidean distance between two normal batches of powder reagent samples in the original feature space, composed of six characteristic dimensions: initial moisture content, inlet moisture content, d10 particle size, d90 particle size, inlet flow rate, and average surface resistivity, is used to measure the similarity of electrostatic adsorption characteristics between them. The closer the moisture content, particle size distribution, conveying flow rate, and average surface resistivity of the two normal batches of powder reagent samples, the smaller the Euclidean distance, indicating that their electrostatic adsorption characteristics are more similar; conversely, the larger the Euclidean distance, the greater the difference in their electrostatic adsorption characteristics. It is a natural exponential function;

[0101] In this embodiment, the radial basis kernel function maps electrostatic normal batch powder reagent samples in the original feature space to a high-dimensional feature space, and calculates the similarity between samples in the high-dimensional space. When the electrostatic adsorption characteristics of two electrostatic normal batch powder reagent samples are highly similar, for example, they have similar initial moisture value, inlet moisture value, d10 particle size value, d90 particle size value, inlet flow rate value, and average surface resistivity value, Approaching 0, A kernel function value close to 1 indicates that the two samples are highly similar; however, when the electrostatic adsorption properties of the two samples differ significantly, for example, one sample has a significantly higher water content than the other, Then it is relatively large. A kernel function value close to 0 indicates that the similarity between the two samples is very low.

[0102] Next, the sequential minimum optimization algorithm is used to solve the above quadratic programming problem: First, all Lagrange multipliers are initialized; then, sequential minimum optimization is performed iteratively, checking all Lagrange multipliers according to the Karush-Kuhn-Tucker optimization conditions, and selecting the pair of multipliers with the most serious violation of the conditions as the optimization objective of this iteration; all other multipliers except the selected pair are fixed, simplifying the original complex global optimization problem into a quadratic function extremum problem with only these two variables, calculating the optimal update amount of these two multipliers under the current constraints through analytical derivation, and modifying the values ​​of these two multipliers; the above iterative process is repeated until all Lagrange multipliers satisfy the Karush-Kuhn-Tucker conditions, at which point the algorithm converges, and the optimal Lagrange multiplier vector is obtained. ;

[0103] The support vectors are determined based on the Lagrange multiplier vectors obtained from the solution: For The electrostatically normal batch powder reagent samples are the support vectors. These samples lie on and inside the decision boundary of electrostatic adsorption risk, collectively defining the distribution region of electrostatically normal batch powder reagents in terms of electrostatic adsorption characteristics; among them, those satisfying... The samples are located on the decision boundary. Although these electrostatically normal batches of powder reagent samples still have various characteristics within the normal electrostatic range, they are in a critical state where electrostatic accumulation has not yet triggered adsorption risk. Therefore, they directly participate in the construction of the electrostatic adsorption risk decision boundary; satisfying The samples are located inside the decision boundary. These electrostatically normal batch powder reagent samples have relatively stable electrostatic adsorption characteristics, are far from the risk boundary, and contribute little to the formation of the decision boundary.

[0104] Then, based on the obtained Lagrange multiplier vectors and support vectors, the decision threshold for electrostatic adsorption risk is calculated. :

[0105]

[0106] In the formula, For the first Lagrange multipliers of a normal batch of electrostatic powder reagent samples; For the first A normal batch of electrostatic powder reagent samples and the selected boundary support vector Radial basis kernel function values ​​between; For any support vector located on the decision boundary, i.e., satisfying Electrostatic normal batch powder reagent samples;

[0107] The decision threshold in this embodiment Essentially, this is a mathematical expression of the boundary position of the electrostatic safety region for a normal batch of electrostatically controlled powder agents in a six-dimensional feature space. For the normal batch of electrostatically controlled powder agents corresponding to the support vectors located on the decision boundary, although their six characteristics—initial moisture content, inlet moisture content, d10 particle size, d90 particle size, inlet flow rate, and average surface resistivity—fluctuate to some extent, these characteristic combinations are precisely at the critical state where electrostatic accumulation has not yet triggered adsorption risk. Therefore, by calculating the weighted similarity between all normal batches of electrostatically controlled powder agents and the samples at this critical state, the following is obtained: The value serves as the baseline for determining whether a new batch of powdered reagents has entered the electrostatic adsorption risk zone.

[0108] Meanwhile, based on the mathematical principles of single-class support vector machines, for any support vector located on the decision boundary, substituting it into the above formula yields the calculated result. The values ​​are all the same, which means that regardless of which critical state batch of powder reagent is selected as the reference, the determined electrostatic safety zone boundary is unique and definite; therefore, after the model is built, This will be determined as an inherent parameter of the electrostatic adsorption risk identification model. When predicting the electrostatic adsorption risk of a new batch of powdered reagents, it is only necessary to calculate the weighted similarity between this batch and all support vectors and... By comparison, it can be determined whether the electrostatic adsorption risk of the new batch of powdered reagents exceeds the normal fluctuation range of electrostatics;

[0109] Taking a model training process as an example, this training collected 60 normal batch samples, i.e., M=60, and after normalization, a 60×6 feature matrix was obtained; using The parameters were used for training, and the Lagrange multipliers corresponding to 60 samples were obtained by solving the sequence minimum optimization algorithm. Among them, there are 9 samples. The values ​​were 0.35, 0.30, 0.28, 0.25, 0.22, 0.20, 0.18, 0.15, and 0.12, respectively, all greater than 0, and were thus identified as support vectors. Among these 9 support vectors, 5 samples satisfied... Located on the decision boundary, these samples correspond to powder reagent batches that are in a critical state of electrostatic accumulation; the remaining four samples... The vector lies inside the decision boundary; one of the five support vectors located on the decision boundary is randomly selected, for example, one... that sample Calculate the kernel function value of the sample and all 60 training samples, and substitute it into the formula to obtain the decision threshold. This threshold is the benchmark value for judging whether the electrostatic adsorption risk of a new batch of powdered medicine is abnormal.

[0110] Finally, the support vectors obtained from the above training, along with their corresponding Lagrange multipliers, decision thresholds, and the minimum and maximum values ​​of each feature in the training dataset of the powder agent electrostatic adsorption risk identification model, are saved together to complete the powder agent electrostatic adsorption risk identification model; for any batch of powder agent, the corresponding 6-dimensional feature vector... First, the minimum and maximum values ​​of each feature are stored and subjected to min-max normalization to obtain the normalized feature vector. Then, the decision function value of the batch of powder reagent to be tested is calculated by substituting it into the decision function:

[0111] ;

[0112] In the formula, The decision function value is the batch of powder reagent to be tested, which characterizes the position of the electrostatic adsorption state of the batch of powder reagent in the kernel space relative to the decision boundary; The number of support vectors; For the first Support vectors; For the first Lagrange multipliers corresponding to each support vector; For the first Each support vector and input sample Radial basis kernel function values ​​between; This is the decision threshold;

[0113] To convert the decision function value into an intuitive risk quantification indicator, the Sigmoid function is used to map it to the (0,1) interval, thus obtaining the electrostatic adsorption risk value of the batch of powder reagent to be tested:

[0114]

[0115] In the formula, The electrostatic adsorption risk value for the batch of powdered reagent to be tested ranges from (0,1), representing the degree of deviation of the electrostatic adsorption state of this batch of powdered reagent compared to batches with normal electrostatic properties, and its value range is the set of real numbers; when The closer the value is to 1, the greater the deviation of the electrostatic adsorption state of the batch of powdered reagent being tested from the normal electrostatic distribution, and the greater the risk of electrostatic adsorption; when The closer the value is to 0, the more likely the electrostatic adsorption state of the batch of powder reagent to be tested is within the normal electrostatic distribution range, and the lower the risk of electrostatic adsorption. When the electrostatic adsorption state of the batch of powdered reagent to be tested is exactly on the decision boundary, the electrostatic adsorption risk of this batch is at the critical state between normal and abnormal electrostatic adsorption.

[0116] The process of constructing the electrostatic adsorption risk identification model for powdered pharmaceuticals in this embodiment can accurately assess the degree of deviation of the current batch from the normal baseline, realize early identification and quantitative warning of electrostatic adsorption risks, and provide a reliable technical means for quality control and electrostatic adsorption risk prevention in the production process of powdered pharmaceuticals.

[0117] The risk diagnosis module is used to collect baseline state characteristic data, flowability assessment index data, and electrostatic assessment index data of the powdered agent to be tested in real time during the hopper conveying stage. Combined with the powdered agent flowability decay assessment model and the powdered agent electrostatic adsorption risk identification model, the production risk of the powdered agent to be tested is analyzed.

[0118] Please see Figure 5 , Figure 5 This is a flowchart of risk analysis for powder pharmaceutical production provided in an embodiment of this application. The specific steps are as follows:

[0119] Step S41: Read the real-time production data of the current batch of powder reagent to be tested during the hopper conveying stage from the production process control system, and extract the 5-dimensional feature vector required for assessing the degree of flowability degradation, denoted as... ,in This is the initial moisture value. This is the inlet moisture value. This is the median particle size value. Set the feeding speed value. This represents the average surface roughness value of the corresponding hopper; simultaneously, the 6-dimensional feature vector required for electrostatic adsorption risk identification is extracted, denoted as... ,in This is the initial moisture value. This is the inlet moisture value. The d10 particle size value The d90 particle size value This is the inlet flow rate value. This represents the average surface resistivity value of the corresponding hopper;

[0120] Step S42: Transfer the liquidity feature vector Input the powder agent flowability degradation assessment model to obtain the flowability degradation value of the current batch of powder agent to be tested; and input the electrostatic feature vector. Input the electrostatic adsorption risk identification model for powdered reagents to obtain the electrostatic adsorption risk value of the current batch of powdered reagents to be tested;

[0121] Step S43: Based on the fluidity degradation value and electrostatic adsorption risk value of the current batch of powder reagent to be tested, the entropy weight method is used to determine the weight coefficients of the fluidity degradation value and electrostatic adsorption risk value, and then a weighted sum is performed to calculate the comprehensive production risk value of the batch of powder reagent to be tested; the specific steps are as follows:

[0122] First, considering that the weighting coefficients need to objectively reflect the contribution of the two indicators to the overall production risk, and that recent production status is more representative of the fluctuation characteristics under current process conditions, data from the most recent 30 normal powder reagent batches were selected from historical production batches. The entropy weight method was used to determine the weighting coefficients for the fluidity degradation value and the electrostatic adsorption risk value. The specific calculation process is as follows:

[0123] Based on the degree of fluidity degradation of these 30 normal powder reagent batches and electrostatic adsorption risk value As evaluation indicators, an original evaluation matrix is ​​constructed. The first column Corresponding liquidity indicators, second column The corresponding electrostatic index has a matrix dimension of ;

[0124] Subsequently, the original evaluation matrix was normalized, and the first evaluation matrix was calculated. The first indicator The proportion of the index value of each batch ,in The normalized matrix is ​​obtained by corresponding to the two indicators of liquidity and static electricity respectively. ; Calculate the first Information entropy of each indicator ,when Time definition Information entropy The smaller the value, the greater the degree of variation of the indicator, the more information it contains, and the larger the weighting coefficient accordingly.

[0125] Finally, the first number is calculated based on the information entropy. Weighting coefficients of each indicator The weighting coefficients for the degree of liquidity recession are obtained. Weighting coefficient of electrostatic adsorption risk value And satisfy ;

[0126] Secondly, the flowability degradation value and electrostatic adsorption risk value of the current batch of powder reagent to be tested are weighted and summed with the corresponding weighting coefficients to obtain the comprehensive production risk value of the batch of powder reagent to be tested.

[0127] The powder agent production risk analysis process in this embodiment effectively quantifies the complex situation of synergistic deterioration of fluidity decline and electrostatic adsorption. The final output comprehensive production risk value can truly reflect the overall risk level of the powder agent transportation process, providing technical support with both theoretical depth and practical value for risk warning, key batch identification and differentiated control in the production process.

[0128] The control output module is used to output control suggestions for the production process of the powder reagent to be tested based on the production risks of the powder reagent to be tested.

[0129] Please see Figure 6 , Figure 6 This is a flowchart of the production process control suggestions for the powder reagent to be tested provided in the embodiments of this application. The specific steps are as follows:

[0130] Step S51: Based on the distribution of the comprehensive production risk values ​​of all normal batches of powdered reagents, a risk level classification threshold is set. The comprehensive production risk value of the batch of powdered reagent to be tested is compared with the risk level classification threshold to determine the production risk level of the batch of powdered reagent to be tested. The specific steps are as follows:

[0131] First, the data of all normal batches of powdered reagent samples were re-entered into the powdered reagent flowability degradation assessment model and the powdered reagent electrostatic adsorption risk identification model to obtain the corresponding flowability degradation value and electrostatic adsorption risk value. Since normal batches of powdered reagents themselves fluctuate, their flowability degradation value and electrostatic adsorption risk value are not zero, but show a certain distribution range.

[0132] Secondly, the flowability degradation value and electrostatic adsorption risk value of all normal batches of powdered reagents are input into the risk diagnosis module to obtain the corresponding comprehensive production risk value. The comprehensive production risk values ​​of all normal batches of powdered reagents are arranged in ascending order, and the 95th percentile is taken as the risk threshold. ;

[0133] Finally, based on the production risk of the powdered reagent to be tested Risk level determination: If This indicates that the production risk of the powdered reagent being tested is within the normal fluctuation range, and the production risk level of the current batch of powdered reagent is determined to be a normal risk level; if If the production risk of the powder agent to be tested exceeds the normal fluctuation range, the production risk level of the current batch of powder agent is determined to be a level of concern.

[0134] Step S52: Based on the production risk level of the batch of powder reagent to be tested, generate differentiated production process control suggestions for the powder reagent to be tested, specifically including setting corresponding control operations, monitoring frequencies, and handling measures for different production risk levels; the specific steps are as follows:

[0135] First, for batches of powder reagents to be tested that are determined to be at a normal risk level, the following recommendations are generated for the production process control of the powder reagents to be tested: maintain the existing process parameters and continue production, maintain the current monitoring frequency, and no additional intervention measures are required;

[0136] Secondly, for batches of powder reagents to be tested that are determined to be at a risk level of concern, suggestions for adjusting the production process of the powder reagents to be tested are generated:

[0137] If the flowability degradation value of the powder reagent to be tested exceeds the 95th percentile of the flowability degradation value of all normal batches of powder reagents, while the electrostatic adsorption risk value of the powder reagent to be tested does not exceed the 95th percentile of the electrostatic adsorption risk value of all normal batches of powder reagents, then the risk source is determined to be abnormal flowability. At this time, the feeding speed setting value is reduced by 10% from the original value, and the powder moisture content and median particle size are checked. The monitoring frequency is increased to once per hour, and three batches are monitored continuously.

[0138] If the electrostatic adsorption risk value of the powder agent to be tested exceeds the 95th percentile of the electrostatic adsorption risk value of all normal batches of powder agents, while the flowability degradation value of the powder agent to be tested does not exceed the 95th percentile of the flowability degradation value of all normal batches of powder agents, then the source of risk can be determined to be electrostatic anomaly. At this time, reduce the conveying speed setting by 8% from the original value, adjust the ambient humidity to within the 45%RH range, check the hopper grounding system, and increase the monitoring frequency to once per hour for continuous monitoring of three batches.

[0139] If the fluidity degradation value and electrostatic adsorption risk value of the powder reagent to be tested both exceed their respective 95th percentiles, the risk source can be determined to be a dual abnormality of fluidity and electrostatics. At this time, reduce the feeding speed setting value by 10% and the conveying speed setting value by 10% from the original value. Adjust the ambient humidity to about 50%RH, check the powder moisture content, d10 particle size value and d90 particle size value, confirm that the hopper grounding system is intact, and increase the monitoring frequency to test twice per batch. If necessary, stop the machine for troubleshooting.

[0140] The output process of the production process control recommendations for the powdered reagents to be tested in this embodiment achieves a complete closed loop from risk identification to precise intervention by differentiating and binding production risk levels with control measures. This makes the control recommendations both clear at the operational level and adaptable to risk response, providing a systematic technical guarantee for the safe and stable operation of the powdered reagent production process.

[0141] This application provides a method for monitoring drug production based on big data, including the following steps:

[0142] S1. Obtain baseline state characteristic data, flowability assessment index data, and electrostatic assessment index data of powdered agents during the hopper conveying stage.

[0143] S2. Correlate the baseline state characteristic data and flowability assessment index data of powdered agents to construct an assessment model for the degree of flowability degradation of powdered agents.

[0144] S3. Combine the baseline state characteristic data and electrostatic assessment index data of powdered agents during the hopper conveying stage to construct a risk identification model for electrostatic adsorption of powdered agents.

[0145] S4. Real-time acquisition of baseline state characteristic data, flowability assessment index data, and electrostatic assessment index data of the powder agent to be tested during the hopper conveying stage. Combined with the powder agent flowability decay assessment model and the powder agent electrostatic adsorption risk identification model, the production risk of the powder agent to be tested is analyzed.

[0146] S5. Based on the production risks of the powder reagent to be tested, provide suggestions for the control of the production process of the powder reagent to be tested.

[0147] This application provides an electronic device, including a memory, a processor, and a communication bus; the memory and the processor are connected via the communication bus; the memory stores a drug production monitoring method based on big data that can be loaded by the processor and executed as provided in the above embodiments.

[0148] The memory can be used to store instructions, programs, code, code sets, or instruction sets; the memory may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function, and instructions for implementing the big data-based drug production monitoring method provided in the above embodiments, etc.; the data storage area may store data involved in the big data-based drug production monitoring method provided in the above embodiments, etc.

[0149] The processor may include one or more processing cores; the processor executes or runs instructions, programs, code sets or instruction sets stored in memory, calls data stored in memory, and performs various functions and processes data in this application; the processor may be at least one of the following: Application-Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), controller, microcontroller and microprocessor; it is understood that for different devices, the electronic device used to implement the above processor functions may also be other, and the embodiments of this application do not specifically limit it.

[0150] A communication bus may include a path for transmitting information between the aforementioned components; the communication bus may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc.; the communication bus may be divided into address bus, data bus, control bus, etc.

[0151] This application provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described in the above embodiments, which is a big data-based drug production monitoring method.

[0152] In this embodiment, a computer-readable storage medium can be a tangible device that holds and stores instructions used by an instruction execution device; a computer-readable storage medium can be, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof; specifically, a computer-readable storage medium can be a portable computer disk, a hard disk, a USB flash drive, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital multifunction disc (DVD), a memory stick, a floppy disk, an optical disk, a magnetic disk, a mechanical encoding device, or any combination thereof.

[0153] The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0154] The above description is merely a preferred embodiment of this application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of this application is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the foregoing application concept; for example, technical solutions formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions applied in this application.

Claims

1. A drug production monitoring system based on big data, characterized in that, The system includes: The data acquisition module is used to acquire baseline state characteristic data, flowability assessment index data, and electrostatic assessment index data of powdered agents during the hopper conveying stage. The rheological analysis module is used to correlate and process the baseline state characteristic data and flowability assessment index data of powdered agents to construct an assessment model for the degree of flowability degradation of powdered agents. Includes the following specific content: Mass flow time series data of each batch of powdered agent during the feeding process are extracted from the baseline state characteristic data and flowability assessment index data of powdered agent. By analyzing the fluctuation of mass flow, the initial moisture value, inlet moisture value, median particle size value, feeding speed set value and the average surface roughness value of the corresponding hopper of the powdered agent with normal flow rate are identified and extracted. At the same time, these five data values ​​of each batch of powdered agent with normal flow rate are normalized to generate the training dataset of the powdered agent flowability degradation assessment model. Based on the training dataset of the powder agent flowability degradation assessment model, a subsample set is constructed for each isolated tree. On each subsample set, isolated trees are recursively constructed by randomly selecting segmentation features and randomly generating segmentation values. The generated multiple isolated trees are combined to form an isolated forest, thereby completing the powder agent flowability degradation assessment model. The electrostatic analysis module is used to jointly analyze the baseline state characteristic data and electrostatic assessment index data of powdered agents during the hopper conveying stage, and construct a risk identification model for electrostatic adsorption of powdered agents. Includes the following specific content: The average wall potential value of the corresponding hopper during the feeding process of each batch of powdered agent is extracted from the baseline state characteristic data and electrostatic evaluation index data of powdered agent. In this way, the initial moisture value, inlet moisture value, d10 particle size value, d90 particle size value, inlet flow rate value and average surface resistivity value of the powdered agent in the electrostatic normal batch are identified and extracted. At the same time, these six data values ​​of each powdered agent in the electrostatic normal batch are normalized to generate the training dataset of the powdered agent electrostatic adsorption risk identification model. Based on the training dataset of the powder agent electrostatic adsorption risk identification model, the single-class support vector machine algorithm is used to fit the electrostatic normal samples in the kernel space. The support vectors and their corresponding weight coefficients and decision thresholds are determined by solving the quadratic programming problem, thereby completing the powder agent electrostatic adsorption risk identification model. The risk diagnosis module is used to collect baseline state characteristic data, flowability assessment index data and electrostatic assessment index data of the powder agent to be tested in real time during the hopper conveying stage. Combined with the powder agent flowability decay assessment model and the powder agent electrostatic adsorption risk identification model, the production risk of the powder agent to be tested is analyzed. The control output module is used to output control suggestions for the production process of the powder reagent to be tested based on the production risks of the powder reagent to be tested.

2. The big data-based drug production monitoring system according to claim 1, characterized in that, The acquisition of baseline state characteristic data, flowability assessment index data, and electrostatic assessment index data of powdered agents during the hopper conveying stage includes the following specific contents: Obtain baseline state characteristic data of powdered agents: Obtain the initial moisture value of each batch of powdered agent measured before feeding from the production records of multiple batches of the same powdered agent; The inlet moisture value of the same powder agent was collected from multiple batches during the conveying process using a near-infrared moisture meter installed on the conveying pipeline at the hopper inlet. Obtain flowability assessment data for powdered agents: collect the median particle size of multiple batches of the same powdered agent during the conveying process from an online particle size analyzer installed on the conveying pipeline at the hopper inlet; obtain the set value of the feeding speed of each batch of powdered agent from the hopper production process parameter records of multiple batches of the same powdered agent. The flow meter installed on the conveying pipe at the hopper inlet collects the mass flow time series data of each batch of powder agent during the conveying process; the inner wall of the hopper is divided into multiple monitoring areas according to the principle of equal area; before multiple batches of the same powder agent are fed, the surface roughness of the inner wall of the hopper in each monitoring area is measured by a surface roughness meter, and the arithmetic mean is calculated as the average surface roughness value of the hopper corresponding to each batch of powder agent; To obtain electrostatic evaluation index data for powdered agents: An online particle size analyzer installed on the conveying pipe at the hopper inlet was used to collect the d10 and d90 particle size values ​​of multiple batches of the same powdered agent during the conveying process; a flow rate sensor installed on the conveying pipe at the hopper inlet was used to collect the inlet flow rate values ​​of multiple batches of the same powdered agent during the conveying process; the inner wall of the hopper was divided into multiple monitoring areas according to the principle of equal area, and a non-contact electrometer was placed in each area. During the feeding process of multiple batches of the same powdered agent, the wall potential values ​​of all monitoring points were collected, and their arithmetic mean was calculated as the average wall potential value of the hopper corresponding to each batch of powdered agent; according to the same equal area division, before feeding multiple batches of the same powdered agent, a surface resistivity tester was used to measure the surface resistivity of the inner wall of the hopper in each monitoring area, and its arithmetic mean was calculated as the average surface resistivity value of the hopper corresponding to each batch of powdered agent.

3. The big data-based drug production monitoring system according to claim 2, characterized in that, The system collects real-time baseline state characteristic data, flowability assessment index data, and electrostatic assessment index data of the powdered agent under test during the hopper conveying stage. Combined with the powdered agent flowability degradation assessment model and the powdered agent electrostatic adsorption risk identification model, the system analyzes the production risk of the powdered agent under test, including the following specific contents: Real-time production data of the current batch of powder reagent to be tested during the hopper conveying stage is read from the production process control system to extract the 5-dimensional feature vector required for assessing the degree of flowability degradation, denoted as... ,in This is the initial moisture value. This is the inlet moisture value. This is the median particle size value. Set the feeding speed value. This represents the average surface roughness value of the corresponding hopper; simultaneously, the 6-dimensional feature vector required for electrostatic adsorption risk identification is extracted, denoted as... ,in This is the initial moisture value. This is the inlet moisture value. The d10 particle size value The d90 particle size value This is the inlet flow rate value. This represents the average surface resistivity value of the corresponding hopper; Liquidity feature vector Input the powder agent flowability degradation assessment model to obtain the flowability degradation value of the current batch of powder agent to be tested; and input the electrostatic feature vector. Input the electrostatic adsorption risk identification model for powdered reagents to obtain the electrostatic adsorption risk value of the current batch of powdered reagents to be tested; Based on the current flowability degradation value and electrostatic adsorption risk value of the powder reagent batch to be tested, the entropy weight method is used to determine the weight coefficients of the flowability degradation value and electrostatic adsorption risk value, and then a weighted sum is performed to calculate the comprehensive production risk value of the powder reagent batch to be tested.

4. The big data-based drug production monitoring system according to claim 3, characterized in that, Based on the production risks of the powdered reagent to be tested, suggestions for controlling the production process of the powdered reagent to be tested are output, including the following steps: Based on the distribution of the comprehensive production risk value of all normal batches of powdered reagents, a risk level classification threshold is set. The comprehensive production risk value of the batch of powdered reagents to be tested is compared with the risk level classification threshold to determine the production risk level of the batch of powdered reagents to be tested. Based on the production risk level of the batch of powder reagent to be tested, differentiated recommendations for the production process control of the powder reagent to be tested are generated, including setting corresponding control operations, monitoring frequencies and handling measures for different production risk levels.

5. A big data-based drug production monitoring method, applied to any one of the big data-based drug production monitoring systems described in claims 1-4, characterized in that, Includes the following steps: S1. Obtain baseline state characteristic data, flowability assessment index data, and electrostatic assessment index data of powdered agents during the hopper conveying stage. S2. Correlate the baseline state characteristic data and flowability assessment index data of powdered agents to construct an assessment model for the degree of flowability degradation of powdered agents. S3. Combine the baseline state characteristic data and electrostatic assessment index data of powdered agents during the hopper conveying stage to construct a risk identification model for electrostatic adsorption of powdered agents. S4. Real-time acquisition of baseline state characteristic data, flowability assessment index data, and electrostatic assessment index data of the powder agent to be tested during the hopper conveying stage. Combined with the powder agent flowability decay assessment model and the powder agent electrostatic adsorption risk identification model, the production risk of the powder agent to be tested is analyzed. S5. Based on the production risks of the powder reagent to be tested, provide suggestions for the control of the production process of the powder reagent to be tested.

6. An electronic device comprising a processor and a memory, characterized in that, The memory stores computer programs that can be called by the processor; the processor executes the big data-based drug production monitoring method as described in claim 5 by calling the computer programs stored in the memory.

7. A computer-readable storage medium storing instructions, characterized in that, When the instructions are executed on a computer, the computer performs the big data-based drug production monitoring method as described in claim 5.

Images