An electronic irradiation dose real-time calibration management method and system

Abstract

The application provides an electron irradiation dose real-time calibration management method and system, relates to the technical field of data processing, and comprises the following steps: collecting real-time dose original data at different positions, performing mapping processing on the real-time dose original data according to transmission positions and time parameters, and forming a real-time dose matrix containing spatial distribution characteristics; comparing dose values of each point with preset dose reference threshold values, and generating a dose deviation distribution map containing deviation sizes and deviation positions; identifying local areas with deviations exceeding a preset deviation threshold range, and combining product batch numbers to generate abnormal dose batch information; extracting transmission speed parameters, stacking thickness parameters and irradiation power parameters of corresponding batch products to form a set of running state parameters; performing hierarchical calibration processing; and performing calibration verification processing; and the application improves the autonomy and accuracy of electron irradiation dose real-time calibration.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for real-time calibration and management of electron irradiation dose. Background Technology

[0002] Existing electron irradiation dose calibration methods are mostly based on offline computer data processing models. For example, raw data is collected over a certain period of time using a dose detector, and then this data is input into independent analysis software. A fixed calculation program is used to calculate the cumulative dose, thereby obtaining the calibration result. This method usually adopts a batch data acquisition and centralized calculation mode. Although it can ensure the accuracy of the values, its data processing often relies on post-processing modeling and result export, lacking synchronous mapping of acquired data and transmission parameters, and cannot directly generate a real-time spatial distribution matrix.

[0003] In industrial production scenarios involving food irradiation sterilization, products move continuously along conveyor belts, and there are variations in stacking thickness and packaging form between batches. If the aforementioned offline processing mode is still used, the computer can only output dosage results after irradiation is complete, preventing the system from performing real-time calculations and dynamic corrections for deviations during operation. For example, if a batch of products experiences dosage shifts due to localized fluctuations in conveyor belt speed, the computer results in offline mode can only be identified afterward. Real-time matrix comparison and deviation mapping cannot trigger calibration, potentially resulting in the failure to promptly label and trace abnormal dosage distributions in that batch of products. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for real-time calibration and management of electron irradiation dose, which aims to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] In a first aspect, a method for real-time calibration and management of electron irradiation dose, the method comprising:

[0007] Real-time dose raw data from different locations are collected, and the real-time dose raw data are mapped according to the transmission location and time parameters to form a real-time dose matrix containing spatial distribution characteristics.

[0008] Based on the real-time dose matrix, the dose values ​​at each point are compared with the preset dose reference threshold to generate a dose deviation distribution map that includes the magnitude and location of the deviation.

[0009] Based on the dose deviation distribution map, local areas where the deviation exceeds the preset deviation threshold are identified, and abnormal dose batch information is generated by combining the product batch number.

[0010] Based on the abnormal dose batch information, the transmission speed parameters, stacking thickness parameters and irradiation power parameters of the corresponding batch of products are extracted to form a set of operating status parameters.

[0011] Based on the operating status parameter set and dose deviation distribution map, a graded calibration process is performed: when the deviation is linearly distributed, the transmission speed is adjusted to generate a speed calibration result; when the deviation is concentrated, the local power is adjusted to generate a power calibration result; when the deviation is generally offset, the irradiation time is adjusted to generate a time calibration result, and the results are unified into a dynamic calibration record.

[0012] Based on the dynamic calibration record, a calibration verification process is performed: real-time dose raw data is collected again to form a new real-time dose matrix, which is compared with the dose reference threshold to generate a calibration verification result. If there is still a deviation, the graded calibration process is repeated until the deviation is eliminated.

[0013] Preferably, the raw real-time dose data is mapped according to transmission location and time parameters to form a real-time dose matrix containing spatial distribution characteristics, including:

[0014] The raw real-time dose data is sorted chronologically to obtain a dose time series arranged according to the acquisition time.

[0015] Based on the conveyor belt speed and product positioning marks, the dose time series is matched to obtain dose positioning data containing time and position coordinates;

[0016] The dose localization data is divided into two-dimensional grids and interpolated to obtain dose grid data that fully covers the transmission path;

[0017] The dose grid data is smoothed to eliminate random fluctuations and form a real-time dose matrix that reflects the spatial distribution characteristics.

[0018] Preferably, based on the real-time dose matrix, the dose values ​​at each point are compared with a preset dose reference threshold to generate a dose deviation distribution map containing the magnitude and location of the deviation, including:

[0019] The difference between the dose value and the dose reference threshold at each point in the real-time dose matrix is ​​calculated to obtain the point deviation data including the magnitude of the difference.

[0020] Based on the sign and range of the location deviation data, it is classified into ultra-high deviation, ultra-low deviation, and normal range, thus obtaining deviation classification data with classification attributes.

[0021] Based on the deviation classification data and the correspondence between the spatial locations of the points, a deviation label set with coordinate identifiers is generated;

[0022] By spatially combining and connecting the deviation markers, a dose deviation distribution map that reflects the magnitude and positional relationship of the deviation can be obtained.

[0023] Preferably, based on the dose deviation distribution map, local areas where the deviation exceeds a preset deviation threshold are identified, and abnormal dose batch information is generated by combining the product batch number, including:

[0024] Based on the dose deviation distribution map, the deviation markers are filtered by threshold to obtain the set of all out-of-range points that exceed the preset deviation threshold range;

[0025] Clustering is performed based on the spatial proximity of the out-of-range point set to obtain continuous or adjacent out-of-range area data;

[0026] Calculate the boundary range and time interval based on the abnormal area data to obtain the area index information that includes the location coverage and the time period of occurrence;

[0027] By associating regional index information with production batch numbers, abnormal dose batch information is generated, which includes the regional range, degree of deviation, and batch number.

[0028] Preferably, the graded calibration process includes:

[0029] Based on the dose deviation distribution map, the deviation values ​​at different points are trend-fitted to obtain deviation pattern data describing the directionality and amplitude changes of the deviation.

[0030] Based on the deviation pattern data, the overall change direction and local concentration characteristics are extracted to form a deviation feature vector for judgment.

[0031] Based on the deviation feature vector, perform type identification to determine whether the deviation belongs to linear distribution, centralized distribution or overall offset, and generate deviation type determination result;

[0032] Based on the deviation type determination result and the operating status parameter set, perform the corresponding calibration process:

[0033] When the deviation type determination result is linear distribution, the correction coefficient is calculated based on the relationship between the linear fitting slope and the transmission speed, and the speed calibration amount is obtained from the preset correction coefficient to generate the corresponding speed calibration result.

[0034] When the deviation type determination result is a concentrated distribution, the local power adjustment coefficient is calculated based on the extreme value of the dose deviation and the coverage area of ​​the concentrated area, and the power calibration amount is obtained from the preset adjustment coefficient to generate the corresponding power calibration result.

[0035] When the deviation type determination result is an overall offset, the time extension or shortening is calculated based on the difference between the overall average deviation and the target dose, the time calibration amount is obtained, and the corresponding time calibration result is generated.

[0036] The speed calibration results, power calibration results, and time calibration results are summarized and corresponding trigger conditions are added to form a dynamic calibration record that includes calibration actions, calibration range, and applicable scope.

[0037] Preferably, the calibration verification process includes:

[0038] Based on the adjusted operating parameters from the dynamic calibration record, the real-time dose data of the product along the transmission path is re-acquired to obtain the calibrated dose data.

[0039] The calibrated dose data is mapped to generate a new calibrated real-time dose matrix.

[0040] The new calibrated dose deviation data is obtained by comparing the real-time dose matrix after calibration with the dose reference threshold point by point.

[0041] Based on the new calibrated dose deviation data, the deviation values ​​of all points are statistically summarized, the overall average residual value and the local maximum residual value are calculated, and the residual statistics are formed.

[0042] The calibration verification results, which include pass or fail indicators, are generated by comparing the residual statistics with the preset allowable threshold.

[0043] When the calibration verification result is unqualified, the residual statistics result is compared with the dynamic calibration record to locate the uncorrected deviation pattern, and the process is returned to the graded calibration processing step for targeted adjustment.

[0044] When the calibration verification result is qualified, the final result after calibration is output and marked as qualified verification information.

[0045] Preferably, based on the dose deviation distribution map, trend fitting is performed on the deviation values ​​at different points to obtain deviation pattern data describing the directionality and amplitude changes of the deviation, including:

[0046] Based on the dose deviation distribution map, the deviation values ​​at each point are extracted and sorted according to the acquisition time to obtain the deviation time series;

[0047] Missing points are filled in the biased time series to form a complete corrected biased series;

[0048] Based on the corrected bias sequence, multinomial regression, piecewise linear regression, and weighted average fitting based on a sliding window are performed to obtain multiple candidate trend curves.

[0049] Based on the candidate trend curves, the residual sequence is calculated, and the mean, variance, and maximum deviation of the residuals are statistically analyzed to form a set of residual evaluation indicators.

[0050] Based on the residual evaluation index set, the candidate trend curve with the smallest residual and the highest stability is selected to form the optimal trend curve;

[0051] Based on the optimal trend curve, the overall slope, local inflection point position, and fluctuation amplitude are extracted to obtain deviation pattern data that reflects the directionality and amplitude changes of the deviation.

[0052] Preferably, based on the deviation pattern data, the overall change direction and local concentration characteristics are extracted to form a deviation feature vector for judgment, including:

[0053] Based on the deviation pattern data, calculate the average deviation value, average rate of change and trend stability index of all points to form a set of indicators related to the overall direction of change.

[0054] Based on the deviation pattern data, the points are divided into multiple spatial regions, and the local deviation extreme value, local variance and point density index are calculated in each region to form a set of local concentration related indicators.

[0055] Based on the set of indicators related to the overall direction of change, the average deviation value, average rate of change and trend stability index are weighted and integrated to obtain a single value for the overall direction of change.

[0056] Based on the set of local concentration-related indicators, the local deviation extreme value, local variance, and point density indicators are weighted and fused to obtain a single local concentration value.

[0057] The overall direction of change and the local concentration are combined to form the final deviation feature vector.

[0058] Preferably, based on the deviation feature vector, type identification is performed to determine whether the deviation belongs to a linear distribution, a concentrated distribution, or an overall shift, and a deviation type determination result is generated, including:

[0059] Based on the overall change direction value and local concentration value in the deviation feature vector, they are compared with the corresponding preset thresholds. When the overall change direction value exceeds the overall threshold, an overall offset candidate result is generated. When the local concentration value exceeds the concentration threshold, a concentrated distribution candidate result is generated. When neither exceeds the threshold, a linear distribution candidate result is generated. The rule confidence index is calculated based on the difference between the value and the threshold, and the confidence index is normalized to obtain the first type of judgment result.

[0060] The deviation feature vector is input into the trained classification model. The classification model calculates the matching probability for each deviation type based on the learning results of historical deviation samples, calculates the model confidence index based on the difference between the highest probability and the second highest probability, and normalizes the confidence index to obtain the second-class judgment result.

[0061] When the first type of judgment result is consistent with the second type of judgment result, the result of that type is directly output, and the normalized confidence index of the two is weighted and combined to obtain the final comprehensive confidence score.

[0062] When the first type of judgment result is inconsistent with the second type of judgment result, the normalized confidence index of the two is compared, the judgment result with higher confidence is selected as the final deviation type judgment result, and the corresponding confidence is output.

[0063] Secondly, an electron irradiation dose real-time calibration and management system, the system comprising:

[0064] The data acquisition module is used to acquire real-time dose raw data from different locations, and to map the real-time dose raw data according to the transmission location and time parameters to form a real-time dose matrix containing spatial distribution characteristics.

[0065] The deviation comparison module is used to compare the dose values ​​at each point with the preset dose reference threshold based on the real-time dose matrix, and generate a dose deviation distribution map that includes the magnitude and location of the deviation.

[0066] The anomaly detection module is used to identify local areas where the deviation exceeds the preset deviation threshold range based on the dose deviation distribution map, and generate abnormal dose batch information by combining the product batch number.

[0067] The parameter extraction module is used to extract the transmission speed parameters, stacking thickness parameters, and irradiation power parameters of the corresponding batch of products based on the abnormal dose batch information, and form a set of operating status parameters.

[0068] The graded calibration module is used to perform graded calibration processing based on the set of operating status parameters and the dose deviation distribution map: when the deviation is linearly distributed, the transmission speed is adjusted to generate a speed calibration result; when the deviation is concentrated, the local power is adjusted to generate a power calibration result; when the deviation is generally offset, the irradiation time is adjusted to generate a time calibration result, and the results are unified into a dynamic calibration record.

[0069] The calibration verification module is used to perform calibration verification processing based on the dynamic calibration record: real-time dose raw data is collected again to form a new real-time dose matrix, which is compared with the dose reference threshold to generate calibration verification results. If there is still a deviation, the graded calibration process is repeated until the deviation is eliminated.

[0070] The above-described solution of the present invention has at least the following beneficial effects:

[0071] First, by collecting raw real-time dose data from different locations and mapping it using transmission location and time parameters, a real-time dose matrix containing spatial distribution characteristics can be directly generated. Unlike traditional offline modes that can only perform cumulative value calculations, this matrix data not only preserves the temporal continuity of the dose but also achieves distributed characterization in the spatial dimension. This allows the system to generate dose distribution information in real time throughout the entire product transmission process, thus compensating for the lack of real-time spatial mapping in existing technologies.

[0072] Furthermore, by comparing the values ​​in the real-time dose matrix with a preset dose reference threshold, a deviation distribution map can be generated instantly, and the deviation area can be located using spatial positioning. This comparison mechanism can quickly reveal local anomalies, allowing the system to trigger a judgment at the initial stage of the deviation, rather than relying on centralized calculations after the entire irradiation process is completed. This approach transforms deviation detection from "post-event identification" to "online judgment," effectively improving the timeliness of anomaly localization.

[0073] Based on this, the present invention can automatically generate abnormal dose batch information according to the deviation distribution and batch number, and further extract the corresponding operational status parameters such as transmission speed, stacking thickness and irradiation power of the batch to construct an operational status parameter set. This synchronous binding mechanism between data and operational conditions enables deviation information to not only remain at the level of numerical judgment, but also to establish a direct link with the adjustable parameters of the production process, thereby realizing parameter-based targeted analysis and traceability.

[0074] Furthermore, by combining the set of operating status parameters and the deviation distribution map, the system can perform graded calibration. For different deviation patterns, corresponding calibration measures such as speed adjustment, power regulation, or time extension are adopted. This dynamic correction mode based on deviation type avoids blind adjustment of single parameters, achieves automatic classification and precise matching of calibration actions, and significantly improves calibration efficiency and the accuracy of results.

[0075] Finally, through a closed-loop mechanism of dynamic calibration recording and calibration verification, this invention ensures that every calibration result is verified in real time. When a residual deviation is detected, the system automatically returns to the calibration step and performs targeted corrections until the deviation is completely eliminated. This not only ensures the reliability of the output results but also makes the entire process adaptive and traceable, providing a long-term stable guarantee for dose control in large-scale continuous production. Attached Figure Description

[0076] Figure 1 This is a flowchart of a real-time calibration and management method for electron irradiation dose provided by an embodiment of the present invention. Detailed Implementation

[0077] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0078] like Figure 1 As shown, an embodiment of the present invention proposes a method for real-time calibration and management of electron irradiation dose, the method comprising:

[0079] Real-time dose raw data from different locations are collected, and the real-time dose raw data are mapped according to the transmission location and time parameters to form a real-time dose matrix containing spatial distribution characteristics.

[0080] Based on the real-time dose matrix, the dose values ​​at each point are compared with the preset dose reference threshold to generate a dose deviation distribution map that includes the magnitude and location of the deviation.

[0081] Based on the dose deviation distribution map, local areas where the deviation exceeds the preset deviation threshold are identified, and abnormal dose batch information is generated by combining the product batch number.

[0082] Based on the abnormal dose batch information, the transmission speed parameters, stacking thickness parameters and irradiation power parameters of the corresponding batch of products are extracted to form a set of operating status parameters.

[0083] Based on the operating status parameter set and dose deviation distribution map, a graded calibration process is performed: when the deviation is linearly distributed, the transmission speed is adjusted to generate a speed calibration result; when the deviation is concentrated, the local power is adjusted to generate a power calibration result; when the deviation is generally offset, the irradiation time is adjusted to generate a time calibration result, and the results are unified into a dynamic calibration record.

[0084] Based on the dynamic calibration record, a calibration verification process is performed: real-time dose raw data is collected again to form a new real-time dose matrix, which is compared with the dose reference threshold to generate a calibration verification result. If there is still a deviation, the graded calibration process is repeated until the deviation is eliminated.

[0085] In this embodiment of the invention, the method acquires and maps real-time dose data from different locations to form a real-time dose matrix reflecting spatial distribution characteristics. Compared to traditional methods relying on offline detection, this processing can directly acquire the dynamic situation of dose distribution during irradiation, transforming dose monitoring from a delayed, post-hoc assessment to real-time online monitoring, thus improving the timeliness and operability of the data.

[0086] By comparing the dose values ​​at each point in the real-time dose matrix with a preset dose reference threshold point by point, a dose deviation distribution map containing both the magnitude and location of the deviation can be generated. This processing method can not only identify the specific locations where the dose is too high or too low, but also mark them in real time when the deviation occurs, giving the deviation both spatial coordinates and numerical attributes, which facilitates rapid subsequent location.

[0087] Based on the generated deviation data, the method can further identify local areas where the deviation exceeds a preset deviation threshold and generate abnormal dosage batch data by combining product batch information. This processing enables automatic association between deviation information and production batches, allowing abnormal situations to be accurately traced and classified, avoiding the problems of discontinuous product batch records and difficulty in precise location in traditional methods.

[0088] After obtaining information on abnormal batches, the method can extract the corresponding operating status parameters for that batch of products, including transmission speed, stack thickness, and irradiation power. By establishing the correspondence between deviation distribution and operating status, parameterized analysis can be achieved, providing clear input conditions for subsequent graded calibration, thereby ensuring the pertinence and effectiveness of calibration measures.

[0089] During graded calibration, the method can take corresponding adjustment measures according to different deviation types. For deviations exhibiting a linear distribution, calibration can be achieved by adjusting the transmission speed; for deviations with a concentrated distribution, correction can be achieved by adjusting the local power; and for deviations with an overall offset, the difference can be eliminated by extending or shortening the irradiation time. This approach eliminates the reliance on human experience for calibration actions, instead enabling automatic selection through data-driven methods, significantly improving the accuracy and automation level of calibration.

[0090] After parameter adjustments, the method performs calibration verification by re-acquiring real-time dose data and generating a new real-time matrix. By comparing this matrix with a reference threshold, verification results are generated, and the calibration process is repeated if residual deviations are detected. This closed-loop processing mechanism ensures that the final output dose data fully meets the preset requirements, thereby achieving dynamic and iterative automatic calibration and improving the system's stability and reliability.

[0091] For example, in a food irradiation sterilization production line, a batch of products may experience insufficient dosage in some areas due to unstable conveyor belt speed. Using the method of this invention, the system first generates a dose deviation distribution in real time and identifies abnormal areas. Then, it associates these areas with the specific batch and automatically extracts the batch's operating speed parameters. Next, the method determines that the deviation is linearly distributed and performs calibration by adjusting the conveyor speed, followed by data collection for verification. If residual errors still exist, iterative correction continues until the deviation is eliminated. Finally, the system automatically outputs a qualified calibration result and marks it as traceable production information, achieving an integrated process of online detection, real-time calibration, and quality tracking.

[0092] In a preferred embodiment of the present invention, the raw real-time dose data is mapped according to transmission location and time parameters to form a real-time dose matrix containing spatial distribution characteristics, including:

[0093] The raw real-time dose data is sorted chronologically to obtain a dose time series arranged according to the acquisition time.

[0094] Based on the conveyor belt speed and product positioning marks, the dose time series is matched to obtain dose positioning data containing time and position coordinates;

[0095] The dose localization data is divided into two-dimensional grids and interpolated to obtain dose grid data that fully covers the transmission path;

[0096] The dose grid data is smoothed to eliminate random fluctuations and form a real-time dose matrix that reflects the spatial distribution characteristics.

[0097] In this embodiment of the invention, by organizing the raw real-time dose data in chronological order and mapping it with transmission location parameters, a dose matrix covering the transmission path can be generated, thereby achieving a visual representation of spatial distribution characteristics. This processing method can transform the originally scattered dose measurement point data into an information matrix with temporal and spatial characteristics, so that the dose data not only has temporal continuity but also reflects the positional distribution during the operation of the transmission belt.

[0098] By performing two-dimensional grid partitioning and interpolation on the dose localization data, a complete data distribution covering the entire transmission path can be formed, effectively solving the spatial blind zone problem caused by insufficient number of sensors or missing individual acquisition points. At the same time, smoothing operations further reduce random fluctuations caused by noise during acquisition, enabling the matrix data to more stably reflect the true dose distribution characteristics.

[0099] For example, on a food irradiation sterilization production line, the system collects real-time raw dose data from multiple sensors. By sorting this data according to the time of collection and matching the positions with the conveyor belt speed, a two-dimensional dose distribution matrix is ​​ultimately formed. This matrix can intuitively reflect the dose uniformity at different product locations, enabling production personnel to promptly detect dose defects caused by product stacking or abnormal conveying.

[0100] Specifically, the dose localization data is divided into two-dimensional grids and interpolated to obtain dose grid data that fully covers the transmission path, including:

[0101] After acquiring dose positioning data containing time and location coordinates, the entire transmission area is first divided into a regular two-dimensional coordinate grid based on the geometric layout of the conveyor belt and the product's movement path. The rows of this grid correspond to the longitudinal position of the transmission path, and the columns correspond to the temporal sequence during transmission. Each grid cell corresponds to a dose value to be estimated. Since the original acquisition points may only cover a portion of the grid area, interpolation algorithms are needed to fill in the missing points. Common interpolation methods include weighted average interpolation based on adjacent known points, or numerical inference based on the spatial distance relationship between multiple surrounding points. Through this process, continuously distributed dose values ​​can be obtained within the complete coverage area of ​​the transmission path, ensuring that the corresponding dose value can be found regardless of the product's location, thus realizing the transformation from discrete point data to continuously distributed data.

[0102] This includes smoothing the dose grid data to eliminate random fluctuations and forming a real-time dose matrix that reflects the spatial distribution characteristics. Specifically, this includes:

[0103] After completing the two-dimensional grid interpolation, the resulting dose grid data may contain local irregularities caused by detector noise, environmental interference, or instantaneous fluctuations. To avoid these abnormal fluctuations interfering with the overall judgment, the data needs to be smoothed. Common smoothing methods include: weighted averaging of adjacent points to reduce the impact of individual outliers; using a sliding window approach to average a continuous set of grid points to suppress instantaneous spikes; or employing curve fitting-based smoothing methods to make the dose data present a more stable and continuous trend overall. Through smoothing, the overall dose distribution characteristics can be effectively preserved while reducing local random disturbances, ultimately yielding a real-time dose matrix that accurately reflects the spatial dose distribution of the product along the transmission path.

[0104] In a preferred embodiment of the present invention, based on the real-time dose matrix, the dose values ​​at each point are compared with a preset dose reference threshold to generate a dose deviation distribution map containing the magnitude and location of the deviation, including:

[0105] The difference between the dose value and the dose reference threshold at each point in the real-time dose matrix is ​​calculated to obtain the point deviation data including the magnitude of the difference.

[0106] Based on the sign and range of the location deviation data, it is classified into ultra-high deviation, ultra-low deviation, and normal range, thus obtaining deviation classification data with classification attributes.

[0107] Based on the deviation classification data and the correspondence between the spatial locations of the points, a deviation label set with coordinate identifiers is generated;

[0108] By spatially combining and connecting the deviation markers, a dose deviation distribution map that reflects the magnitude and positional relationship of the deviation can be obtained.

[0109] In this embodiment of the invention, by comparing the real-time dose matrix with a preset threshold point by point, a dose deviation distribution map containing information on the magnitude and spatial location of the deviation can be generated. This method can not only quickly calculate the difference between each location and the target dose, but also present the deviation information in layers through classification and labeling, thereby providing an intuitive and visual effect of the deviation situation.

[0110] By calculating the difference and classifying positive and negative values, areas of excessive and insufficient dosage can be effectively distinguished, thus avoiding the shortcomings of traditional methods that only provide the overall average dosage and cannot reflect local anomalies. Furthermore, the deviation distribution map generated by spatial combination and connection forms a clear pattern of deviation in two-dimensional space, providing an intuitive basis for subsequent calibration.

[0111] For example, on a food irradiation sterilization production line, a batch of products may experience localized insufficient doses due to uneven stacking thickness. The system calculates the difference between each point and the threshold based on the real-time dose matrix and marks these deviations on a spatial distribution map, visually displaying the location and size of the insufficient dose areas. This allows managers to quickly identify abnormal areas without waiting for subsequent testing results.

[0112] The dose reference threshold specifically includes:

[0113] In electron irradiation treatment, a dose reference threshold is used as a criterion for determining whether the absorbed dose of a product meets the standard. This threshold is typically determined based on process standards, product category, and national or industry irradiation regulations. For example, different product categories have specified minimum effective dose and maximum safe dose ranges for food irradiation, medical device sterilization, or polymer material modification. In practice, the target dose range is first determined based on the process requirements of the target product; then, the lower limit of this target range is used as the minimum reference threshold, and the upper limit as the maximum reference threshold. Subsequently, when comparing the dose at each point in the real-time dose matrix, if the dose at a point is lower than the minimum reference threshold, it is determined that there is a risk of insufficient dose at that point; if the dose at a point is higher than the maximum reference threshold, it is determined that there is a risk of excessive irradiation at that point. This method enables compliance testing of dose values ​​at all points, ensuring that the judgment standard is clear, operable, and consistent.

[0114] The preset deviation threshold range specifically includes:

[0115] A preset deviation threshold range is used to identify and filter out points and regions exceeding tolerance in the dose deviation distribution map. This threshold range is typically set based on statistical analysis and process tolerance. First, based on long-term accumulated production data, dose deviations under different operating conditions are statistically analyzed to obtain the average deviation level and its fluctuation range. Second, combined with the process-allowed dose fluctuation range, a numerical range is set as the acceptable deviation limit. For example, if the process requires the product dose to have an allowable error of ± a certain percentage, then the range corresponding to that percentage is used as the preset deviation threshold range. During implementation, the difference value of each point in the deviation distribution map is judged: when the difference falls within the threshold range, the point is considered to be within the allowable deviation range; when the difference exceeds the threshold range, it is marked as an out-of-tolerance point. Subsequently, clustering and spatial proximity analysis methods are used to combine these out-of-tolerance points into local abnormal regions, thus providing a reliable basis for generating abnormal dose batch information. In this way, non-critical deviations caused by normal fluctuations can be effectively filtered out, and abnormal regions that pose a risk to product quality can be identified and located in a timely manner.

[0116] In a preferred embodiment of the present invention, based on the dose deviation distribution map, local areas where the deviation exceeds a preset deviation threshold are identified, and abnormal dose batch information is generated by combining the product batch number, including:

[0117] Based on the dose deviation distribution map, the deviation markers are filtered by threshold to obtain the set of all out-of-range points that exceed the preset deviation threshold range;

[0118] Clustering is performed based on the spatial proximity of the out-of-range point set to obtain continuous or adjacent out-of-range area data;

[0119] Calculate the boundary range and time interval based on the abnormal area data to obtain the area index information that includes the location coverage and the time period of occurrence;

[0120] By associating regional index information with production batch numbers, abnormal dose batch information is generated, which includes the regional range, degree of deviation, and batch number.

[0121] In this embodiment of the invention, by filtering and clustering the dose deviation distribution map, abnormal areas where the deviation exceeds a preset deviation threshold can be automatically identified, and abnormal dose batch information can be generated by combining the production batch number. This method simplifies complex deviation data into traceable abnormal batch records, directly linking deviation information to specific production processes, thereby achieving anomaly tracing and quality tracking.

[0122] The threshold-based selection of out-of-tolerance locations eliminates those within a reasonable fluctuation range, preventing false alarms. Using spatial clustering to generate continuous or adjacent anomalous region data further identifies the concentrated distribution of deviations along the transmission path. By associating region index information with production batch numbers, the final generated anomalous batch data not only includes the range of location and time but also marks the specific degree of deviation, providing a complete basis for subsequent graded calibration and batch tracking.

[0123] For example, on a food irradiation sterilization production line, a batch experienced insufficient dosage in multiple adjacent areas due to power fluctuations. The system uses threshold filtering and clustering of the deviation data to mark these areas as abnormal areas and generates an abnormal batch record based on the batch number. This record not only indicates the location and time of the abnormal area but also clarifies its correspondence with the production batch, enabling the abnormal batch to be isolated and handled immediately.

[0124] In a preferred embodiment of the present invention, the graded calibration process includes:

[0125] Based on the dose deviation distribution map, the deviation values ​​at different points are trend-fitted to obtain deviation pattern data describing the directionality and amplitude changes of the deviation.

[0126] Based on the deviation pattern data, the overall change direction and local concentration characteristics are extracted to form a deviation feature vector for judgment.

[0127] Based on the deviation feature vector, perform type identification to determine whether the deviation belongs to linear distribution, centralized distribution or overall offset, and generate deviation type determination result;

[0128] Based on the deviation type determination result and the operating status parameter set, perform the corresponding calibration process:

[0129] When the deviation type determination result is linear distribution, the correction coefficient is calculated based on the relationship between the linear fitting slope and the transmission speed, and the speed calibration amount is obtained from the preset correction coefficient to generate the corresponding speed calibration result.

[0130] When the deviation type determination result is a concentrated distribution, the local power adjustment coefficient is calculated based on the extreme value of the dose deviation and the coverage area of ​​the concentrated area, and the power calibration amount is obtained from the preset adjustment coefficient to generate the corresponding power calibration result.

[0131] When the deviation type determination result is an overall offset, the time extension or shortening is calculated based on the difference between the overall average deviation and the target dose, the time calibration amount is obtained, and the corresponding time calibration result is generated.

[0132] The speed calibration results, power calibration results, and time calibration results are summarized and corresponding trigger conditions are added to form a dynamic calibration record that includes calibration actions, calibration range, and applicable scope.

[0133] In this embodiment of the invention, by performing trend fitting on the dose deviation distribution map, pattern information reflecting the directionality and amplitude changes of the deviation can be obtained, so that the deviation is no longer just an isolated numerical difference, but can form a continuous trend feature. This pattern processing enables the system to capture the overall trend of the deviation as well as local concentrated anomalies, thereby providing a scientific basis for subsequent calibration.

[0134] After obtaining the deviation pattern, the method further extracts the overall direction of change and local concentration features, and transforms them into feature vectors. This vectorization not only compresses the original complex data but also strengthens the data's discriminative properties, making subsequent classification and recognition processes more efficient. By identifying the type of the deviation feature vector, the method can automatically determine whether the deviation belongs to a linear distribution, a concentrated distribution, or an overall shift, providing a direct basis for selecting appropriate calibration measures.

[0135] Once the type of deviation is identified, the system can perform targeted calibration based on the set of operating status parameters. For example, linearly distributed deviations can be corrected by adjusting the transmission speed; concentrated deviations can be compensated for by adjusting local power; and overall offset deviations can be corrected by extending or shortening the irradiation time. In this way, calibration measures no longer rely on human experience but are based on data-driven intelligent selection, thereby improving calibration efficiency and accuracy.

[0136] For example, on a food irradiation sterilization production line, a certain batch exhibited a trend of gradually decreasing product dose in the middle section of the conveyor belt. The system identified this deviation as linearly distributed through trend fitting and extracted the overall direction of change. Combined with the conveyor speed parameter, the system automatically generated a correction scheme, achieving dose compensation by reducing the conveyor speed, ultimately ensuring that the irradiation dose of this batch of products met the set requirements.

[0137] The correction factor specifically includes:

[0138] The correction factor is used to adjust the transmission speed when the deviation is linearly distributed. Its core function is to convert the trend of dose variation with location into a speed adjustment amount. The specific process is as follows: First, based on the deviation data at each point in the dose deviation distribution map, an overall trend curve is extracted, and its change is judged to have a linear pattern. When it is confirmed that the deviation and the transmission path position have an approximately linear relationship, the slope of the trend curve is used as a reflection of the rate of deviation change. Subsequently, a correspondence is established between this slope and the current operating speed of the transmission belt to measure the degree of speed's influence on the dose. Based on this, the required speed adjustment amount is derived by comparing the target dose with the actual deviation amplitude. Finally, the correction factor, as a scaling factor, is used to convert the speed adjustment amount into an actually executable speed calibration command, enabling the system to gradually eliminate the linear deviation caused by speed mismatch during subsequent operation.

[0139] The adjustment coefficient specifically includes:

[0140] The adjustment coefficient is used to adjust local power when the deviation is concentrated. The specific process is as follows: First, the deviation distribution map is clustered to identify concentrated areas of dose anomalies, and the extreme dose deviation values ​​and coverage areas of these areas are statistically analyzed. Next, based on the magnitude of the dose difference reflected by the extreme deviation values ​​and the influence range represented by the coverage area, the required power adjustment magnitude is comprehensively assessed. When the deviation is too high, the adjustment coefficient is used to reduce the power output corresponding to that area; when the deviation is too low, the adjustment coefficient is used to increase the power output of that area. The value of this coefficient is determined by the difference between the deviation magnitude and the target dose, while also considering the power adjustment sensitivity of the equipment to ensure that the adjustment can quickly eliminate the deviation without introducing new overshoot or undershoot. Finally, the adjustment coefficient is converted into a corresponding local power calibration value to guide the system in making targeted corrections to the radiation power of specific areas.

[0141] The process involves summarizing the speed calibration results, power calibration results, and time calibration results, and adding corresponding trigger conditions to form a dynamic calibration record that includes calibration actions, calibration amplitudes, and applicable ranges. Specifically, this includes:

[0142] After identifying and calibrating different deviation types, the calibration results need to be uniformly organized for dynamic retrieval during system operation. The specific process is as follows: First, the speed, power, and time calibration results are converted into standardized calibration instructions. Each instruction includes a calibration action (i.e., the type of adjustment, such as speed adjustment, power adjustment, or time adjustment) and a calibration magnitude (i.e., the corresponding adjustment amount). Second, based on the deviation distribution map and operating status parameters, the triggering conditions for each calibration instruction are determined, such as "triggering speed calibration when the deviation continuously exceeds a threshold" or "triggering power calibration when a power deviation persists in a certain area." Finally, all calibration instructions and their triggering conditions are combined to form a dynamic calibration record, and its applicable scope is defined, such as applicable to a specific batch, a specific area, or the entire transmission process. Through this process, the system can automatically invoke the corresponding calibration action based on the real-time collected deviation data during operation, achieving hierarchical and dynamic management of different deviation modes.

[0143] In a preferred embodiment of the present invention, the calibration verification process includes:

[0144] Based on the adjusted operating parameters from the dynamic calibration record, the real-time dose data of the product along the transmission path is re-acquired to obtain the calibrated dose data.

[0145] The calibrated dose data is mapped to generate a new calibrated real-time dose matrix.

[0146] The new calibrated dose deviation data is obtained by comparing the real-time dose matrix after calibration with the dose reference threshold point by point.

[0147] Based on the new calibrated dose deviation data, the deviation values ​​of all points are statistically summarized, the overall average residual value and the local maximum residual value are calculated, and the residual statistics are formed.

[0148] The calibration verification results, which include pass or fail indicators, are generated by comparing the residual statistics with the preset allowable threshold.

[0149] When the calibration verification result is unqualified, the residual statistics result is compared with the dynamic calibration record to locate the uncorrected deviation pattern, and the process is returned to the graded calibration processing step for targeted adjustment.

[0150] When the calibration verification result is qualified, the final result after calibration is output and marked as qualified verification information.

[0151] In this embodiment of the invention, by re-collecting and processing the adjusted operating parameters from the dynamic calibration records, a new real-time dose matrix can be formed to verify the calibration effect. This closed-loop verification method ensures that calibration is not a one-time action, but a repeatable dynamic iterative process, thereby guaranteeing the reliability and consistency of the results.

[0152] By comparing the new real-time dose matrix point by point with the reference threshold, the calibrated deviation data can be obtained, and the overall residual and local maximum residual can be further statistically analyzed. This statistical result comprehensively reflects the calibration effect, not only confirming whether most areas have met the standard, but also identifying any potential local residual anomalies.

[0153] After comparing the residual statistical results with the allowable threshold, if a deviation still exists, the system can automatically backtrack to the graded calibration step for targeted adjustments. If the result is satisfactory, a final output with a verification mark can be generated, achieving traceable quality control. This verification mechanism significantly reduces errors from human judgment, ensuring that the production line can maintain a stable dosage level even during long-term operation.

[0154] For example, on a food irradiation sterilization production line, the system adjusts the irradiation time based on the initial calibration record. Subsequently, by collecting dose data again and generating a real-time matrix, slight deviations were found in some local areas. The system then performs local power calibration again until the residual statistical results show that all points are within the allowable range. The final output is automatically marked as qualified verification information. The entire process requires no manual intervention, significantly improving production efficiency and reliability.

[0155] In a preferred embodiment of the present invention, based on the dose deviation distribution map, trend fitting is performed on the deviation values ​​at different points to obtain deviation pattern data describing the directionality and amplitude changes of the deviation, including:

[0156] Based on the dose deviation distribution map, the deviation values ​​at each point are extracted and sorted according to the acquisition time to obtain the deviation time series;

[0157] Missing points are filled in the biased time series to form a complete corrected biased series;

[0158] Based on the corrected bias sequence, multinomial regression, piecewise linear regression, and weighted average fitting based on a sliding window are performed to obtain multiple candidate trend curves.

[0159] Based on the candidate trend curves, the residual sequence is calculated, and the mean, variance, and maximum deviation of the residuals are statistically analyzed to form a set of residual evaluation indicators.

[0160] Based on the residual evaluation index set, the candidate trend curve with the smallest residual and the highest stability is selected to form the optimal trend curve;

[0161] Based on the optimal trend curve, the overall slope, local inflection point position, and fluctuation amplitude are extracted to obtain deviation pattern data that reflects the directionality and amplitude changes of the deviation.

[0162] In this embodiment of the invention, by performing trend fitting on the dose deviation distribution map, scattered point deviations can be transformed into trend curves with continuous and pattern characteristics. This approach not only reveals the dynamic changes of deviations over time and space but also effectively distinguishes between local random fluctuations and systematic deviations, providing valuable input data for subsequent classification and calibration.

[0163] By incorporating missing point completion, multiple regression methods, and weighted averaging into trend fitting, the integrity and accuracy of the trend curve can be guaranteed. Furthermore, through residual analysis and the construction of an evaluation index set, the merits of different trend curves can be quantified, thereby selecting the optimal trend curve and avoiding bias or distortion caused by a single fitting method. Finally, by extracting the overall slope, local inflection point positions, and fluctuation amplitude of the optimal curve, pattern data reflecting the directionality and magnitude of deviation can be generated, laying a solid foundation for feature extraction and type identification.

[0164] For example, in a food irradiation sterilization production line, the deviation data collected by the system contained some missing sensor data. Through a combination of automatic completion and multiple fitting methods, the system ultimately selected the trend curve with the smallest residual and extracted the feature of a clear inflection point in its middle section. This result indicates that the batch of products exhibits a systematic dose decrease trend during transmission. The system thus determines in the subsequent type identification stage that it belongs to a linear deviation mode, providing a basis for speed adjustment.

[0165] This includes filling in missing points in the biased time series to form a complete corrected biased series, specifically including:

[0166] During the acquisition of deviation time series data, deviation values ​​at some time points may be missing due to factors such as detector momentary failure, signal loss, or data transmission delay. Without imputation, these missing points will disrupt the continuity of the time series, thus affecting the accuracy of subsequent trend fitting. Therefore, it is necessary to imput these missing points. The specific method is as follows: First, the acquired deviation time series is scanned to identify the locations of the missing points. Then, based on the known data before and after the missing point, interpolation or extrapolation methods are used for imputation. For example, linear interpolation can be used to estimate the reasonable value of the missing point based on the values ​​and time intervals of two adjacent known points; alternatively, a local averaging-based imputation method can be used, taking the average of several known points near the missing point as the imputation value; for cases with continuous missing values, extrapolation based on the time series pattern can be used to estimate the missing segment based on the overall trend. Through these methods, a continuous and complete corrected deviation series on the time axis can be obtained, providing reliable input data for subsequent trend fitting.

[0167] Specifically, based on the corrected deviation sequence, multinomial regression, piecewise linear regression, and weighted average fitting based on a sliding window are performed to obtain multiple candidate trend curves, including:

[0168] After obtaining the complete corrected deviation sequence, trend fitting is required to extract potential patterns of change. The specific process is as follows: First, multinomial regression is performed, where an appropriate polynomial order is chosen to fit the deviation sequence into a smooth curve reflecting the overall trend. Second, piecewise linear regression is performed, dividing the entire time series into several continuous intervals and fitting a linear function within each interval. This method better captures local trend changes and inflection point characteristics. Third, a sliding window-based weighted average fitting process is performed, setting a fixed-length window on the sequence and weighting the data points within the window according to distance weights to obtain a smooth curve that updates over time, effectively eliminating random fluctuations. Through these three different fitting methods, candidate trend curves with overall, local, and stable characteristics can be obtained, providing diverse candidate bases for subsequent residual analysis and selection of the optimal trend curve.

[0169] In a preferred embodiment of the present invention, based on the deviation pattern data, the overall change direction and local concentration characteristics are extracted to form a deviation feature vector for judgment, including:

[0170] Based on the deviation pattern data, calculate the average deviation value, average rate of change and trend stability index of all points to form a set of indicators related to the overall direction of change.

[0171] Based on the deviation pattern data, the points are divided into multiple spatial regions, and the local deviation extreme value, local variance and point density index are calculated in each region to form a set of local concentration related indicators.

[0172] Based on the set of indicators related to the overall direction of change, the average deviation value, average rate of change and trend stability index are weighted and integrated to obtain a single value for the overall direction of change.

[0173] Based on the set of local concentration-related indicators, the local deviation extreme value, local variance, and point density indicators are weighted and fused to obtain a single local concentration value.

[0174] The overall direction of change and the local concentration are combined to form the final deviation feature vector.

[0175] In this embodiment of the invention, by indexing and weighted fusion processing of deviation pattern data, the overall direction of change and local concentration features can be effectively extracted, thereby transforming complex raw data into numerical features with discriminative significance. The numerical value of the overall direction of change is used to reflect the trend of overall deviation over time or space, while the numerical value of local concentration is used to reflect whether anomalies are clustered in a specific region. This decomposition method can comprehensively characterize deviation features from both macroscopic and microscopic dimensions.

[0176] By weighted fusion of the overall and local indicator sets, multiple different statistics are compressed into a single value, avoiding mutual interference between multidimensional indicators and making the features simpler and easier to compare. Furthermore, by combining two values ​​into a deviation feature vector, it can be used as a unified input in subsequent recognition stages, achieving a complete mapping from the original deviation pattern to identifiable features. This process not only reduces data dimensionality but also enhances the stability and robustness of the features.

[0177] For example, in a food irradiation sterilization production line, the deviation pattern data collected by the system contains a large number of irregular fluctuations. By weighting and fusing the average deviation, rate of change, and stability index, an overall value representing the direction of change is formed. Simultaneously, local extrema, variance, and density are weighted and fusing to form a local concentration value. The resulting deviation feature vector can intuitively show whether the batch of products deviates gradually as a whole during transport or whether concentrated anomalies occur in local areas, providing a clear basis for subsequent type determination.

[0178] Based on the deviation pattern data, the average deviation value, average rate of change, and trend stability index of all points are calculated to form a set of indicators related to the overall direction of change, specifically including:

[0179] After obtaining the deviation pattern data, the first step is to perform overall statistical analysis on the deviation values ​​of all points. Specifically, the deviations of all collected points are summed and divided by the total number of points to obtain the average deviation value, which reflects the magnitude and direction of the overall deviation. Subsequently, the differences between adjacent points in the deviation time series are statistically analyzed to calculate the average rate of change, reflecting the overall trend of deviation over time or location. Finally, the fluctuation of the deviation series is measured, for example, by examining the variance or range of the deviation over a period or region, as an indicator of trend stability, reflecting whether the overall change is stable or exhibits abrupt changes. Through these steps, the resulting set of indicators related to the overall direction of change provides a quantitative basis for subsequent judgments on whether the deviation exhibits an overall shift.

[0180] Based on the deviation pattern data, the locations are divided into multiple spatial regions, and the local deviation extreme value, local variance, and location density index are calculated in each region to form a set of local concentration-related indicators, specifically including:

[0181] In terms of spatial distribution, the deviation data needs to be regionalized. The specific method is as follows: First, the transmission path or irradiation area is divided into several adjacent sub-regions, each corresponding to a certain spatial range. Then, within each sub-region, the maximum and minimum values ​​of the local deviation are extracted to reflect extreme deviation situations in that region. Next, the dispersion of deviations at each point within the region is statistically analyzed, for example, by calculating local variance to measure the concentration or dispersion of deviation values. Finally, based on the number and distribution of collection points within the sub-regions, a point density index is calculated to reflect the completeness and representativeness of the data distribution in that region. Through the above processing, a set of local concentration-related indicators can be obtained to determine whether there are localized regional dose anomalies.

[0182] Specifically, based on the set of indicators related to the overall direction of change, the average deviation value, average rate of change, and trend stability index are weighted and fused to obtain a single overall direction of change value, which includes:

[0183] To make the judgment of the overall direction of change more objective, multiple indicators need to be combined into a single value. The specific process is as follows: First, based on empirical data or process requirements, weights are assigned to the average deviation value, average rate of change, and trend stability indicators, with the weight reflecting their importance in the overall deviation judgment. Then, the three indicators are weighted and summed or equivalently combined according to their weights to obtain a unified numerical result. This value not only reflects the magnitude and direction of the overall deviation but also takes into account the rate of change and trend stability. This fusion process avoids the bias caused by a single indicator, improving the accuracy and reliability of the overall deviation direction determination.

[0184] Specifically, based on the local concentration index set, the local deviation extreme value, local variance, and point density index are weighted and fused to obtain a single local concentration value, which includes:

[0185] In determining local deviations, multiple indicators need to be converted into a single, comparable, and quantifiable value. The specific method is as follows: First, based on the importance of different indicators, weights are assigned to local deviation extremes, local variance, and point density. For example, if greater emphasis is placed on the degree of anomaly, the weight of deviation extremes is increased; if greater emphasis is placed on deviation stability, the weight of variance is increased. Then, the indicators are combined according to their weights to obtain a comprehensive value. This value can intuitively reflect the concentration and severity of deviations within a certain area. When this value exceeds a set threshold, it can be determined that there is a significant concentrated distribution deviation in that area. This process provides a clear quantitative basis for subsequent identification of deviation types.

[0186] In a preferred embodiment of the present invention, based on the deviation feature vector, type identification is performed to determine whether the deviation belongs to a linear distribution, a concentrated distribution, or an overall shift, and a deviation type determination result is generated, including:

[0187] Based on the overall change direction value and local concentration value in the deviation feature vector, they are compared with the corresponding preset thresholds. When the overall change direction value exceeds the overall threshold, an overall offset candidate result is generated. When the local concentration value exceeds the concentration threshold, a concentrated distribution candidate result is generated. When neither exceeds the threshold, a linear distribution candidate result is generated. The rule confidence index is calculated based on the difference between the value and the threshold, and the confidence index is normalized to obtain the first type of judgment result.

[0188] The deviation feature vector is input into the trained classification model. The classification model calculates the matching probability for each deviation type based on the learning results of historical deviation samples, calculates the model confidence index based on the difference between the highest probability and the second highest probability, and normalizes the confidence index to obtain the second-class judgment result.

[0189] When the first type of judgment result is consistent with the second type of judgment result, the result of that type is directly output, and the normalized confidence index of the two is weighted and combined to obtain the final comprehensive confidence score.

[0190] When the first type of judgment result is inconsistent with the second type of judgment result, the normalized confidence index of the two is compared, the judgment result with higher confidence is selected as the final deviation type judgment result, and the corresponding confidence is output.

[0191] In a preferred embodiment of the present invention, by analyzing the deviation feature vector, the deviation type can be accurately classified into three categories: linear distribution, concentrated distribution, or overall offset, and a corresponding confidence index is generated for each category. This typified processing method can significantly improve the targeting of calibration, enabling different calibration measures to correspond to different deviation patterns, thereby achieving automated intelligent decision-making.

[0192] In the type determination process, the method relies not only on the results of rule comparison but also on the classification results based on the trained model. By normalizing the confidence scores obtained from rule-based determination and model-based determination, the two types of results are made comparable. Thus, when the two types of results are consistent, a fused comprehensive confidence score can be output; when the results are inconsistent, the more reliable result can be selected by comparing the confidence scores. This dual-channel fusion mechanism effectively improves the stability and reliability of the determination, avoiding misjudgments that might arise from a single method.

[0193] For example, on a food irradiation sterilization production line, a batch of products exhibited both an overall decreasing trend in dose distribution and distinct low-dose points in localized areas. The system first uses rule-based judgment to identify that the overall change direction exceeds a threshold, suggesting an overall shift; simultaneously, the classification model, based on historical sample training results, determines a concentrated distribution. After confidence level normalization and comparison, the system ultimately selects the judgment with higher confidence as the output, along with its reliability. In this way, the calibration module can take targeted measures to ensure the uniformity and safety of the irradiation dose for that batch of products.

[0194] The overall threshold specifically includes:

[0195] The overall threshold is used to determine the overall direction of change in the deviation feature vector to ascertain whether an overall deviation exists. The specific process is as follows: First, based on historical process data and deviation samples accumulated in actual production, the distribution range and common fluctuation amplitude of the overall deviation values ​​are statistically analyzed. Then, in conjunction with industry standards or product dosage control specifications, a numerical range capable of distinguishing between normal fluctuations and abnormal overall deviations is set. When the overall direction of change value in the deviation feature vector exceeds the upper limit of this set range, the system automatically identifies it as a candidate result for overall deviation; if the value is within the range, the overall deviation is considered to be within an acceptable range. In this way, the overall threshold provides an objective basis for distinguishing between linear deviations and overall deviations.

[0196] The concentration threshold specifically includes:

[0197] The concentration threshold is used to determine the local concentration values ​​in the deviation feature vector, identifying whether there is a significant concentrated distribution deviation. The specific process is as follows: First, based on the deviation distribution of multiple batches of products, the typical range of local concentration values ​​is statistically analyzed, and the degree of random clustering that may occur under normal process conditions is assessed. Then, combined with product quality standards or allowable dose distribution non-uniformity, a threshold range is set. When the local concentration value exceeds this threshold, the region is determined to have a candidate concentrated distribution result. This threshold typically considers both the deviation magnitude and spatial clustering characteristics, thus avoiding misjudgments due to a small number of random outliers. In this way, the concentration threshold can effectively identify local abnormal areas that significantly affect product consistency.

[0198] The classification model specifically includes:

[0199] A classification model is used to intelligently identify the type of deviation feature vectors, thereby improving the accuracy and reliability of the judgment results. The specific process is as follows: First, a training set is constructed using historical production data and known deviation samples. The samples contain three typical deviation types: linear distribution, concentrated distribution, and overall deviation. Then, an appropriate machine learning model is selected for training, such as a support vector machine, random forest, or neural network. The overall direction of change and local concentration are input into the model for learning through feature extraction methods. After training, the classification model can output the matching probability of each type based on the input deviation feature vector. Finally, the system uses the deviation type with the highest matching probability as the model's judgment result and compares it with the first type judgment result based on a threshold. If they match, the confidence of the final judgment is increased; if they do not match, the result with higher confidence is selected as the final output. By introducing a classification model, the ability to identify complex deviation patterns can be significantly enhanced, achieving efficient classification and accurate judgment of different types of deviations.

[0200] Embodiments of the present invention also provide a real-time electron irradiation dose calibration and management system, the system comprising:

[0201] The data acquisition module is used to acquire real-time dose raw data from different locations, and to map the real-time dose raw data according to the transmission location and time parameters to form a real-time dose matrix containing spatial distribution characteristics.

[0202] The deviation comparison module is used to compare the dose values ​​at each point with the preset dose reference threshold based on the real-time dose matrix, and generate a dose deviation distribution map that includes the magnitude and location of the deviation.

[0203] The anomaly detection module is used to identify local areas where the deviation exceeds the preset deviation threshold range based on the dose deviation distribution map, and generate abnormal dose batch information by combining the product batch number.

[0204] The parameter extraction module is used to extract the transmission speed parameters, stacking thickness parameters, and irradiation power parameters of the corresponding batch of products based on the abnormal dose batch information, and form a set of operating status parameters.

[0205] The graded calibration module is used to perform graded calibration processing based on the set of operating status parameters and the dose deviation distribution map: when the deviation is linearly distributed, the transmission speed is adjusted to generate a speed calibration result; when the deviation is concentrated, the local power is adjusted to generate a power calibration result; when the deviation is generally offset, the irradiation time is adjusted to generate a time calibration result, and the results are unified into a dynamic calibration record.

[0206] The calibration verification module is used to perform calibration verification processing based on the dynamic calibration record: real-time dose raw data is collected again to form a new real-time dose matrix, which is compared with the dose reference threshold to generate calibration verification results. If there is still a deviation, the graded calibration process is repeated until the deviation is eliminated.

[0207] It should be noted that this system is the system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0208] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0209] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0210] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for real-time calibration and management of electron irradiation dose, characterized in that, The method includes: Collect raw real-time dose data from different locations, and map the raw real-time dose data according to the product's transmission location and time parameters to form a real-time dose matrix containing spatial distribution characteristics. Based on the real-time dose matrix, the dose values ​​at each point are compared with the preset dose reference threshold to generate a dose deviation distribution map that includes the magnitude and location of the deviation. Based on the dose deviation distribution map, local areas where the deviation exceeds the preset deviation threshold are identified, and abnormal dose batch information is generated by combining the product batch number. Based on the abnormal dose batch information, the transmission speed parameters, stacking thickness parameters and irradiation power parameters of the corresponding batch of products are extracted to form a set of operating status parameters. Based on the operating status parameter set and dose deviation distribution map, perform graded calibration: Based on the dose deviation distribution map, trend fitting was performed on the deviation values ​​at different points to obtain deviation pattern data describing the directionality and amplitude changes of the deviation, specifically including: Based on the dose deviation distribution map, the deviation values ​​at each point are extracted and sorted according to the acquisition time to obtain the deviation time series; Missing points are filled in the biased time series to form a complete corrected biased series; Based on the corrected bias sequence, multinomial regression, piecewise linear regression, and weighted average fitting based on a sliding window are performed to obtain multiple candidate trend curves. Based on the candidate trend curves, the residual sequence is calculated, and the mean, variance, and maximum deviation of the residuals are statistically analyzed to form a set of residual evaluation indicators. Based on the residual evaluation index set, the candidate trend curve with the smallest residual and the highest stability is selected to form the optimal trend curve; Based on the optimal trend curve, the overall slope, local inflection point position, and fluctuation amplitude are extracted to obtain deviation pattern data that reflects the directionality and amplitude changes of the deviation. Based on the deviation pattern data, the overall direction of change and local concentration features are extracted to form a deviation feature vector for judgment, specifically including: Based on the deviation pattern data, calculate the average deviation value, average rate of change and trend stability index of all points to form a set of indicators related to the overall direction of change. Based on the deviation pattern data, the points are divided into multiple spatial regions, and the local deviation extreme value, local variance and point density index are calculated in each region to form a set of local concentration related indicators. Based on the set of indicators related to the overall direction of change, the average deviation value, average rate of change and trend stability index are weighted and integrated to obtain a single value for the overall direction of change. Based on the set of local concentration-related indicators, the local deviation extreme value, local variance, and point density indicators are weighted and fused to obtain a single local concentration value. The overall direction of change and the local concentration are combined to form the final deviation feature vector; Based on the deviation feature vector, type identification is performed to determine whether the deviation belongs to a linear distribution, a concentrated distribution, or an overall shift, generating a deviation type determination result, specifically including: Based on the overall change direction value and local concentration value in the deviation feature vector, they are compared with the corresponding preset thresholds. When the overall change direction value exceeds the overall threshold, an overall offset candidate result is generated. When the local concentration value exceeds the concentration threshold, a concentrated distribution candidate result is generated. When neither exceeds the threshold, a linear distribution candidate result is generated. The rule confidence index is calculated based on the difference between the value and the threshold, and the confidence index is normalized to obtain the first type of judgment result. The deviation feature vector is input into the trained classification model. The classification model calculates the matching probability for each deviation type based on the learning results of historical deviation samples, calculates the model confidence index based on the difference between the highest probability and the second highest probability, and normalizes the confidence index to obtain the second-class judgment result. When the first type of judgment result is consistent with the second type of judgment result, the result of that type is directly output, and the normalized confidence index of the two is weighted and combined to obtain the final comprehensive confidence score. When the first type of judgment result is inconsistent with the second type of judgment result, the normalized confidence index of the two is compared, the judgment result with higher confidence is selected as the final deviation type judgment result, and the corresponding confidence is output. Based on the deviation type determination result and the operating status parameter set, perform the corresponding calibration process: When the deviation is linearly distributed, the transmission speed is adjusted to generate a speed calibration result; when the deviation is concentrated, the local power is adjusted to generate a power calibration result; when the deviation is overall offset, the irradiation time is adjusted to generate a time calibration result, and the results are unified into a dynamic calibration record. Based on the dynamic calibration record, a calibration verification process is performed: real-time dose raw data is collected again to form a new real-time dose matrix, which is compared with the dose reference threshold to generate a calibration verification result. If there is still a deviation, the graded calibration process is repeated until the deviation is eliminated.

2. The method for real-time calibration and management of electron irradiation dose according to claim 1, characterized in that, The raw real-time dose data is mapped based on transmission location and time parameters to form a real-time dose matrix containing spatial distribution characteristics, including: The raw real-time dose data is sorted chronologically to obtain a dose time series arranged according to the acquisition time. Based on the conveyor belt speed and product positioning marks, the dose time series is matched to obtain dose positioning data containing time and position coordinates; The dose localization data is divided into two-dimensional grids and interpolated to obtain dose grid data that fully covers the transmission path; The dose grid data is smoothed to eliminate random fluctuations and form a real-time dose matrix that reflects the spatial distribution characteristics.

3. The method for real-time calibration and management of electron irradiation dose according to claim 1, characterized in that, Based on the real-time dose matrix, the dose values ​​at each point are compared with preset dose reference thresholds to generate a dose deviation distribution map that includes the magnitude and location of the deviation, including: The difference between the dose value and the dose reference threshold at each point in the real-time dose matrix is ​​calculated to obtain the point deviation data including the magnitude of the difference. Based on the sign and range of the location deviation data, it is classified into ultra-high deviation, ultra-low deviation, and normal range, thus obtaining deviation classification data with classification attributes. Based on the correspondence between deviation classification data and spatial location of points, a deviation label set with coordinate identifiers is generated; By spatially combining and connecting the deviation markers, a dose deviation distribution map that reflects the magnitude and positional relationship of the deviation can be obtained.

4. The method for real-time calibration and management of electron irradiation dose according to claim 3, characterized in that, Based on the dose deviation distribution map, local areas where the deviation exceeds the preset deviation threshold are identified, and abnormal dose batch information is generated by combining the product batch number, including: Based on the dose deviation distribution map, the deviation markers are filtered by threshold to obtain the set of all out-of-range points that exceed the preset deviation threshold range; Clustering is performed based on the spatial proximity of the out-of-poor point set to obtain continuous or adjacent out-of-poor area data; Calculate the boundary range and time interval based on the abnormal area data to obtain the area index information that includes the location coverage and the time period of occurrence; By associating regional index information with production batch numbers, abnormal dose batch information is generated, which includes the regional range, degree of deviation, and batch number.

5. The method for real-time calibration and management of electron irradiation dose according to claim 1, characterized in that, The step of performing corresponding calibration processing based on the deviation type determination result and the operating status parameter set includes: When the deviation type determination result is linear distribution, the correction coefficient is calculated based on the relationship between the linear fitting slope and the transmission speed, and the speed calibration amount is obtained from the preset correction coefficient to generate the corresponding speed calibration result. When the deviation type determination result is a concentrated distribution, the local power adjustment coefficient is calculated based on the extreme value of the dose deviation and the coverage area of ​​the concentrated area, and the power calibration amount is obtained from the preset adjustment coefficient to generate the corresponding power calibration result. When the deviation type determination result is an overall offset, the time extension or shortening is calculated based on the difference between the overall average deviation and the target dose, the time calibration amount is obtained, and the corresponding time calibration result is generated. The speed calibration results, power calibration results, and time calibration results are summarized and corresponding trigger conditions are added to form a dynamic calibration record that includes calibration actions, calibration range, and applicable scope.

6. The method for real-time calibration and management of electron irradiation dose according to claim 1, characterized in that, The calibration verification process includes: Based on the adjusted operating parameters from the dynamic calibration record, the real-time dose data of the product along the transmission path is re-acquired to obtain the calibrated dose data. The calibrated dose data is mapped to generate a new calibrated real-time dose matrix. The new calibrated dose deviation data is obtained by comparing the real-time dose matrix after calibration with the dose reference threshold point by point. Based on the new calibrated dose deviation data, the deviation values ​​of all points are statistically summarized, the overall average residual value and the local maximum residual value are calculated, and the residual statistics are formed. The calibration verification results, which include pass or fail indicators, are generated by comparing the residual statistics with the preset allowable threshold. When the calibration verification result is unqualified, the residual statistics result is compared with the dynamic calibration record to locate the uncorrected deviation pattern, and the process is returned to the graded calibration processing step for targeted adjustment. When the calibration verification result is qualified, the final result after calibration is output and marked as qualified verification information.

7. A real-time calibration and management system for electron irradiation dose, characterized in that, The system, used in the method of any one of claims 1 to 6, comprises: The data acquisition module is used to acquire real-time dose raw data from different locations, and to map the real-time dose raw data according to the transmission location and time parameters to form a real-time dose matrix containing spatial distribution characteristics. The deviation comparison module is used to compare the dose values ​​at each point with the preset dose reference threshold based on the real-time dose matrix, and generate a dose deviation distribution map that includes the magnitude and location of the deviation. The anomaly detection module is used to identify local areas where the deviation exceeds the preset deviation threshold range based on the dose deviation distribution map, and generate abnormal dose batch information by combining the product batch number. The parameter extraction module is used to extract the transmission speed parameters, stacking thickness parameters, and irradiation power parameters of the corresponding batch of products based on the abnormal dose batch information, and form a set of operating status parameters. The graded calibration module is used to perform graded calibration processing based on the set of operating status parameters and the dose deviation distribution map: when the deviation is linearly distributed, the transmission speed is adjusted to generate a speed calibration result; when the deviation is concentrated, the local power is adjusted to generate a power calibration result; when the deviation is generally offset, the irradiation time is adjusted to generate a time calibration result, and the results are unified into a dynamic calibration record. The calibration verification module is used to perform calibration verification processing based on the dynamic calibration record: real-time dose raw data is collected again to form a new real-time dose matrix, which is compared with the dose reference threshold to generate calibration verification results. If there is still a deviation, the graded calibration process is repeated until the deviation is eliminated.

Images