Drug filling and sealing sterilization post-light inspection data monitoring system based on deep learning

Abstract

The application is based on a deep learning-based pharmaceutical post-filling sterilization lamp inspection data monitoring system, which comprises: through a dose integration algorithm, a determined intensity change sequence is accumulated and calculated, position coordinates and stay time length information are fused, and an actual received dose value of each pharmaceutical product is obtained; if the obtained actual received dose value is lower than a preset sterilization threshold value, an environmental factor adjustment module is triggered to re-simulate a three-dimensional attenuation map, and whether the adjusted three-dimensional attenuation map improves the coverage of the low-intensity area is judged; by comparing the difference between the adjusted three-dimensional attenuation map and the original map, the optimized intensity distribution data is obtained, and the expected dose improvement amplitude of the pharmaceutical re-path is determined; according to the determined improvement amplitude, the input parameters of the dose integration algorithm are updated, the dose values of all pharmaceutical products are recalculated, and the sterilization consistency index of the batch as a whole is obtained; and a personalized evaluation report is output, and the sterilization state of each pharmaceutical product is judged to have met the requirements.

Description

Technical Field

[0001] This invention relates to the field of pharmaceutical technology, and in particular to a deep learning-based system for monitoring light inspection data after drug filling and sterilization. Background Technology

[0002] In modern pharmaceutical manufacturing, the precision and reliability of sterilization technology are directly related to drug safety and quality, and are one of the core pillars of the industry's development. During radiation sterilization, ensuring consistent sterilization effects for each batch of drugs under complex environments has become a crucial issue for safeguarding public health. Existing methods often struggle to adapt to the variable conditions in production environments when applying radiation sterilization in practice.

[0003] Many technologies focus on the output power of the equipment itself, neglecting the propagation characteristics of radiation in space and the impact of variations in drug placement. This oversight leads to uncertainty in sterilization effectiveness, especially in large-scale production, making it difficult to guarantee that every drug meets the expected sterilization standards. A deeper technical challenge lies in the fact that the distribution of radiation intensity in space is not uniform, but rather exhibits a complex attenuation pattern with distance and environmental factors. After emanating from the source, radiation gradually weakens with increasing distance, and is further affected by the material of the drug container and its placement, resulting in significant differences in sterilization intensity in different areas. This uneven spatial distribution further exacerbates the difficulty of monitoring, because without accurately grasping the cumulative effect of radiation at each location, it is difficult to determine whether the drug has truly met sterilization requirements. Summary of the Invention

[0004] This invention provides a deep learning-based monitoring system for light inspection data after drug filling and sterilization, the method comprising:

[0005] The output power of the sterilization lamp radiation source and environmental interference factors are collected in real time by a sensor array. A spatial intensity distribution model is constructed by processing the radiation propagation simulation algorithm to obtain a three-dimensional attenuation spectrum of radiation in the production environment.

[0006] Based on the obtained three-dimensional attenuation map and the conveyor belt speed parameters, the real-time position coordinates and corresponding dwell time of each medicine in the sterilization lamp area are tracked to determine the intensity change sequence on the medicine path;

[0007] The actual dose value of each drug is obtained by accumulating the determined intensity change sequence through a dose integral algorithm and fusing location coordinates and dwell time information.

[0008] If the actual received dose value is lower than the preset sterilization threshold, the environmental factor adjustment module is triggered to re-simulate the three-dimensional attenuation map and determine whether the adjusted three-dimensional attenuation map has improved the coverage of the low-intensity area.

[0009] By comparing the differences between the adjusted three-dimensional attenuation map and the original map, the optimized intensity distribution data is obtained, and the expected dose increase under the new drug pathway is determined.

[0010] The input parameters of the dose integral algorithm are updated according to the determined increase, the dose values ​​of all drugs are recalculated, and the overall sterilization consistency index of the batch is obtained.

[0011] If the batch consistency index meets the preset standard, a personalized evaluation report will be output to determine that the sterilization status of each drug meets the requirements.

[0012] Optionally, the 3D modeling module for the ray intensity field includes:

[0013] The power of the sterilization lamp radiation source and environmental interference in the production environment are collected in real time by a sensor array to obtain initial power data and interference data.

[0014] The power data and interference data are cleaned to remove outliers, resulting in a clean dataset.

[0015] For the cleaned data set after processing, a ray propagation simulation algorithm is applied to calculate and construct a preliminary framework for the spatial intensity distribution to obtain the initial distribution results;

[0016] Extract key points of spatial intensity change from the initial distribution results, combine them with the specific layout information of the production environment, adjust the distribution framework, and determine the corrected intensity distribution data.

[0017] If the corrected intensity distribution data exceeds the preset threshold in some areas, a second simulation calculation is performed on the data in the corresponding areas to obtain local attenuation information;

[0018] By integrating the corrected intensity distribution data with local attenuation information, a complete three-dimensional attenuation map is constructed, resulting in the final spatial distribution view.

[0019] Visualization tools are used to render the final spatial distribution view and generate a three-dimensional attenuation map.

[0020] Optionally, the intensity tracking module includes:

[0021] The initial position information of the drug within the sterilization lamp area is obtained from the three-dimensional attenuation map. Combined with the conveying speed data, the current coordinate data of the drug is determined through real-time calculation.

[0022] Based on the current coordinate data of the drug, determine whether it is within the sterilization lamp area. If the coordinate data falls within the preset area division, record its entry time and start accumulating the dwell time.

[0023] Based on the cumulative dwell time, combined with the conveying speed and path tracking data, the trajectory of the medicine moving within the sterilization lamp area is calculated, and the sequence of its position changes over time is obtained;

[0024] By using the sequence of location changes over time, the distribution of drug dwell time in different regional divisions can be obtained, and the key point information of the drug on the path can be determined.

[0025] Based on key location information and monitoring data on intensity changes, the intensity change values ​​of the drug at each segment along the route are calculated, and a complete intensity change sequence is derived.

[0026] Optionally, the strong cumulative dose accurate inversion module includes:

[0027] By processing the intensity change sequence, the cumulative dose value of each drug at different time periods is obtained;

[0028] Based on the cumulative value and location coordinate data, the dosage distribution of each medicine within the spatial range is determined using a preset mapping rule;

[0029] If the dose distribution exceeds the preset threshold range, the dwell time data is extracted to determine whether there is a dose deviation caused by abnormal dwell time.

[0030] By analyzing the dwell time data, the location coordinates of the abnormal periods can be obtained to determine the specific source of the deviation.

[0031] For the source of the deviation, a data fusion method is used to integrate the duration records and coordinate positioning information to obtain the corrected dose distribution data;

[0032] Based on the revised dose distribution data, the actual dose value of each drug is recalculated to obtain the actual received dose value.

[0033] Optionally, the environmental factor regulation module includes:

[0034] By analyzing the actual received dose values, the specific data of the dose values ​​are obtained and compared with the preset sterilization threshold to determine whether the standard requirements are met.

[0035] If the comparison results show that the dose value is lower than the sterilization threshold, the environmental factor adjustment module is triggered to correct the parameters of the environmental factor and obtain the adjusted factor data.

[0036] Regenerate the three-dimensional decay map based on the adjusted factor data to obtain a new three-dimensional decay map.

[0037] The distribution of low-intensity regions was extracted from the new three-dimensional attenuation map to determine the specific location and extent of the low-intensity regions;

[0038] By assessing the distribution of low-intensity areas, calculating changes in regional coverage, and determining whether coverage has improved;

[0039] If the coverage assessment results show that the coverage of low-intensity areas is not as expected, the environmental factor adjustment module will be activated again to obtain new adjustment parameters and generate an adjusted three-dimensional attenuation map.

[0040] Optionally, the evaluation module includes:

[0041] By obtaining difference data from the adjusted three-dimensional attenuation spectrum and the original spectrum, an initial comparative analysis result is constructed to obtain the specific manifestation of the spectrum differences;

[0042] Based on the specific manifestations of the differences in the intensity distribution, data extraction methods are used to separate the varying parts of the intensity distribution and determine the optimized data for the intensity distribution.

[0043] Based on the optimized intensity distribution data, comparative analysis was used to calculate the distribution change characteristics under the new path, and the intensity distribution results after the drug path adjustment were obtained.

[0044] By simulating the dose change trend under the new pathway using the intensity distribution results after the drug pathway adjustment, the initial magnitude of dose increase can be determined.

[0045] Based on the initial magnitude of the dose increase, a preset threshold is used for comparison. If the initial magnitude exceeds the threshold, the distribution calculation result is corrected a second time to obtain the corrected value of the dose increase.

[0046] For the correction value of the dose increase, combined with the specific adjustment parameters of the drug pathway, the stability of the dose distribution under the new pathway is analyzed, and the expected dose increase under the new pathway is determined.

[0047] Optionally, the global consistency iterative analysis module obtains the overall sterilization consistency index of the batch, including:

[0048] Based on the expected dose increase under the determined drug repath, algorithm parameters related to dose integration are extracted from the system to form an initial input adjustment scheme and determine the list of drugs to be processed.

[0049] For the list of drug types, the algorithm parameters for updating the dose integral are updated by inputting the adjustment scheme, and the preliminary dose data for each drug is obtained by matching them one by one.

[0050] From the preliminary dose data, the dose distribution characteristics of each drug are extracted. Statistical analysis tools are used to summarize the data for the whole batch. It is determined whether the fluctuation range of the batch data meets the preset threshold. If it exceeds the threshold, the abnormal data is marked and the marked dose dataset is output.

[0051] Based on the labeled dose dataset, the dose value of each drug is recalculated. By comparing the differences between the data before and after, the stable state after dose adjustment is determined, and a dose record after adjustment is generated.

[0052] From the adjusted dosage records, the overall dosage distribution information of the batch is extracted, and the sterilization consistency is analyzed using a logistic regression model to obtain the overall sterilization consistency index of the batch.

[0053] Optionally, the sterilization compliance determination module includes:

[0054] A preliminary analysis of the overall sterilization consistency indicators of the batch was conducted to obtain preliminary results on whether the indicators met the preset standards.

[0055] If the preliminary analysis results show that the sterilization consistency index meets the preset standard, then a deep comparison is performed on the sterilization data of each medicine to determine whether the sterilization status meets the requirements.

[0056] Based on the results of in-depth comparison, details of the sterilization status of each medicine are obtained, targeted data records are generated, and it is determined whether further processing is required.

[0057] If the data records show that the sterilization status of some drugs is not up to standard, the abnormal data will be extracted through the information processing stage to determine the classification of the abnormality.

[0058] Based on the results of the anomaly cause classification, a logistic regression model is used to predict and analyze the anomaly data to obtain the distribution of potential risk levels.

[0059] By analyzing the risk level distribution, personalized assessment data is generated to determine whether the sterilization status of each drug meets the requirements. If the final status assessment result shows that all meet the standards, all data are integrated to generate a complete batch assessment record and determine the overall batch qualification.

[0060] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0061] This invention achieves precise inversion of sterilization dosage and single-item-level tracking, eliminating detection blind spots. By constructing a three-dimensional modeling module and an intensity tracking module for the radiation intensity field, it changes the traditional crude mode of estimating dosage solely based on lamp power or single-point sensors. The system constructs a visualized three-dimensional attenuation map through a sensor array combined with a radiation propagation simulation algorithm. Combined with conveyor belt speed parameters, it can track the specific path and dwell time of each medicine in the sterilization area in real time. Through the cumulative dosage precise inversion module, the spatial intensity distribution and time dimension are deeply integrated, enabling accurate calculation of the actual cumulative dosage received by each bottle of medicine. This effectively avoids the problem of insufficient dosage caused by equipment vibration, conveyor belt speed fluctuations, or spatial obstruction, significantly improving the granularity and accuracy of sterilization detection.

[0062] The sterilization compliance judgment module of this invention adopts a deep comparison and logistic regression prediction model, which can not only output conventional qualified / unqualified judgments, but also classify the potential risk levels for abnormal data, and generate personalized assessment reports containing detailed path and dosage data for each drug. This provides detailed and traceable data support for the Good Manufacturing Practice (GMP) of pharmaceuticals, making the sterilization status of each drug leaving the factory verifiable, and significantly improving the safety and compliance management level of pharmaceutical production. Attached Figure Description

[0063] Figure 1 This is a flowchart of the deep learning-based drug filling and sterilization post-light inspection data monitoring system of the present invention. Detailed Implementation

[0064] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] like Figure 1 This embodiment of the deep learning-based drug filling and sterilization post-light inspection data monitoring system includes:

[0066] The output power of the sterilization lamp radiation source and environmental interference factors are collected in real time by a sensor array. The data are processed by a radiation propagation simulation algorithm to construct a spatial intensity distribution model and obtain a three-dimensional attenuation spectrum of radiation in the production environment.

[0067] A sensor array is used to collect real-time data on the power of the sterilization lamp radiation source and environmental interference in the production environment, obtaining initial power and interference data to establish a basic information set. Based on the collected power and interference data, a pre-established filtering mechanism is used to clean the data, removing outliers to obtain a clean data set. For this clean data set, a radiation propagation simulation algorithm is applied to construct a preliminary framework for spatial intensity distribution, obtaining initial distribution results. Key points of spatial intensity change are extracted from the initial distribution results, and the distribution framework is adjusted based on the specific layout information of the production environment to determine the corrected intensity distribution data. If the corrected intensity distribution data exceeds a preset threshold in certain areas, a second simulation calculation is performed on the data in those areas to obtain more accurate local attenuation information. By integrating the corrected intensity distribution data with the local attenuation information, a complete three-dimensional attenuation map is constructed, resulting in the final spatial distribution view. A visualization tool is used to render the final spatial distribution view, generating a three-dimensional attenuation map to assess the radiation attenuation trend in the production environment.

[0068] The output power of the sterilization lamp's radiation source and environmental interference factors are collected in real time using a sensor array. Specifically, a high-precision radiation sensor array is deployed at key locations in the production environment. For example, 20 sensor nodes are evenly distributed within a 10m × 10m × 5m workshop space. Each node collects radiation power data once per second. Assuming the output power of the sterilization lamp's radiation source in a given measurement is 500 watts, environmental interference factors such as temperature (25 degrees Celsius) and humidity (60%) are recorded simultaneously. The data is transmitted to the central processing unit via a wireless network. A radiation propagation simulation algorithm, specifically the Monte Carlo simulation method, is used to process the data. Based on the collected power data and environmental parameters, the propagation path of the radiation in three-dimensional space is simulated. Assuming the sterilization lamp's radiation source is located at the center point (5, 5, 2.5), the algorithm calculates the energy loss of the radiation in the air. Considering the influence coefficient of humidity on radiation attenuation as 0.02 per meter, the energy attenuation value along each path is obtained. For example, the power attenuates to 400 watts at a distance of 3 meters from the source, forming a preliminary attenuation dataset. A spatial intensity distribution model is constructed, using interpolation algorithms such as Kriging interpolation to transform discrete attenuation data points into a continuous three-dimensional intensity distribution. Assuming an intensity of 380 watts at the spatial point (3,3,2), the intensity at surrounding points is calculated through interpolation, forming a smooth distribution model. A three-dimensional attenuation map of the radiation in the production environment is generated, and visualization software is used to map the intensity distribution data into a heat map. Color depth represents intensity magnitude; for example, areas with intensity below 200 watts are marked in blue, and those above 400 watts are marked in red. The map resolution is 0.1 meters. Analysis of the map reveals that the intensity in areas 5 meters away from the sterilization lamp radiation source is below the safety threshold of 300 watts, which can be used to plan safe zones. This can then be combined with production equipment layout optimization to ensure that personnel and equipment operate within low-radiation areas, forming a complete logical chain from data acquisition to application decision-making.

[0069] Based on the obtained three-dimensional attenuation map and the conveyor belt speed parameters, the real-time position coordinates and corresponding dwell time of each medicine in the sterilization lamp area are tracked to determine the intensity change sequence on the medicine path.

[0070] The initial position information of the drug within the sterilization lamp area is obtained from the 3D attenuation map. Combined with the conveyor speed data, the current coordinate data of the drug is determined through real-time calculation. Based on the current coordinate data, it is determined whether the drug is within the sterilization lamp area. If the coordinate data falls within the preset area division, its entry time is recorded and the dwell time begins to accumulate. For the accumulated dwell time, combined with the conveyor speed and path tracking data, the trajectory of the drug within the sterilization lamp area is calculated, obtaining a sequence of its position changes over time. Through the sequence of position changes over time, the distribution of the drug's dwell time within different area divisions is obtained, determining its key point information on the path. Based on the key point information and the intensity change monitoring data, the intensity change value of the drug in each segment of the path is calculated, deriving a complete intensity change sequence. Using the intensity change sequence, combined with time records and area division, it is determined whether the drug has completed the intensity change within the preset sterilization lamp area. If it has not reached the preset threshold, it is marked as abnormal, and its coordinate data and dwell time are recorded.

[0071] The system utilizes information technology to achieve real-time position tracking and intensity change sequence analysis of pharmaceuticals within the sterilization lamp area. The specific implementation method is as follows: Using three-dimensional attenuation map data combined with conveyor belt speed parameters (assuming a constant conveyor belt speed of 0.5 m / s and a total length of 10 meters in the sterilization lamp area), the system collects pharmaceutical position data every 0.1 seconds using high-precision sensors installed on the conveyor belt, calculating the real-time coordinates of the pharmaceuticals. For example, if a pharmaceutical's initial position is 0 meters and its position is 1 meter after 2 seconds, the coordinate calculation formula is: Position = Initial Position + Speed ​​× Time, i.e., 1 = 0 + 0.5 × 2. Secondly, the system records the duration of the pharmaceuticals' stay within the sterilization lamp area. Assuming it takes 20 seconds for a pharmaceutical to enter and leave the area, the system records the entry time t1 and exit time t2 using timestamps, with the stay duration = t2 - t1 = 20 seconds. This data is then stored in a database for subsequent analysis. The system determines the intensity change sequence along the drug's path, assuming the intensity within the sterilization lamp area changes linearly with location, increasing by 10 units per meter. The system calculates the corresponding intensity value based on the location coordinates; for example, if the intensity is 10 units at 1 meter and 50 units at 5 meters, an intensity sequence [10, 20, 30, 40, 50] is generated. An algorithm is used to fit the intensity change curve using a linear regression model y=10x, where x is location and y is intensity. The correlation coefficient R² reaches 0.99, indicating a good fit. Location, duration, and intensity data are integrated into a cloud platform, and a dynamic path map is generated using data visualization tools. This analysis determines whether the drug meets sterilization standards. If the cumulative intensity value falls below a threshold of 500 units, the system automatically triggers an alarm and records the abnormal data in the log, linking it to subsequent quality inspection processes to ensure product quality traceability and form a complete business loop.

[0072] The actual dose value of each drug is obtained by accumulating the determined intensity change sequence through a dose integral algorithm and fusing location coordinates and dwell time information.

[0073] By collecting and processing the intensity change sequence, the cumulative dose value of each drug at different time periods is obtained. Based on the cumulative value and location coordinate data, the dose distribution of each drug within a spatial range is determined using a preset mapping rule. If the dose distribution exceeds a preset threshold range, the dwell time data is extracted to determine if there is a dose deviation caused by abnormal dwell time. By analyzing the dwell time data, the location coordinate information corresponding to the abnormal time period is obtained to determine the specific source of the deviation. For the source of the deviation, a data fusion method is used to integrate the duration record and coordinate positioning information to obtain corrected dose distribution data. Based on the corrected dose distribution data, the actual dose value of each drug is recalculated to obtain the actual received dose value. After obtaining the final dose result, corresponding dose allocation records are generated for different attributes of individual drugs, completing the data storage.

[0074] In the process of accumulating and calculating the determined intensity change sequence using a dose integration algorithm and fusing location coordinates and dwell time information to obtain the actual dose value received by each drug, the following specific implementation method can be used. Assume we have intensity change sequence data recording the radiation intensity values ​​of a certain drug at different time points. For example, the intensity is 2.5 units at time point t1=0 seconds, 3.0 units at t2=10 seconds, and 2.8 units at t3=20 seconds. Using the dose integration algorithm, these intensity values ​​are accumulated over time intervals, with the formula: dose D=Σ(intensity × time interval). The calculated dose from t1 to t2 is 2.5 × 10 = 25 units per second, and from t2 to t3 it is 3.0 × 10 = 30 units per second, for a total dose of 55 units per second. Analysis shows that the intensity reaches its peak at t2, and the cumulative dose increases rapidly, reflecting the concentration of radiation exposure. By integrating location coordinate information, assuming the drug is located at coordinates (x1, y1) = (10, 20) at time t1, (15, 25) at time t2, and (20, 30) at time t3, the movement distance is calculated based on the positional changes. The distance from t1 to t2 is 7.07 units, and the distance from t2 to t3 is also 7.07 units. Combined with intensity data analysis, there may be a correlation between positional movement and intensity changes. The intensity fluctuates slightly during the movement, and further verification is needed to determine whether it is affected by the environment. Then, by integrating dwell time information, assuming a dwell time of 10 seconds at coordinates (10, 20), 8 seconds at (15, 25), and 12 seconds at (20, 30), and combining this with dose calculation, the dwell time is used as a weight to adjust the dose allocation. The final dose proportions at each location are 33.3%, 26.7%, and 40.0%, respectively. The analysis shows that the dose proportion is higher at locations with longer dwell times, which is consistent with the expected logic. Based on the above data, the actual dose received by each drug was calculated. Assuming a total dose of 55 units per second distributed among the three drugs, and considering the weights of position and duration, drug A receives a dose of 18.3 units per second at (10,20), drug B receives 14.7 units per second at (15,25), and drug C receives 22.0 units per second at (20,30). Analysis shows that dose allocation is highly correlated with dwell time and position intensity, ensuring the rationality of the calculation. Through this method, the system automatically completes the entire process of data acquisition, integration calculation, position fusion, and dose allocation, forming a rigorous logical chain to ensure accurate results.

[0075] If the actual received dose value is lower than the preset sterilization threshold, the environmental factor adjustment module is triggered to re-simulate the three-dimensional attenuation map and determine whether the adjusted three-dimensional attenuation map has improved the coverage of the low-intensity area.

[0076] By analyzing the test results, specific dose values ​​are obtained and compared with preset sterilization thresholds to determine if they meet the standard requirements. If the comparison results show that the dose value is lower than the sterilization threshold, the environmental factor adjustment module is triggered to correct the parameters of the environmental factors and obtain adjusted factor data. Based on the adjusted factor data, a new three-dimensional attenuation map is generated to obtain new map information for subsequent coverage analysis. For the newly generated three-dimensional attenuation map, the distribution of low-intensity areas is extracted to determine the specific location and range of low-intensity areas. By evaluating the distribution of low-intensity areas, the change in regional coverage is calculated to determine whether the coverage rate has improved. If the coverage rate assessment results show that the coverage of low-intensity areas does not meet expectations, the environmental factor adjustment module is activated again to obtain new adjustment parameters and generate an updated three-dimensional attenuation map. By iteratively analyzing the updated three-dimensional attenuation map, the coverage changes of low-intensity areas are continuously monitored to determine the final coverage assessment result.

[0077] In practical applications, when the system detects that the obtained dose value is lower than the preset sterilization threshold—for example, the dose value at a certain point in the detection area is 3.5 units, while the preset threshold is 5.0 units—the system will automatically trigger the environmental factor adjustment module to re-simulate the three-dimensional attenuation spectrum. The system will collect current environmental data, such as a temperature of 25 degrees Celsius and humidity of 60%, and combine this with historical data analysis. It will then use a gradient descent-based optimization algorithm to adjust the environmental parameters, assuming the humidity is increased to 65% and the temperature is decreased to 23 degrees Celsius, and recalculate the dose three-dimensional attenuation spectrum to generate new simulation results. The system will perform coverage analysis on the adjusted three-dimensional attenuation spectrum. Specifically, it will compare the low-intensity areas (areas with doses below 5.0 units) before and after the adjustment, calculating the percentage increase in coverage. For example, if the low-intensity area percentage was 30% before the adjustment and decreased to 20% after the adjustment, the coverage rate has increased by 10 percentage points. The system will record this change and generate an analysis report. Simultaneously, the system further utilizes spatial distribution algorithms, such as a Gaussian distribution-based prediction model, to assess the uniformity of low-intensity areas in the new map. The standard deviation decreased from 2.5 to 1.8, indicating a more uniform distribution. If the coverage improvement does not meet expectations (e.g., a target increase of 15 percentage points), the system automatically iteratively adjusts parameters, incorporating business logic such as coordinating with the ventilation system to optimize airflow. Assuming the wind speed increases from 2 meters per second to 3 meters per second, the system simulates and analyzes again until the threshold requirement is met. The entire process is automated, with real-time data updates, ensuring rigorous logic and traceable results.

[0078] By comparing the differences between the adjusted three-dimensional attenuation map and the original map, the optimized intensity distribution data is obtained, and the expected dose increase under the new drug pathway is determined.

[0079] By extracting difference data from the adjusted and original drug pathway maps, an initial comparative analysis is constructed to obtain the specific manifestations of the map differences. Based on these differences, data extraction methods are used to separate the varying intensity distribution, determining the optimized intensity distribution data. For the optimized intensity distribution data, comparative analysis is employed to calculate the distribution change characteristics under the new pathway, obtaining the intensity distribution results after drug pathway adjustment. Using the adjusted intensity distribution results and distribution extrapolation techniques, the dose change trend under the new pathway is simulated to determine the initial dose increase. Based on the initial dose increase, a preset threshold is used for comparison. If the initial increase exceeds the threshold, the distribution extrapolation results are corrected a second time to obtain the corrected dose increase value. Based on the corrected dose increase value and the specific adjustment parameters of the drug pathway, the stability of the dose distribution under the new pathway is analyzed to determine the final dose increase. Based on the final dose increase, corresponding data records are generated and stored in a pre-established database to obtain complete business analysis results.

[0080] By comparing the adjusted 3D attenuation map with the original map using technical means, and obtaining optimized intensity distribution data, the expected dose increase under the new drug pathway was ultimately determined. Image processing algorithms were used to perform pixel-level comparisons of the original and adjusted 3D attenuation maps. Assuming the average intensity value of the original map was 50.5 units and the average intensity value of the adjusted map was 58.3 units, the difference was calculated to yield an intensity increase of 7.8 units. The formula is: Increase = Adjusted Intensity - Original Intensity, i.e., 58.3 - 50.5 = 7.8. Data analysis software was used to fit the intensity distribution to a normal distribution. The standard deviation of the original map was 5.2, and the standard deviation of the adjusted map was 4.8, indicating that the adjusted data distribution was more concentrated and the optimization effect was significant. The least squares method was used during the fitting process to ensure the error was less than 0.1%. Intensity distribution data is input into the dose prediction model. Based on historical data, assuming a dose increase of 0.02 mg per unit intensity, the expected dose increase is 7.8 × 0.02 = 0.156 mg. The prediction model uses a linear regression algorithm with a correlation coefficient of 0.95, indicating high prediction reliability. To form a logical chain, the dose increase is linked to drug pathway optimization. Assuming the new pathway reduces metabolic losses by 10%, and considering the 0.156 mg increase, the actual dose gain is calculated to be 0.156 × 1.1 = 0.1716 mg. Database comparison confirms this value is within the safe range (upper limit 0.2 mg). The entire process utilizes automated scripts for data extraction, calculation, and analysis. The intensity map comparison module uses an open-source image processing library for pixel analysis, while distribution fitting and dose prediction rely on statistical analysis tools to automatically generate reports, ensuring accurate and traceable results. Logically, this forms a complete closed loop from map differences to dose increase and pathway optimization.

[0081] The input parameters of the dose integration algorithm are updated based on the determined increase, and the dose values ​​of all drugs are recalculated to obtain the overall sterilization consistency index of the batch.

[0082] The process begins by acquiring preset boost level data, extracting algorithm parameters related to dose integration from the system to form an initial input adjustment scheme, and determining the list of drug types to be processed. For each drug type, the input adjustment scheme is used to update the dose integration algorithm parameters, and preliminary dose data for each drug is obtained through a one-to-one matching process. From the preliminary dose data, the dose distribution characteristics of each drug are extracted, and statistical analysis tools are used to summarize the data for the entire batch. The fluctuation range of the batch data is assessed to determine if it meets a preset threshold. If it exceeds the threshold, abnormal data is marked, and a marked dose dataset is output. Based on the marked dose dataset, the dose value of each drug is recalculated. By comparing the differences before and after the data, the stable state after dose adjustment is determined, and a dose record of the adjusted dose is generated. From the adjusted dose record, the overall dose distribution information of the batch is extracted, and a logistic regression model is used to analyze sterilization consistency, obtaining an overall sterilization consistency index for the batch. Based on the sterilization consistency index, combined with drug type and dose data, an overall batch evaluation dataset is generated. The final calculation results are saved through a data storage module, completing the data output.

[0083] In calculating the sterilization consistency index of drug batches, the input parameters of the dose integration algorithm need to be updated according to the determined improvement rate. Assuming the original baseline parameter of the dose integration algorithm is a radiation intensity of 10 units and the improvement rate is 20%, the updated parameter is 12 units. The system automatically reads the improvement rate data stored in the database and applies the formula: New parameter = Original parameter × (1 + Improvement rate) to complete the parameter update. The update result is stored in the system parameter table to ensure consistency in subsequent calculations. The dose values ​​of all drugs are recalculated based on the updated parameters. Assuming the batch contains 1000 drugs, the original average dose value is 50 units. Combining this with the new parameter of 12 units, the algorithm is: New dose value = Original dose value × (New parameter / Original parameter) = 50 × (12 / 10) = 60 units. The system processes all drug data in batches, generates a new dose value list, and stores it in the database. The calculation process is automatically recorded for traceability. The overall sterilization consistency index of the batch is then calculated. Assuming the new dosage range is 58 to 62 units, the system calculates a standard deviation of 1.5 and a mean of 60 units through statistical analysis. The consistency index is calculated using the formula: Consistency Index = 1 - (Standard Deviation / Mean) = 1 - (1.5 / 60) = 0.975. This index value reflects the uniformity of batch sterilization effectiveness. The system compares the results with historical data. If the result is below the threshold of 0.95, an anomaly alarm is triggered, and the quality inspection module is linked to generate rectification suggestions to ensure a closed-loop workflow. The above process is completed automatically by the system. Parameter updates, dosage calculations, and consistency analysis form a complete logical chain, and data storage and anomaly handling further ensure business continuity.

[0084] If the batch consistency index meets the preset standard, a personalized evaluation report will be output to determine that the sterilization status of each drug meets the requirements.

[0085] By obtaining consistency indicators from batch data and conducting preliminary analysis using a pre-established detection model, preliminary results are obtained regarding whether the indicators meet preset standards. If the preliminary analysis shows that the consistency indicators meet the preset standards, a deep comparison is performed on the sterilization data of each drug to determine whether the sterilization status meets the requirements. Based on the results of the deep comparison, details of the sterilization status of each drug are obtained, and targeted data records are generated to determine whether further processing is needed. If the data records show that the sterilization status of some drugs does not meet the standards, abnormal data is extracted through the information processing stage to determine the classification of abnormal causes. Based on the results of the abnormal cause classification, a logistic regression model is used to predict and analyze the abnormal data to obtain the distribution of potential risk levels. Through the analysis of the risk level distribution, personalized assessment data is generated to determine whether the final status of each drug meets the standards. If the final status assessment results show that all meet the standards, all data are integrated to generate a complete batch assessment record, determining the overall batch compliance.

[0086] In the process of evaluating batch consistency indicators and generating personalized reports, the system automatically collects sterilization data for each batch of medicines. For example, for a batch of 1000 medicines, the sterilization temperature data is set at a standard temperature of 121 degrees Celsius for a duration of 30 minutes. The system compares the actual sterilization temperature and time of each medicine one by one, calculating the percentage that meets the standard. Assuming the test result is that 980 items meet the standard, accounting for 98%, and the preset standard is 95%, the system determines that the batch's consistency indicator is qualified. The system then enters the consistency indicator analysis stage, using statistical algorithms to calculate the standard deviation of the data within the batch. Assuming the temperature standard deviation is 0.5 degrees Celsius and the time standard deviation is 1 minute, both within the allowable range (temperature ±1 degree Celsius, time ±2 minutes), the system automatically generates a consistency analysis report, including the mean, standard deviation, and distribution chart, for subsequent traceability. For each medicine, the system performs a personalized assessment based on the collected data. For example, for medicine A001, with a sterilization temperature of 121.2 degrees Celsius and a sterilization time of 31 minutes, the system compares it with the standard values, determines it to be qualified, and records the deviation values ​​(temperature deviation 0.2 degrees Celsius, time deviation 1 minute). If the deviation exceeds the range, it is marked as unqualified. The system integrates all data and automatically generates a personalized assessment report. The report lists the medicine number, sterilization parameters, deviation values, and status determination for each medicine, and also includes the statistical results of batch consistency indicators, such as a pass rate of 98%. The report is stored in the database in PDF format for subsequent query and auditing, ensuring that the entire process is traceable and meets regulatory requirements. Through the above steps, the system achieves fully automated processing from data collection to report output, with data at each stage closely linked to form a complete logical chain.

[0087] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A deep learning-based data monitoring system for light inspection of pharmaceutical filling and sterilization, characterized in that, include: The three-dimensional modeling module for radiation intensity field collects data on the output power of the sterilization lamp radiation source and environmental interference factors in real time through a sensor array. It then uses a radiation propagation simulation algorithm to process and construct a spatial intensity distribution model, thereby obtaining a three-dimensional attenuation spectrum of radiation in the production environment. The intensity tracking module tracks the real-time position coordinates and corresponding dwell time of each medicine in the sterilization lamp area based on the obtained three-dimensional attenuation map and the conveyor belt speed parameters, and determines the intensity change sequence on the medicine path; The cumulative dose accurate inversion module uses a dose integration algorithm to accumulate and calculate a determined intensity change sequence, and integrates location coordinates and dwell time information to obtain the actual received dose value for each drug. If the actual received dose value is lower than the preset sterilization threshold, the environmental factor regulation module will be triggered to re-simulate the three-dimensional attenuation spectrum and determine whether the adjusted three-dimensional attenuation spectrum has improved the coverage of the low-intensity area. The evaluation module compares the differences between the adjusted three-dimensional attenuation map and the original map to obtain optimized intensity distribution data and determine the expected increase in drug dosage. The global consistency iterative analysis module updates the input parameters of the dose integral algorithm according to the determined improvement range, recalculates the dose values ​​of all drugs, and obtains the overall sterilization consistency index of the batch. The sterilization compliance judgment module outputs a personalized assessment report if the batch consistency index meets the preset standard, determining that the sterilization status of each drug meets the requirements. The intensity tracking module includes: The initial position information of the drug within the sterilization lamp area is obtained from the three-dimensional attenuation map. Combined with the conveying speed data, the current coordinate data of the drug is determined through real-time calculation. Based on the current coordinate data of the drug, determine whether it is within the sterilization lamp area. If the coordinate data falls within the preset area division, record its entry time and start accumulating the dwell time. Based on the cumulative dwell time, combined with the conveying speed and path tracking data, the trajectory of the medicine moving within the sterilization lamp area is calculated, and the sequence of its position changes over time is obtained; By using the sequence of location changes over time, the distribution of drug dwell time in different regional divisions can be obtained, and the key point information of the drug on the path can be determined. Based on key location information and monitoring data on intensity changes, the intensity change values ​​of the drug at each segment along the route are calculated, and a complete intensity change sequence is derived.

2. The deep learning-based drug filling and sterilization post-light inspection data monitoring system according to claim 1, characterized in that, The three-dimensional modeling module for the ray intensity field includes: The power of the sterilization lamp radiation source and environmental interference in the production environment are collected in real time by a sensor array to obtain initial power data and interference data. The power data and interference data are cleaned to remove outliers, resulting in a clean dataset. For the cleaned data set after processing, a ray propagation simulation algorithm is applied to calculate and construct a preliminary framework for the spatial intensity distribution to obtain the initial distribution results; Extract key points of spatial intensity change from the initial distribution results, combine them with the specific layout information of the production environment, adjust the distribution framework, and determine the corrected intensity distribution data. If the corrected intensity distribution data exceeds the preset threshold in some areas, a second simulation calculation is performed on the data in the corresponding areas to obtain local attenuation information; By integrating the corrected intensity distribution data with local attenuation information, a complete three-dimensional attenuation map is constructed, resulting in the final spatial distribution view. Visualization tools are used to render the final spatial distribution view and generate a three-dimensional attenuation map.

3. The deep learning-based drug filling and sterilization post-discharge light inspection data monitoring system according to claim 1, characterized in that, The cumulative dose accurate inversion module includes: By processing the intensity change sequence, the cumulative dose value of each drug at different time periods is obtained; Based on the cumulative value and location coordinate data, the dosage distribution of each medicine within the spatial range is determined using a preset mapping rule; If the dose distribution exceeds the preset threshold range, the dwell time data is extracted to determine whether there is a dose deviation caused by abnormal dwell time. By analyzing the dwell time data, the location coordinates of the abnormal periods can be obtained to determine the specific source of the deviation. For the source of the deviation, a data fusion method is used to integrate the duration records and coordinate positioning information to obtain the corrected dose distribution data; Based on the revised dose distribution data, the actual dose value of each drug is recalculated to obtain the actual received dose value.

4. The deep learning-based drug filling and sterilization post-discharge inspection data monitoring system according to claim 1, characterized in that, The environmental factor regulation module includes: By analyzing the actual received dose values, the specific data of the dose values ​​are obtained and compared with the preset sterilization threshold to determine whether the standard requirements are met. If the comparison results show that the dose value is lower than the sterilization threshold, the environmental factor regulation module is triggered to correct the parameters of the environmental factors and obtain the adjusted factor data. Regenerate the three-dimensional decay map based on the adjusted factor data to obtain a new three-dimensional decay map. The distribution of low-intensity regions was extracted from the new three-dimensional attenuation map to determine the specific location and extent of the low-intensity regions; By assessing the distribution of low-intensity areas, calculating changes in regional coverage, and determining whether coverage has improved; If the coverage assessment results show that the coverage of low-intensity areas is not as expected, the environmental factor regulation module will be activated again to obtain new adjustment parameters and generate an adjusted three-dimensional attenuation map.

5. The deep learning-based drug filling and sterilization post-light inspection data monitoring system according to claim 1, characterized in that, The evaluation module includes: By obtaining difference data from the adjusted three-dimensional attenuation spectrum and the original spectrum, an initial comparative analysis result is constructed to obtain the specific manifestation of the spectrum differences; Based on the specific manifestations of the differences in the intensity distribution, data extraction methods are used to separate the varying parts of the intensity distribution and determine the optimized data for the intensity distribution. Based on the optimized intensity distribution data, comparative analysis was used to calculate the distribution change characteristics under the new path, and the intensity distribution results after the drug path adjustment were obtained. By simulating the dose change trend under the new pathway using the intensity distribution results after the drug pathway adjustment, the initial magnitude of dose increase can be determined. Based on the initial magnitude of the dose increase, a preset threshold is used for comparison. If the initial magnitude exceeds the threshold, the distribution calculation result is corrected a second time to obtain the corrected value of the dose increase. For the correction value of the dose increase, combined with the specific adjustment parameters of the drug pathway, the stability of the dose distribution under the new pathway is analyzed, and the expected dose increase under the new pathway is determined.

6. The deep learning-based drug filling and sterilization post-discharge inspection data monitoring system according to claim 1, characterized in that, The global consistency iterative analysis module obtains the overall sterilization consistency index of the batch, including: Based on the expected dose increase under the determined drug repath, algorithm parameters related to dose integration are extracted from the system to form an initial input adjustment scheme and determine the list of drugs to be processed. For the list of drug types, the algorithm parameters for updating the dose integral are updated by inputting the adjustment scheme, and the preliminary dose data for each drug is obtained by matching them one by one. From the preliminary dose data, the dose distribution characteristics of each drug are extracted. Statistical analysis tools are used to summarize the data for the whole batch. It is determined whether the fluctuation range of the batch data meets the preset threshold. If it exceeds the threshold, the abnormal data is marked and the marked dose dataset is output. Based on the labeled dose dataset, the dose value of each drug is recalculated. By comparing the differences between the data before and after, the stable state after dose adjustment is determined, and a dose record after adjustment is generated. From the adjusted dosage records, the overall dosage distribution information of the batch is extracted, and the sterilization consistency is analyzed using a logistic regression model to obtain the overall sterilization consistency index of the batch.

7. The deep learning-based drug filling and sterilization post-discharge inspection data monitoring system according to claim 1, characterized in that, The sterilization compliance determination module includes: A preliminary analysis of the overall sterilization consistency indicators of the batch was conducted to obtain preliminary results on whether the indicators met the preset standards. If the preliminary analysis results show that the sterilization consistency index meets the preset standard, then a deep comparison is performed on the sterilization data of each medicine to determine whether the sterilization status meets the requirements. Based on the results of in-depth comparison, details of the sterilization status of each medicine are obtained, targeted data records are generated, and it is determined whether further processing is required. If the data records show that the sterilization status of some drugs is not up to standard, the abnormal data will be extracted through the information processing stage to determine the classification of the abnormality. Based on the results of the anomaly cause classification, a logistic regression model is used to predict and analyze the anomaly data to obtain the distribution of potential risk levels. By analyzing the risk level distribution, personalized assessment data is generated to determine whether the sterilization status of each drug meets the requirements. If the final status assessment result shows that all meet the standards, all data are integrated to generate a complete batch assessment record and determine the overall batch qualification.

Images