Intelligent sorting control method and system

Abstract

The invention discloses an intelligent sorting control method and system, and belongs to the technical field of article sorting, and the method comprises the following steps: activating a combined sensor group to execute target medicine bottle data acquisition; reading drug data of the cold-chain drug, configuring a microenvironment monitoring network based on the drug data, and outputting a microenvironment anomaly index; the medicine bottles are divided into multiple stages of sorting channels according to the microenvironment anomaly indexes; executing a random sampling instruction, and carrying out random trace nondestructive testing on the target medicine bottle; and in the continuous sorting process, batch abnormity identification is carried out based on the channel binding data set, and sorting isolation management is carried out by utilizing a batch abnormity identification result. The technical problems that in the prior art, during cold chain sorting of biological drugs, the drug microenvironment state monitoring dimensionality is insufficient, anomaly recognition is single, closed-loop management from individuals to batches is lacked, and therefore the sorting accuracy is low, and the quality risk is difficult to control are solved.

Description

Technical Field

[0001] This invention relates to the field of goods sorting technology, specifically to intelligent sorting control methods and systems. Background Technology

[0002] Biopharmaceuticals are temperature-sensitive and susceptible to humidity fluctuations. Cold chain sorting is a core link in their supply chain to ensure efficacy and medication safety, directly determining product quality stability and industry compliance levels. With the continued growth in demand for biopharmaceuticals and the expansion of sorting scale, efficient and precise sorting control has become a key support for the industry's high-quality development. Existing sorting systems mostly use a binary "qualified / unqualified" channel, lacking a review level. Full inspection or destructive testing is inefficient and wasteful of drugs; moreover, there is no batch-level risk association mechanism, making it difficult to detect batch anomalies caused by equipment failure in a timely manner, posing safety hazards. Summary of the Invention

[0003] This application provides an intelligent sorting control method and system, aiming to solve the technical problems in the existing technology of cold chain sorting of biopharmaceuticals, such as insufficient monitoring of the drug microenvironment state, single anomaly identification, and lack of closed-loop management from individual to batch, which leads to low sorting accuracy and difficulty in controlling quality risks.

[0004] In view of the above problems, this application provides an intelligent sorting control method and system.

[0005] The first aspect disclosed in this application provides an intelligent sorting control method, which includes: activating a joint sensor group to collect data from target drug bottles on a cold chain drug sorting line, establishing a status dataset, the status dataset including micro-images of condensation on the bottle surface, infrared thermal distribution images, local humidity at the bottle opening, and gas sensing signals; reading drug data of the cold chain drug, the drug data including drug attribute data and drug packaging data, configuring a microenvironment monitoring network based on the drug data, using the microenvironment monitoring network to perform anomaly identification of the status dataset, and outputting a microenvironment anomaly index; dividing the target drug bottles into multi-level sorting channels according to the microenvironment anomaly index, the multi-level sorting channels including a high-confidence qualified channel, a secondary verification channel, and an isolation channel; executing a random sampling command to perform random micro-quantity non-destructive testing on the target drug bottles while the target drug bottles are being sorted along the corresponding multi-level sorting channels, binding the test results with the corresponding multi-level sorting channels, and establishing a channel binding dataset; and performing batch anomaly identification based on the channel binding dataset during continuous sorting, and using the batch anomaly identification results for sorting and isolation management.

[0006] Another aspect of this application discloses an intelligent sorting control system, which includes: a status dataset establishment module, used to activate a joint sensor group to perform target drug bottle data acquisition and establish a status dataset on a cold chain drug sorting line, the status dataset including micro-images of condensation on the bottle surface, infrared thermal distribution images, local humidity at the bottle opening, and gas sensing signals; and a microenvironment anomaly index output module, used to read drug data of the cold chain drugs, the drug data including drug attribute data and drug packaging data, configure a microenvironment monitoring network based on the drug data, use the microenvironment monitoring network to perform anomaly identification of the status dataset, and output microenvironment anomaly index. The system includes: a constant index; a multi-level sorting channel module, used to divide the target drug bottles into multi-level sorting channels according to the microenvironment anomaly index, the multi-level sorting channels including a high-confidence qualified channel, a secondary verification channel, and an isolation channel; a channel binding dataset establishment module, used to execute random sampling instructions to perform random micro-destructive testing on the target drug bottles when they are sorted along the corresponding multi-level sorting channels, and bind the test results to the corresponding multi-level sorting channels to establish a channel binding dataset; and a batch anomaly identification module, used to identify batch anomalies based on the channel binding dataset during continuous sorting, and to use the batch anomaly identification results for sorting and isolation management.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: By employing a technical solution that collects multi-dimensional status data through a combined sensor array on the cold chain drug sorting line, combines drug data with a micro-environment monitoring network to identify anomalies and sorts drugs through multi-level channels, and supplements this with random sampling detection and batch anomaly identification and isolation management during continuous sorting, this solution solves the technical problems of insufficient monitoring dimensions, single anomaly identification, and lack of closed-loop management from individual to batch in the cold chain sorting of biological drugs. These problems lead to low sorting accuracy and difficulty in controlling quality risks. The solution achieves the technical effect of improving the level of sorting intelligence and ensuring a balance between drug quality stability and sorting efficiency.

[0008] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0009] Figure 1 A flowchart illustrating an intelligent sorting control method is provided for embodiments of this application; Figure 2 This application provides a flowchart illustrating the output of a microenvironment anomaly index in an intelligent sorting control method. Figure 3A schematic diagram of the structure of an intelligent sorting control system is provided for the embodiments of this application.

[0010] Figure labeling: 11 Status dataset creation module, 12 Microenvironment anomaly index output module, 13 Multi-level sorting channel module, 14 Channel binding dataset creation module, 15 Batch anomaly identification module. Detailed Implementation

[0011] The overall concept of the technical solution provided in this application is as follows: This application provides an intelligent sorting control method and system. In a cold chain drug sorting line, a combined sensor group collects multi-dimensional status data of drug bottles. Combined with the drug data, a monitoring network is configured to identify anomalies and output an index. Based on this, the bottles are sorted into multiple channels. Random sampling and detection are performed during sorting, and data is bound to the system. Batch anomalies are identified and managed continuously throughout the process.

[0012] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0013] Example 1 like Figure 1 As shown in the embodiment of this application, an intelligent sorting control method is provided, the method comprising: Step S100: On the cold chain drug sorting line, activate the joint sensor group to perform target drug bottle data acquisition and establish a status dataset, which includes micro-images of condensation on the bottle surface, infrared thermal distribution images, local humidity at the bottle opening, and gas sensing signals.

[0014] Specifically, the combined sensor array is an integrated detection device composed of multiple sensors with different functions, capable of collaboratively acquiring multi-dimensional data. The micro-image of condensation on the bottle surface refers to the distribution and morphology of tiny condensation droplets on the surface of the drug bottle, captured using microscopic imaging technology. The infrared thermal distribution image is a heat map generated using infrared imaging technology, reflecting the temperature distribution on the surface of the drug bottle and displaying local temperature differences. The local humidity at the bottle opening is the real-time humidity value at the sealed area of ​​the drug bottle opening detected by a humidity sensor. The gas sensing signal is the electrical signal of the concentration of characteristic gases related to the drug, such as trace amounts of formaldehyde in vaccines or volatile components in biological agents, detected by a gas sensor.

[0015] Specifically, during the operation of the cold chain drug sorting line, when the target drug vial enters the detection area along the conveyor belt, the combined sensor array is automatically activated. A high-definition microscope camera captures the surface of the vial, generating a micro-image of condensation; an infrared thermal imager scans the vial, generating an infrared thermal distribution image; a humidity sensor collects the local humidity at the vial opening in real time; and a gas sensor array detects the surrounding gas, outputting gas sensing signals. After synchronous processing, the above data is integrated into a status dataset reflecting the condition of the vaccine vial.

[0016] This step achieves multi-dimensional data acquisition through a combined sensor array, providing comprehensive and accurate raw data support for subsequent anomaly identification, and significantly improving the integrity and sensitivity of cold chain drug status monitoring.

[0017] Step S200: Read the drug data of the cold chain drug, the drug data including drug attribute data and drug packaging data, configure a microenvironment monitoring network based on the drug data, use the microenvironment monitoring network to perform anomaly identification of the status dataset, and output the microenvironment anomaly index.

[0018] Specifically, drug property data reflects parameters that indicate the drug's inherent characteristics, including storage temperature range, humidity tolerance threshold, stability period, and characteristic components. Drug packaging data describes the characteristics of drug packaging, including packaging materials, sealing structure, and anti-condensation performance. The microenvironment monitoring network is an intelligent monitoring system based on drug data, composed of algorithmic models and detection logic, which can be tailored to adjust monitoring dimensions, weights, and judgment criteria. The microenvironment anomaly index is used to quantify the degree of anomalies in the drug bottle's microenvironment. It includes multiple functional sub-modules, such as a temperature recognition layer, a micro-pattern recognition layer, and a gas leakage layer, comprehensively reflecting the degree of deviation in dimensions such as temperature, humidity, and sealing.

[0019] Specifically, drug data for cold chain drugs is read through industrial database interfaces or electronic tags attached to drug packaging, such as RFID, to obtain the corresponding attribute and packaging parameters. Then, a parameter configuration engine is invoked to determine the core focus dimensions for microenvironment monitoring based on the drug attribute data. The detection logic is adjusted according to the drug packaging data to complete the customized configuration of the microenvironment monitoring network. The microenvironment monitoring network utilizes deep learning models and multi-dimensional signal analysis algorithms to synchronously analyze the state dataset, comparing the deviation of each monitoring indicator with preset thresholds. A weighted calculation model then integrates the multi-dimensional deviation results to ultimately generate a quantified microenvironment anomaly index, which serves as the core basis for subsequent sorting decisions.

[0020] This step, through the customization of a drug data-driven microenvironment monitoring network, breaks through the limitations of traditional fixed-threshold monitoring systems that apply a one-size-fits-all approach. It enables the monitoring logic to accurately match the characteristic requirements of different drugs, effectively avoiding misjudgments or omissions caused by the mismatch between general standards and individual drug requirements. This improves the accuracy and comprehensiveness of anomaly identification and provides an objective and quantifiable basis for subsequent multi-level sorting channel division.

[0021] Step S300: Based on the microenvironment anomaly index, the target drug bottle is divided into multi-level sorting channels, including a high-confidence qualified channel, a secondary verification channel, and an isolation channel.

[0022] Specifically, multi-level sorting channels are physical or logical pathways divided according to the degree of anomaly in the drug bottles, enabling differentiated processing. The high-confidence qualified channel is for drug bottles with microenvironmental anomaly indices far below the risk threshold; they do not require additional verification and proceed directly to the next stage of circulation. The secondary verification channel is for drug bottles with anomaly indices close to the threshold; they require further precise testing, such as manual verification or deep non-destructive testing, to confirm their status. The isolation channel is for drug bottles with anomaly indices exceeding the threshold; it is used for temporary storage and to initiate quality assessments, preventing abnormal drugs from flowing downstream.

[0023] Specifically, the system acquires the microenvironmental anomaly index of the target drug bottle, calls a pre-set multi-level sorting threshold database, and compares the index with the thresholds of each channel using a threshold matching algorithm to determine the matching target sorting channel. Simultaneously, machine vision positioning technology acquires the real-time position coordinates of the drug bottle on the sorting line, and the PLC sends control commands to the sorting execution mechanism to guide the drug bottle precisely into the matching channel. For different channels, corresponding processing mechanisms are triggered synchronously: high-confidence qualified channels activate fast passage mode, secondary verification channels activate a secondary detection module, and isolated channels activate an environmental locking function and issue an evaluation signal.

[0024] This step enables refined management of drug sorting. The high-confidence qualified channel significantly improves sorting efficiency, the secondary verification channel reduces misjudgments of critically ill drugs through secondary verification, and the isolation channel effectively blocks the circulation risk of abnormal drugs. It balances sorting efficiency and quality control, providing a more reliable safety guarantee for the circulation of biopharmaceuticals.

[0025] Step S400: When the target drug bottle is sorted along the corresponding multi-level sorting channel, execute the random sampling instruction to perform random micro-destructive testing on the target drug bottle, bind the test results to the corresponding multi-level sorting channel, and establish a channel binding dataset.

[0026] Specifically, the random sampling instruction is an instruction generated according to a preset probability rule to randomly select a portion of the target drug bottles for testing, ensuring unbiased sample selection and avoiding insufficient representativeness caused by fixed sampling.

[0027] Specifically, as the target drug bottle is conveyed along the corresponding multi-level sorting channels, a random sampling instruction is generated, containing the sampling probability of each channel. When the drug bottle passes through the detection trigger area within the channel, the machine vision sensor identifies its position and determines whether to trigger detection based on a random number. If triggered, the micro-volume non-destructive testing equipment is activated. After detection, the data processing module binds the detection result with the identifier of the channel to which the drug bottle belongs. Specifically, it first extracts the unique identifier of the detection result, such as the detection timestamp, drug bottle ID, and core data, and simultaneously obtains the physical identifier of the sorting channel to which the drug bottle belongs, such as the channel number. Through a data association algorithm, the identifier information of the two is mapped and matched, embedding metadata such as drug batch and sorting time, forming structured data containing channel, detection, and drug association relationships, which is stored in the channel binding data table of the distributed database, achieving accurate binding of detection results with channels. This data is then used for subsequent batch analysis.

[0028] This step enables dynamic verification of the sorting accuracy of each channel without affecting sorting efficiency and drug integrity.

[0029] Step S500: During continuous sorting, batch anomaly identification is performed based on the channel binding dataset, and the batch anomaly identification results are used for sorting isolation management.

[0030] Specifically, the channel-bound dataset refers to the data set that stores the random sampling and testing results of each sorting channel, including channel identifiers, drug batch information, testing indicators, and anomaly judgment results.

[0031] During continuous sorting, defect records from each stage are aggregated through a real-time data interface to form a sorting defect set. A first identification anomaly and a second identification anomaly are established, and a joint anomaly analysis is performed on the first and second identification anomalies. If the comprehensive score exceeds a threshold, the batch anomaly identification result is output. The location of all drugs in the batch is located via RFID traceability, the PLC is controlled to pause the batch sorting process, and a robotic arm equipped with a force control sensor is instructed to transfer the drugs to a temperature-controlled isolation area.

[0032] This step overcomes the limitation that individual testing is insufficient to detect batch risks and strengthens the overall control capabilities of the distribution of biopharmaceuticals.

[0033] Furthermore, such as Figure 2 As shown, the microenvironment monitoring network is used to identify anomalies in the execution status dataset and outputs a microenvironment anomaly index, including: Activate the temperature recognition layer of the microenvironment monitoring network, and use the temperature recognition layer to extract features from the state dataset to establish a temperature feature set. The temperature feature set includes the maximum temperature difference on the bottle surface, the minimum temperature difference on the bottle surface, temperature gradient features, hotspot coordinate stability, and local temperature rise rate. The temperature feature set is used to identify temperature anomalies under five-dimensional features, and a temperature anomaly index is established. Activate the micro-texture pattern recognition layer of the microenvironment monitoring network, and use the micro-texture pattern recognition layer to extract the condensation droplet distribution density features, droplet size distribution features, crystal state features, and surface reflectivity features from the state dataset to establish a condensation feature set; Surface condensation anomalies are identified based on the condensation feature set, and a condensation anomaly index is established. Anomaly identification is performed based on the temperature anomaly index and condensation anomaly index, and a microenvironment anomaly index is output.

[0034] Specifically, the temperature recognition layer refers to a submodule in the microenvironment monitoring network that specifically processes temperature-related data, responsible for extracting temperature features from the state dataset and identifying temperature anomalies. The micro-ripple pattern recognition layer refers to a submodule in the microenvironment monitoring network that specifically processes condensation-related data, responsible for extracting condensation features from the state dataset and identifying condensation anomalies.

[0035] Specifically, drug attribute data and packaging data are integrated to form functional layers, including a temperature recognition layer and a micro-texture pattern recognition layer. For each layer, feature extraction rules are defined based on drug characteristics, and anomaly detection thresholds are set in conjunction with drug standards. By integrating deep learning models and multi-dimensional algorithms, anomaly detection logic for each layer is constructed. Then, a weighted fusion algorithm is used to comprehensively calculate the multi-dimensional anomaly index. Simultaneously, a dynamic optimization module is embedded to adjust feature weights and thresholds in real time based on actual detection feedback, forming a microenvironment monitoring network adapted to specific drugs.

[0036] The temperature recognition layer takes infrared thermal distribution images as input, uses a CNN model to extract temperature field features on the bottle surface, calculates the maximum / minimum temperature difference using a temperature field segmentation algorithm, generates temperature gradient features using a gradient operator, records hotspot coordinate stability using a time-series tracking algorithm, and calculates the local temperature rise rate using a slope model. Based on temperature standards in the drug property data, an SVM classifier is trained to establish a five-dimensional anomaly judgment logic, dynamically adjusting feature weights and thresholds. The micro-texture pattern recognition layer takes condensation micro-images as input, uses a U-Net model to segment water droplet regions, obtains distribution density using a density statistics algorithm, analyzes water droplet size distribution using K-means clustering, determines crystal state using a morphology recognition model, and extracts reflection features using a light reflectance algorithm. Based on anti-condensation parameters in the packaging data, a rule base is constructed to generate condensation anomaly judgment logic and optimize feature recognition thresholds.

[0037] The temperature recognition layer of the microenvironment monitoring network is activated, and infrared image processing algorithms are used to analyze the infrared thermal distribution images in the state dataset. This includes calculating the maximum and minimum temperature differences on the bottle surface using a temperature field segmentation algorithm; extracting temperature gradient features using a gradient operator; recording the stability of hotspot coordinates using a time-series tracking algorithm (instability is determined if a hotspot drifts from the bottom to the top of the bottle within 30 seconds); and deriving the local temperature rise rate using a slope calculation model. These indicators are then integrated into a temperature feature set. Subsequently, a multi-dimensional anomaly detection model is used to perform a five-dimensional joint analysis of the temperature feature set, comparing it with the temperature standard in the drug data to output a temperature anomaly index. Next, the micro-texture pattern recognition layer is activated, and microscopic image analysis tools are used to process the condensation micro-images on the bottle surface in the state dataset. This involves obtaining the condensation droplet distribution density using a density statistics algorithm; analyzing the droplet size distribution characteristics using a size clustering algorithm; determining the crystal state using a morphology recognition model; and calculating surface reflectivity features using a light reflectivity detection module, integrating these into a condensation feature set. Finally, a condensation anomaly detection rule base is used to identify surface condensation anomalies, outputting a condensation anomaly degree index.

[0038] This step overcomes the limitations of traditional single temperature or humidity detection by extracting and jointly analyzing refined features from both temperature and condensation dimensions: the five-dimensional features of the temperature recognition layer comprehensively capture the static differences, dynamic changes, and spatial distribution of temperature, avoiding missed detections due to local temperature anomalies being masked by the overall mean; the condensation features of the micro-texture pattern recognition layer accurately depict the details of the condensation state, and the weighted fusion of the two-dimensional indices further improves the comprehensiveness and accuracy of anomaly identification, providing a more reliable quantitative basis for subsequent grading and sorting.

[0039] Furthermore, based on the temperature anomaly index and condensation anomaly index, anomaly identification is performed, and a microenvironment anomaly index is output, including: Activate the gas leakage layer of the microenvironment monitoring network, and use the gas leakage layer to analyze the instantaneous humidity pulse, humidity offset, and characteristic gas concentration in the status dataset to establish a leakage anomaly index; Anomalies are fused based on the temperature anomaly index, condensation anomaly index, and leakage anomaly index to output a microenvironment anomaly index.

[0040] Specifically, the gas leakage layer refers to a functional submodule within the microenvironment monitoring network that specifically detects the sealing of drug vials. It determines the presence of leaks by analyzing humidity and gas signals. A transient humidity pulse refers to a sudden peak in local humidity at the vial opening within a short period, such as a rapid increase from the normal 30% RH to 60% RH followed by a rapid drop, typically indicating a transient leak. Characteristic gas concentration refers to the concentration of volatile components specific to the drug, detected by gas sensors; an abnormally high concentration indicates a sealing failure.

[0041] Specifically, to construct a gas leakage layer, the humidity signal at the bottle opening and the gas sensor signal from the state dataset are first accessed. A wavelet transform algorithm is used to extract instantaneous humidity pulse features, such as pulse amplitude and duration. Humidity offsets are captured through sliding window trend analysis. A partial least squares regression model is used to analyze the characteristic gas concentration signal and convert it into standardized concentration values. Combining the sealing parameters from the drug packaging data, a three-level risk mapping rule is established, and a logistic regression model is trained to generate a leakage anomaly index.

[0042] The gas leakage layer of the microenvironment monitoring network is activated, and signal processing algorithms are used to analyze the bottle opening humidity signal in the state dataset: instantaneous humidity pulses are identified, and humidity shifts are captured through trend analysis models; simultaneously, characteristic gas concentration signals are processed using gas sensor data analysis models. Based on these analyses, a risk level mapping algorithm is used to convert humidity pulse amplitude, shift, and gas concentration exceedance values ​​into risk scores, establishing a leakage anomaly index.

[0043] This step, by adding a gas leakage layer, compensates for the shortcomings of temperature and condensation dimensions in monitoring sealing performance, improves the detection rate of hidden risks such as drug packaging damage and sealing failure, and provides more comprehensive protection for drug quality and safety.

[0044] Furthermore, based on the aforementioned temperature anomaly index, condensation anomaly index, and leakage anomaly index, anomalies are fused to output a microenvironment anomaly index, including: Obtain a preset anomaly weighting factor, and use the preset anomaly weighting factor to perform a weighted calculation of the temperature anomaly index, condensation anomaly index, and leakage anomaly index to generate an initial anomaly index; Anomalies were traced to their respective sources for the temperature anomaly index, condensation anomaly index, and leakage anomaly index. The anomaly tracing results were used for source tracing interaction authentication, and an interaction enhancement factor was established. After enhancing the initial anomaly index using the aforementioned interaction enhancement factor, the microenvironment anomaly index is output.

[0045] Specifically, the preset anomaly weighting factor is a coefficient pre-set based on drug characteristics, such as storage requirements and packaging type, used to measure the impact of temperature anomaly index, condensation anomaly index, and leakage anomaly index on the overall microenvironment. The traceability interaction authentication is used to determine whether there is a correlation between the traceability results of anomalies in different dimensions, such as the verification process where both temperature anomalies and leakage anomalies are caused by the same equipment failure. The interaction enhancement factor is a coefficient set based on the traceability interaction authentication results; if anomalies are correlated, the coefficient is increased; otherwise, the base value is maintained, used to enhance the accuracy of the initial anomaly index.

[0046] Specifically, a preset anomaly weighting factor is obtained from the drug database, and a weighted summation algorithm is called to substitute the three-dimensional anomaly indices into the calculation to generate an initial anomaly index. Next, a fault tree analysis tool is used to trace the anomalies of each index. For example, temperature anomalies are traced to a malfunctioning temperature controller on sorting line 3, leakage anomalies to minor damage to the packaging rubber stopper, and condensation anomalies to humidity fluctuations in the sorting room. Then, an association rule algorithm is used to perform source tracing interaction authentication, revealing that temperature anomalies and leakage anomalies are not directly related, while condensation anomalies and leakage anomalies are both affected by humidity in the sorting room (a weak correlation exists). Based on this, an interaction enhancement factor is established. Finally, an index enhancement algorithm is called to multiply the initial anomaly index by the interaction enhancement factor, outputting the final microenvironment anomaly index.

[0047] This step overcomes the shortcomings of traditional single-weighted methods that ignore the correlation of anomalies by employing a three-order fusion logic of weighted calculation, traceability authentication, and index enhancement. Compared to simple weighting, this method improves the accuracy of the microenvironment anomaly index, providing a more realistic risk-based decision-making basis for subsequent hierarchical sorting.

[0048] Furthermore, based on the microenvironment anomaly index, the target drug vials are divided into multi-level sorting channels, including: Obtain the threshold space corresponding to the multi-level sorting channel, perform matching analysis based on the microenvironment anomaly index and the threshold space, and establish a mapping result; Determine whether the mapping result is consistent with the current sorting flow of the target drug bottle; If the mapping result is inconsistent with the current sorting flow, a sorting transfer instruction is generated; After locating the target drug bottle using the sorting and transfer command, interception and grabbing are performed, and the target drug bottle is transferred to the sorting channel corresponding to the mapping result.

[0049] Specifically, the threshold space refers to the set of microenvironmental anomaly index intervals corresponding to the high-confidence qualified channel, the secondary review channel, and the isolation channel in the multi-level sorting channel. It is the standard basis for determining the channel to which the drug bottle belongs.

[0050] Specifically, the threshold space corresponding to the multi-level sorting channels is retrieved from the sorting control database. An interval matching algorithm, such as a fast interval positioning model based on the bisection method, is called to compare the microenvironmental anomaly index of the target medicine bottle with the threshold space to obtain the mapping result to which the medicine bottle should belong. Then, the current sorting direction of the medicine bottle is obtained in real time through the dynamic positioning system of the sorting line, and the mapping result is checked with the current direction by the logic judgment module. If they are inconsistent, the system generates a sorting transfer instruction through the PLC. The instruction contains the real-time coordinates of the medicine bottle, the target channel number, and the transfer timing parameters. Then, the dynamic positioning system calculates the movement trajectory of the medicine bottle and completes precise positioning by combining the sorting line conveying speed fed back by the encoder. Subsequently, the instruction interception execution mechanism executes the interception and grabbing, and finally transfers the medicine bottle to the secondary verification channel corresponding to the mapping result.

[0051] This step overcomes the limitations of static and unadjustable sorting allocation, enabling real-time correction of initial sorting deviations and effectively preventing abnormal drugs from entering the qualified channel. This ensures the safe circulation of biological drugs and avoids resource waste caused by misallocation.

[0052] Furthermore, interception and capture are performed, including: Based on the positioning results and the conveying speed of the cold chain drug sorting line, the temporal position of the target drug bottle is fitted, and the temporal position fitting result is established. The timing and position fitting results are used to optimize the motion control of the interception component and establish interception operation commands. The interception operation command is used to control the interception component to intercept the target drug bottle. When the interception component completes the interception operation command action and the interception contact surface trigger signal of the interception component meets the preset threshold, the interception and capture are performed.

[0053] Specifically, the positioning results of the target medicine bottle are acquired through a machine vision system, such as a 2D vision camera. Simultaneously, the conveyor speed of the sorting line is collected via an encoder. A Kalman filter algorithm is used to fit the temporal position, establishing a fitting result. Based on this result, a model predictive control algorithm is used to optimize the motion control of the interception component: optimizing the robotic arm's start-up time, motion trajectory, end effector speed, and contact force parameters to generate interception operation commands. Then, the PLC sends the commands to the robotic arm controller, controlling the robotic arm to move along a preset trajectory. When the robotic arm's end effector contacts the medicine bottle, a pressure sensor provides real-time feedback of a trigger signal. Once the signal is within a preset threshold range and contact is confirmed to be safe and reliable, the grasping action is triggered, completing the interception and grasping process.

[0054] This step ensures the reliable transfer of abnormal drug bottles and avoids damage to drug packaging caused by mechanical operation, providing key support for the efficiency and safety of cold chain drug sorting.

[0055] Furthermore, execute random sampling instructions, including: Obtain the random probability factor for each sorting channel in the multi-level sorting channel; When a sorting individual passes through each sorting channel, the sorting individual is identified by probability sampling based on the random probability factor. If the probability sampling identification result is the sampled result, then a micro-quantity non-destructive test is performed on the corresponding sorted individual. The micro-quantity non-destructive test includes ultrasonic testing and optical scattering testing.

[0056] Specifically, the random probability factor refers to a value set separately for each channel in a multi-level sorting channel to determine the probability of the medicine bottle being sampled in that channel. It is usually expressed as a percentage. The probability of different channels is set according to the difference in risk level, such as higher probability in high-risk channels.

[0057] Specifically, the random probability factors for each sorting channel are retrieved from the sorting control parameter library. When a sorted item is conveyed along its channel, it is identified by a real-time positioning sensor, such as an infrared beam sensor, as entering the sampling detection area. At the same time, an encrypted random number generator, such as one based on the SHA-256 random algorithm, is invoked to generate a random number. This random number is then compared with the random probability factor of the corresponding channel to complete the probability sampling identification. If the result is that the item has been sampled, the micro-destructive testing process is triggered: the ultrasonic detection module is controlled to emit sound waves that penetrate the bottle, receive the reflected signal, and extract features through Fourier transform. Simultaneously, the optical scattering detection module is activated to collect the scattered light intensity distribution data and analyze it through the Mie scattering model. Finally, the two detection results are integrated as the sampling data for the sorted item.

[0058] This step ensures sorting efficiency while enabling dynamic verification of the sorting accuracy of each channel: low-probability sampling in high-confidence channels reduces invalid detections, while high-probability sampling in secondary verification and isolation channels strengthens risk monitoring; the combination of the two non-destructive testing technologies avoids drug waste caused by destructive testing and improves the reliability of sampling results through multi-physics field signals.

[0059] Furthermore, during continuous sorting, batch anomaly identification is performed based on the channel-bound dataset, including: During continuous sorting, obtain the set of sorting defects for continuous sorting; Based on the sorting defect set, perform continuous occurrence anomaly identification of the same type of defect, and establish a first identification anomaly; A second identification anomaly is established based on the channel binding dataset; Based on the first identified anomaly and the second identified anomaly, a joint anomaly analysis is performed, and the batch anomaly identification results are output.

[0060] Specifically, the sorting defect set refers to the collection of defect records identified at each stage of the continuous sorting process for all target drug bottles, including information such as defect type, occurrence time, batch number, and channel. The first identification anomaly refers to the preliminary abnormal result obtained through the continuous occurrence of similar defects, suggesting the possible existence of batch-related defects. The second identification anomaly is based on the channel-bound dataset, obtained by analyzing common deviations in detection indicators of the same batch of drugs.

[0061] Specifically, during continuous sorting, a real-time data acquisition module aggregates defect records from each channel to form a sorting defect set, which is stored in a time-series database. This set includes the type, timestamp, and batch number of each defect. A sliding window algorithm is used to scan the sorting defect set, counting the number of consecutive occurrences of the same type of defect within the same batch. This count is compared to a preset threshold; if the threshold is exceeded, it is marked as the first identified anomaly. Simultaneously, random sampling detection results for that batch are extracted from the channel-bound dataset. The distribution of detection indicators is analyzed using the 3σ principle, and these are marked as the second identified anomaly. Subsequently, a weighted voting model is used for joint anomaly analysis. If the overall score exceeds the batch anomaly threshold, the batch anomaly identification result is output. The weighted voting model is constructed by first determining the weights of the first and second identified anomalies based on historical anomaly datasets using the analytic hierarchy process (AHP). Known batch anomaly cases are collected, and the two types of identification results are converted into 0-1 scores, where 1 represents anomaly and 0 represents normal. The weighted score is calculated using the formula: "Overall score = First anomaly score × weight + Second anomaly score × weight". The model is validated using a confusion matrix, and the weights are optimized to maximize accuracy. Finally, a threshold judgment is embedded, and the weights are dynamically updated based on new data.

[0062] This step overcomes the limitations of single-dimensional batch anomaly identification: firstly, it identifies continuous defect trends during dynamic sorting, avoiding the missed detection of "low-probability, high-frequency" batch problems; secondly, it provides objective verification through sampled data, reducing misjudgments caused by chance factors. This significantly reduces the probability of large-scale drug quality problems.

[0063] Furthermore, the results of batch anomaly identification are used for sorting and isolation management, including: If the batch anomaly identification result is a genuine batch anomaly, a batch anomaly warning will be generated and the batch anomaly warning will be issued. Activate the batch isolation channel and use it to sort and isolate the current batch of cold chain drugs.

[0064] Specifically, if the batch anomaly identification result is verified as a genuine batch anomaly, the early warning management module is invoked to generate a batch anomaly warning. This triggers a continuous beeping and red light flashing from the sorting line's audible and visual alarm, while simultaneously sending a warning message containing the batch number, anomaly type, and risk level to the quality management system. The batch isolation channel is activated, and the real-time location of all drugs in the batch is located via RFID traceability. A control command is sent to the PLC: suspend the regular sorting process for this batch of drugs, and instruct the robotic arms in each channel to transfer the batch of drugs to the batch isolation channel according to the coordinates fed back by the positioning system. The temperature and humidity control system within the channel automatically starts, and the RFID reader records the warehousing information for each bottle of drug, forming a "batch-location-time" traceability chain, thus completing the sorting and isolation management.

[0065] This step, through precise early warning and closed-loop management of the entire batch, prevents secondary quality deterioration of abnormal drugs due to changes in storage conditions. The traceability chain provides complete data support for subsequent cause analysis. This reduces the rate of quality risk diffusion caused by batch anomalies and significantly improves the quality and safety control level of cold chain drugs.

[0066] In summary, the intelligent sorting control method provided in this application has the following technical effects: 1. By collecting key status data through joint sensing and configuring a monitoring network based on drug characteristics, graded sorting and random verification are achieved, forming a closed-loop control from individual to batch, which improves the intelligence level of cold chain drug sorting and ensures a balance between drug quality stability and sorting efficiency.

[0067] 2. By extracting and fusing multi-dimensional features, combined with anomaly tracing and interactive analysis, the accuracy of microenvironment anomaly judgment is enhanced. It can comprehensively capture potential risks such as temperature, condensation, and leakage, reduce misjudgment based on a single indicator, and provide a more reliable basis for subsequent sorting decisions.

[0068] 3. By identifying the continuous occurrence of similar defects and deviations in the channel system, batch risks can be detected in a timely manner. Combined with rapid early warning and isolation mechanisms, abnormal spread can be effectively prevented, the overall control of biopharmaceutical batch quality can be strengthened, and the reliability of the quality assurance system can be improved.

[0069] Example 2 Based on the same inventive concept as the intelligent sorting control method in the foregoing embodiments, such as Figure 3 As shown in the embodiment of this application, an intelligent sorting control system is provided, which includes: The status dataset establishment module 11 is used to activate the joint sensor group to perform target drug bottle data acquisition and establish a status dataset on the cold chain drug sorting line. The status dataset includes micro-images of condensation on the bottle surface, infrared thermal distribution images, local humidity at the bottle opening, and gas sensing signals. The microenvironment anomaly index output module 12 is used to read drug data of cold chain drugs, including drug attribute data and drug packaging data, configure a microenvironment monitoring network based on the drug data, use the microenvironment monitoring network to perform anomaly identification of the status dataset, and output a microenvironment anomaly index. The multi-level sorting channel module 13 is used for... The target drug bottles are divided into multi-level sorting channels based on the microenvironment anomaly index. The multi-level sorting channels include a high-confidence qualified channel, a secondary verification channel, and an isolation channel. The channel binding dataset establishment module 14 is used to execute random sampling instructions to perform random micro-destructive testing on the target drug bottles when they are sorted along the corresponding multi-level sorting channels, and bind the test results to the corresponding multi-level sorting channels to establish a channel binding dataset. The batch anomaly identification module 15 is used to identify batch anomalies based on the channel binding dataset during continuous sorting and to use the batch anomaly identification results for sorting and isolation management.

[0070] Furthermore, the microenvironment anomaly index output module 12 is also used to perform the following steps: activating the temperature recognition layer of the microenvironment monitoring network, using the temperature recognition layer to extract features from the state dataset, and establishing a temperature feature set, the temperature feature set including the maximum temperature difference on the bottle surface, the minimum temperature difference on the bottle surface, temperature gradient features, hotspot coordinate stability, and local temperature rise rate; using the temperature feature set to perform temperature anomaly identification under five-dimensional features, and establishing a temperature anomaly index; activating the micro-texture pattern recognition layer of the microenvironment monitoring network, using the micro-texture pattern recognition layer to extract the condensation droplet distribution density features, droplet size distribution features, crystal state features, and surface reflectivity features from the state dataset, and establishing a condensation feature set; performing surface condensation anomaly identification based on the condensation feature set, and establishing a condensation anomaly degree index; performing anomaly identification according to the temperature anomaly index and the condensation anomaly degree index, and outputting the microenvironment anomaly index.

[0071] Furthermore, the microenvironment anomaly index output module 12 is also used to perform the following steps: activate the gas leakage layer of the microenvironment monitoring network, analyze the instantaneous humidity pulse, humidity offset, and characteristic gas concentration in the state dataset using the gas leakage layer, and establish a leakage anomaly index; perform anomaly fusion based on the temperature anomaly index, condensation anomaly index, and leakage anomaly index, and output the microenvironment anomaly index.

[0072] Furthermore, the microenvironment anomaly index output module 12 is also used to perform the following steps: obtain a preset anomaly weight factor, use the preset anomaly weight factor to perform weighted calculation of temperature anomaly index, condensation anomaly index, and leakage anomaly index to generate an initial anomaly index; perform anomaly tracing on the temperature anomaly index, condensation anomaly index, and leakage anomaly index respectively, use the anomaly tracing results to perform tracing interaction authentication, and establish an interaction enhancement factor; after enhancing the initial anomaly index using the interaction enhancement factor, output the microenvironment anomaly index.

[0073] Furthermore, the multi-level sorting channel module 13 is also used to perform the following steps: obtain the threshold space corresponding to the multi-level sorting channel, perform matching analysis based on the microenvironmental anomaly index and the threshold space, and establish a mapping result; determine whether the mapping result is consistent with the current sorting flow of the target drug bottle; if the mapping result is inconsistent with the current sorting flow, generate a sorting transfer instruction; after locating the target drug bottle using the sorting transfer instruction, perform interception and grabbing, and transfer the target drug bottle to the sorting channel corresponding to the mapping result.

[0074] Furthermore, the multi-level sorting channel module 13 is also used to perform the following steps: fitting the time-series position of the target drug bottle according to the positioning result and the conveying speed of the cold chain drug sorting line, and establishing a time-series position fitting result; using the time-series position fitting result to optimize the action control of the interception component, and establishing an interception operation command; using the interception operation command to control the interception component to intercept the target drug bottle, and when the interception component completes the interception operation command action, and the interception contact surface trigger signal of the interception component meets the preset threshold, the interception and grabbing is performed.

[0075] Furthermore, the channel binding dataset establishment module 14 is also used to perform the following steps: obtain the random probability factor of each sorting channel in the multi-level sorting channel; when a sorting individual passes through each sorting channel, perform probability sampling identification of the sorting individual according to the random probability factor; if the probability sampling identification result is the sampled result, then perform micro-quantity non-destructive testing on the corresponding sorting individual, wherein the micro-quantity non-destructive testing includes ultrasonic testing and optical scattering testing.

[0076] Furthermore, the batch anomaly identification module 15 is also used to perform the following steps: during continuous sorting, acquiring a set of sorting defects for continuous sorting; identifying the continuous occurrence of the same type of defects based on the set of sorting defects, and establishing a first identification anomaly; establishing a second identification anomaly based on the channel binding dataset; performing joint anomaly analysis based on the first identification anomaly and the second identification anomaly, and outputting the batch anomaly identification result.

[0077] Furthermore, the batch anomaly identification module 15 is also used to perform the following steps: if the batch anomaly identification result is a real batch anomaly, then generate a batch anomaly warning and execute a batch anomaly warning; activate the batch isolation channel and use the batch isolation channel to sort and isolate the current batch of cold chain drugs.

[0078] In summary, any step of the method described above can be stored as a computer instruction or program in an unrestricted computer memory, and can be called and identified by an unrestricted computer processor to implement any method in the embodiments of this application, without any additional restrictions.

[0079] Furthermore, the "first" or "second" mentioned above not only represents a sequential relationship but also a specific concept and / or refers to the possibility of selecting individual or all of multiple elements. Clearly, those skilled in the art can make various modifications and variations to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method of intelligent sorting control, characterized in that, The method comprises: At the cold-chain medicine sorting line, activate the combined sensor group to perform target medicine bottle data acquisition, and establish a state data set, the state data set comprising condensation micro-image of bottle surface, infrared thermal distribution image, local humidity of bottle mouth and gas sensing signal; Read medicine data of the cold-chain medicine, the medicine data comprising medicine attribute data and medicine packaging data, configure a micro-environment monitoring network based on the medicine data, perform abnormal identification of the state data set by using the micro-environment monitoring network, and output a micro-environment abnormality index; According to the micro-environment abnormality index, divide the target medicine bottle into a plurality of sorting channels, the plurality of sorting channels comprising a high-confidence qualified channel, a secondary review channel and an isolation channel; When the target medicine bottle is sorted along the corresponding plurality of sorting channels, execute a random sampling instruction, perform random micro-lossless detection of the target medicine bottle, bind the detection result with the corresponding plurality of sorting channels, and establish a channel binding data set; In the continuous sorting process, perform batch abnormality identification based on the channel binding data set, and perform sorting isolation management by using the batch abnormality identification result.

2. The intelligent sorting control method of claim 1, wherein, Performing abnormal identification of the state data set by using the micro-environment monitoring network and outputting a micro-environment abnormality index comprises: Activate a temperature identification layer of the micro-environment monitoring network, perform feature extraction on the state data set by using the temperature identification layer, establish a temperature feature set, and the temperature feature set comprises maximum temperature difference of bottle surface, minimum temperature difference of bottle surface, temperature gradient feature, hotspot coordinate stability and local temperature rise rate; Perform temperature abnormality identification under five-dimensional features by using the temperature feature set, and establish a temperature abnormality index; Activate a micro-line pattern identification layer of the micro-environment monitoring network, extract condensate drop distribution density feature, water drop size distribution feature, crystal state feature and surface reflectivity feature in the state data set by using the micro-line pattern identification layer, and establish a condensation feature set; Perform surface condensation abnormality identification based on the condensation feature set, and establish a condensation abnormality degree index; Perform abnormal identification according to the temperature abnormality index and the condensation abnormality degree index, and output a micro-environment abnormality index.

3. The intelligent sorting control method of claim 2, wherein, Performing abnormal identification according to the temperature abnormality index and the condensation abnormality degree index and outputting a micro-environment abnormality index comprises: Activate a gas leakage layer of the micro-environment monitoring network, analyze instantaneous humidity pulse, humidity offset and characteristic gas concentration in the state data set by using the gas leakage layer, and establish a leakage abnormality index; Perform abnormal fusion according to the temperature abnormality index, the condensation abnormality degree index and the leakage abnormality index, and output a micro-environment abnormality index.

4. The intelligent sorting control method of claim 3, wherein, Performing abnormal fusion according to the temperature abnormality index, the condensation abnormality degree index and the leakage abnormality index and outputting a micro-environment abnormality index comprises: Obtain a preset abnormality weight factor, perform weighted calculation of the temperature abnormality index, the condensation abnormality degree index and the leakage abnormality index by using the preset abnormality weight factor, and generate an initial abnormality index; Perform abnormality tracing on the temperature abnormality index, the condensation abnormality degree index and the leakage abnormality index respectively, perform trace interaction authentication by using the abnormality tracing result, and establish an interaction enhancement factor; After initial abnormal index enhancement is performed by using the interaction enhancement factor, an environment abnormal index of the micro-environment is output.

5. The intelligent sorting control method of claim 1, wherein, The target medicine bottle is divided into a plurality of sorting channels according to the micro-environment abnormal index, including: A threshold space corresponding to the plurality of sorting channels is obtained, and matching analysis is performed according to the micro-environment abnormal index and the threshold space to establish a mapping result; It is judged whether the mapping result is consistent with a current sorting flow direction of the target medicine bottle; If the mapping result is not consistent with the current sorting flow direction, a sorting transfer instruction is generated; After the target medicine bottle is positioned by using the sorting transfer instruction, interception and grabbing are performed, and the target medicine bottle is transferred to a sorting channel corresponding to the mapping result.

6. The intelligent sorting control method of claim 5, wherein, The interception and grabbing are performed, including: Timing position fitting of the target medicine bottle is performed according to the positioning result and a conveying speed of the cold-chain medicine sorting line to establish a timing position fitting result; Action control optimization of an interception assembly is performed by using the timing position fitting result to establish an interception operation command; The interception operation command is used to control the interception assembly to perform interception of the target medicine bottle, and when the interception assembly completes the action of the interception operation command and an interception contact surface trigger signal of the interception assembly meets a preset threshold, the interception and grabbing are performed.

7. The intelligent sorting control method of claim 1, wherein, The random sampling instruction is executed, including: A random probability factor of each sorting channel in the plurality of sorting channels is obtained; When a sorting individual passes through each sorting channel, probability sampling identification of the sorting individual is performed according to the random probability factor; If the probability sampling identification result is a sampled result, micro-loss detection of the corresponding sorting individual is performed, and the micro-loss detection includes ultrasonic detection and optical scattering detection.

8. The intelligent sorting control method of claim 1, wherein, In a continuous sorting process, batch abnormal identification is performed based on a channel binding data set, including: In the continuous sorting process, a sorting defect set of the continuous sorting is obtained; Continuous occurrence abnormal identification of the same type of defect is performed according to the sorting defect set to establish a first identification abnormality; A second identification abnormality is established according to the channel binding data set; Joint abnormal analysis is performed based on the first identification abnormality and the second identification abnormality to output a batch abnormal identification result.

9. The intelligent sorting control method of claim 1, wherein, Sorting isolation management is performed by using the batch abnormal identification result, including: If the batch abnormal identification result is a real batch abnormality, a batch abnormality warning is generated, and a batch abnormality warning report is executed; A batch isolation channel is activated, and the current batch of cold-chain medicines is sorted and managed by using the batch isolation channel.

10. An intelligent sorting control system characterized by, The system is used to execute the intelligent sorting control method in any one of claims 1 to 9, and the system includes: A state data set establishment module is used to activate a joint sensor group to perform target medicine bottle data acquisition on a cold-chain medicine sorting line to establish a state data set, and the state data set includes a condensation micro-image of a bottle surface, an infrared thermal distribution image, a local humidity of a bottle mouth, and a gas sensing signal; A micro-environment abnormal index output module is used to read medicine data of cold-chain medicines, the medicine data includes medicine attribute data and medicine packaging data, a micro-environment monitoring network is configured based on the medicine data, abnormal identification of the state data set is performed by using the micro-environment monitoring network, and a micro-environment abnormal index is output. A multi-stage sorting channel module is used to divide the target medicine bottle into multi-stage sorting channels according to the micro-environment anomaly index, and the multi-stage sorting channels include a high-confidence qualified channel, a secondary review channel, and an isolation channel; A channel binding dataset establishment module is used to execute random sampling instructions when the target medicine bottle is sorted along the corresponding multi-stage sorting channel, to perform random micro-lossless detection of the target medicine bottle, to bind the detection result with the corresponding multi-stage sorting channel, and to establish a channel binding dataset; A batch anomaly identification module is used to identify batch anomalies based on the channel binding dataset during continuous sorting, and to use the batch anomaly identification result for sorting isolation management.

Images