Isolation process data traceability method and system for radiopharmaceutical preparation

Abstract

This invention discloses a method and system for tracing data in the separation process of radiopharmaceutical preparation, including: data reliability processing: performing anti-interference repair and verification on the collected data, and outputting repaired and verified data; correlation map construction: performing correlation analysis on multi-source heterogeneous data in the separation process, generating a correlation map defined with data entities as nodes; process deduction and inverse inference: solving the symbolic differential equations describing the dynamics of the separation process, and outputting deduction data and parameter deviation information; intelligent traceability application: based on the deduction data and parameter deviation information, performing data retrieval and attribution analysis on the separation process, and outputting full-link traceability results. This invention transforms the traditional discrete, passive, and manual-dependent data monitoring and post-analysis mode into an active, continuous, and automated intelligent analysis paradigm. The system can autonomously complete the repair of original data and the correlation of multi-source heterogeneous data.

Description

Technical Field

[0001] This invention belongs to the field of radiopharmaceutical preparation technology, and particularly relates to a method and system for tracing data of the separation process used in radiopharmaceutical preparation. Background Technology

[0002] In the field of radiopharmaceutical preparation, especially in critical separation and purification processes, comprehensive monitoring and precise traceability of process data are crucial for ensuring drug quality and meeting stringent regulatory requirements. Currently, production process data management in this field generally relies on a combined technical architecture. Typically, programmable logic controllers (PLCs) and Supervisory Control and Data Acquisition (SCADA) systems are used to collect and monitor key process parameters such as temperature, pressure, flow rate, and radioactivity in real time, enabling alarms for exceeding limits and basic data storage. Production process data is usually archived in a real-time historical database, while quality data such as raw material properties, intermediate testing, and final product release are recorded in a Laboratory Information Management System (LIMS). When quality deviations occur or process analysis is required, analysts need to manually export massive amounts of data from these separate systems and then use spreadsheets or commercial analytics software for time-consuming post-processing data alignment, filtering, and correlation analysis. The depth of this analysis largely depends on the engineer's personal experience and understanding of the process mechanism.

[0003] However, the aforementioned existing technologies face a series of inherent limitations in practical applications, making it difficult to meet the needs of intelligent and refined traceability. First, the systems form "data silos," lacking effective automated correlation mechanisms, leading to reliance on manual connection during the traceability process, resulting in low efficiency and a high risk of errors, failing to achieve the end-to-end automated intelligent traceability claimed in the claims. Second, existing analyses remain at the level of surface observation of directly measured parameters, lacking both the ability to proactively repair data affected by noise and the means to mine implicit correlations between multi-source data and construct dynamic knowledge graphs. Furthermore, they cannot deduce unmeasurable internal process states through mechanistic models, resulting in insufficient depth and limited accuracy in traceability analysis. Third, traditional data security solutions (such as static encryption) often conflict with efficient computational analysis, and fixed computing resources cannot adapt to the fluctuating real-time demands of traceability tasks, failing to achieve the synergistic optimization of security and efficiency as described in this invention. Finally, existing system architectures are rigid, with high module coupling, making it difficult to flexibly expand or adapt to production lines of different sizes. Compared to the modular and scalable collaborative system architecture of this invention, its applicability and maintainability are significantly inferior. These shortcomings collectively restrict the timeliness, accuracy, and intelligence level of data traceability in the radiopharmaceutical preparation process. Summary of the Invention

[0004] To overcome the aforementioned shortcomings of the prior art, this invention provides a method and system for tracing data of the separation process in radiopharmaceutical preparation, which solves the problems of data silos, reliance on manual analysis, low tracing efficiency, and lack of in-depth intelligent mechanism tracing capabilities in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method and system for tracing data on the separation process in radiopharmaceutical preparation includes the following steps:

[0007] S1, Data Reliability Processing: Perform anti-interference repair and verification on the raw time-series acquisition data output by the radiopharmaceutical separation equipment, and output repaired data with high confidence in the time-series data after repair and verification;

[0008] S2, Association Graph Construction: Based on the repair data output from step S1, association analysis is performed on the multi-source heterogeneous data in the separation process to generate an association graph defined as data entities as nodes, relationships between entities as edges, and with standardized semantic labels.

[0009] S3, Process Deduction and Back-calculation: Using the correlation map generated in step S2 as data and relational constraints, the process data is deduced and key process parameters are back-calculated by solving the symbolic differential equations describing the dynamics of the separation process, and the deduction data and parameter deviation information are output.

[0010] S4, Intelligent Traceability Application: Based on the correlation map generated in step S2, the inference data and parameter deviation information output in step S3, data retrieval and attribution analysis are performed on the separation process, and the full-link traceability results are output.

[0011] During the execution of steps S1 to S4, security and computing power adaptation operations are integrated:

[0012] The privacy data involving core processes and drug quality in steps S1 to S4 are processed using homomorphic encryption technology, and the encryption process supports the association, deduction and retrieval calculations required in steps S2 to S4 directly in the encrypted state.

[0013] Based on the real-time stage of different computing tasks in steps S1 to S4 and the preset urgency tags of the traceability tasks, edge computing resources are allocated through a dynamic scheduling mechanism.

[0014] Preferably, step S1 specifically includes:

[0015] Construct a dedicated dictionary for matching the spectral features of time-series data during the separation process;

[0016] Based on the interference-distortion mapping relationship established by historical data, abnormal data points caused by radiation fluctuations or equipment noise in the original time-series acquisition data are identified and removed, and missing data segments caused by communication packet loss are filled in.

[0017] Record the repair operation logs for abnormal data points and missing data segments, along with the corresponding interference identifiers. The logs and identifiers serve as traceable metadata associated with the output repair data.

[0018] Preferably, step S2 specifically includes:

[0019] A Bayesian network model was used to calculate the association probability of each data entity in the repaired data.

[0020] The Bayesian network model automatically learns and outputs the conditional probability relationships between three types of data entities: separation equipment parameters, material properties, and environmental variables, based on the input repair data.

[0021] Associations with conditional probabilities greater than a set threshold are transformed into standardized semantic tags that are readable by machines, and the association graph is dynamically constructed and updated based on these tags.

[0022] Preferably, the Bayesian network model is incrementally adaptively trained by acquiring a small batch of sample sequences composed of the repair data output in step S1 online, so as to dynamically optimize its network parameters.

[0023] Preferably, step S3 specifically includes:

[0024] The ordinary differential or partial differential equations describing the separation kinetics of radiopharmaceuticals are symbolically bound to the node data in the correlation graph to form a unified computational model.

[0025] Using key process variables in the repair data as initial or boundary conditions, the bound symbolic differential equations are numerically solved to obtain a sequence of intermediate state variables that cannot be directly measured during the separation process, which serves as inference data.

[0026] Using offline or online detection data of the separated products as constraints, the inverse problem of specific process parameters in the separation kinetic equation is solved to obtain parameter deviation information, which is then mapped to the corresponding parameter nodes in the correlation graph.

[0027] Preferably, step S4 specifically includes:

[0028] Using a temporal data semantic embedding model, user requests for natural language or structured queries about the separation process are converted into high-dimensional feature vectors;

[0029] In the feature vector space, standardized semantic labels, inferred data and data repair operation logs from step S1 are retrieved in parallel from the association graph, and the full-link tracing results containing the complete evolution path of the data are returned in order of timestamp and causal logic.

[0030] When the system receives an alarm about abnormal quality of the separated product or detects that a key parameter exceeds the limit, it automatically triggers the attribution analysis process: based on the parameter deviation information, it locates the root cause parameter of the abnormality, and reverses the correlation graph to trace back to the core process link, equipment sensor node and original interference event that caused the abnormality, and generates a structured and visualized attribution report.

[0031] Preferably, the homomorphic encryption technology specifically employs a linear transformation homomorphic encryption algorithm to ensure that after data encryption, the calculation result on the ciphertext is consistent with the result of direct calculation on the plaintext after decryption.

[0032] The method also includes: recording all access, query, and calculation operations on encrypted data, as well as the operation subjects and timestamps, through an audit log module, and using such audit logs as an immutable traceability evidence chain, which is then linked to the final end-to-end traceability result.

[0033] Preferably, the dynamic scheduling mechanism specifically includes:

[0034] Real-time monitoring of the execution status of each subtask in steps S1 to S4, including ready, running, blocked, and completed;

[0035] Query the urgency tags associated with the current traceability task. The tags are set according to the production batch priority or anomaly level.

[0036] Based on a pre-defined resource allocation strategy model that takes task status and urgency labels as input parameters, the CPU and memory resource quotas and execution priorities of each subtask on the edge computing node are dynamically adjusted.

[0037] The resource allocation strategy model is optimized based on historical task execution time and resource utilization data.

[0038] Preferably, a system for tracing data on the separation process in radiopharmaceutical preparation includes:

[0039] The reliability processing module is configured to execute step S1, and its output is connected to the input of the associated construction module via a data bus.

[0040] The association construction module is configured to execute step S2, and its output association graph is pushed to the inference and reverse inference module and the application tracing module through the message middleware interface.

[0041] The deduction and reverse deduction module is configured to execute step S3, and its output is connected to the input of the application tracing module via a data bus.

[0042] The application traceability module is configured to execute step S4, which is used to perform data retrieval and attribution analysis on the separation process based on the correlation map, inference data and parameter deviation information, and output the traceability results.

[0043] The security adaptation module, whose functions are integrated into the above modules in the form of service components, is configured to perform security and computing power adaptation operations;

[0044] The reliability processing module, the association construction module, the deduction and reverse inference module, and the application tracing module constitute the main data processing link in sequence. The output of each module serves as the direct input or key decision constraint for the downstream module, forming a closed loop.

[0045] Preferably, it also includes a collaborative control module;

[0046] The collaborative control module is specifically a central scheduling service program deployed on the edge gateway of the separation equipment or the server of the preparation workshop. It subscribes to the running status messages of each module in real time through the message middleware interface and sends control commands to each module through the data bus.

[0047] The collaborative control module has a pre-set task priority strategy table, which is used to generate specific computing power scheduling instructions based on the running status messages and externally input production scheduling instructions, and send them to the executor of the security adaptation module to drive the dynamic allocation of resources.

[0048] The technical effects and advantages of the separation process data traceability method and system for radiopharmaceutical preparation of this invention are as follows:

[0049] 1. This invention constructs a logically closed-loop and deeply collaborative method and system encompassing "data reliability processing - data association construction - process deduction and reverse inference - intelligent application tracing." This transforms the traditional discrete, passive, and manual-dependent data monitoring and post-event analysis model into a proactive, continuous, and automated intelligent analysis paradigm. The system can autonomously complete the entire chain of operations, from raw data repair, multi-source heterogeneous data association, process mechanism deduction to final root cause localization. This transforms the traditional time-consuming and lengthy post-event investigation into intelligent tracing completed automatically within minutes, significantly improving tracing efficiency and response speed.

[0050] 2. This invention enhances traceability capabilities from multiple fundamental dimensions through multi-level technological collaboration: At the data foundation level, dynamic learning and anti-interference repair technologies ensure the high quality and reliability of input data, avoiding error propagation; at the cognitive discovery level, adaptive learning probabilistic graphical models automatically mine complex explicit and implicit relationships between data, constructing and continuously evolving a system knowledge graph, enabling the discovery of complex causal relationships that are difficult for the human brain to perceive; at the mechanism insight level, by deeply integrating symbolic process models with real-time data and related knowledge, quantitative deduction of key intermediate process states that cannot be directly measured and inversion of process parameters are achieved, enabling traceability analysis to penetrate the surface and reach the root cause at the process mechanism level; at the application interaction level, multi-source information fusion retrieval technology based on semantic understanding can comprehensively and accurately associate all relevant data, events, and knowledge, and generate intuitive and credible visual evidence chains and attribution reports.

[0051] 3. This invention innovatively solves the challenge of balancing security and efficiency by deeply integrating homomorphic encryption technology and dynamic computing power scheduling mechanisms into the core traceability process. On the one hand, it achieves secure computation of core process parameters and privacy data in a "usable but invisible" manner, ensuring that complete traceability analysis can be performed under end-to-end data encryption, thus fully protecting intellectual property rights and sensitive information. On the other hand, through an intelligent resource scheduling model based on task stages and urgency, it can dynamically guarantee the execution priority and resource supply of each high-load traceability module (such as inference calculation and semantic retrieval) in a resource-constrained edge computing environment, thereby ensuring the overall high efficiency and stability of the system while meeting security requirements.

[0052] 4. This invention achieves efficient communication and collaboration among various professional functional modules through a standardized data bus and message middleware interface, and realizes global task scheduling and resource adaptation through an independent collaborative control module. This design not only enables core modules such as data reliability processing, correlation construction, and model inference to be developed, optimized, and deployed independently, but also allows the entire system to flexibly adapt to radiopharmaceutical preparation scenarios of different scales and complexities, possessing excellent maintainability, scalability, and engineering practicality. Attached Figure Description

[0053] Figure 1 This is a flowchart of the separation process data traceability method and system for radiopharmaceutical preparation proposed in this invention;

[0054] Figure 2 This is a system block diagram of the separation process data traceability method and system for radiopharmaceutical preparation proposed in this invention. Detailed Implementation

[0055] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0056] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include," "contain," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "includes..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0057] refer to Figure 1-2 This invention provides a method and system for tracing data in the separation process of radiopharmaceutical preparation. The method achieves end-to-end intelligent traceability by constructing sequentially executed, closed-loop data flow collaborative steps: First, the original time-series data of the equipment is subjected to anti-interference repair and verification, outputting high-quality repaired data; based on this, implicit semantic association mining of multi-source heterogeneous data is performed to construct an association graph with standardized semantic labels; then, using this graph as a constraint, process data deduction and key process parameter back-calculation are achieved by solving the symbolic differential equations describing the separation process, outputting deduction data and parameter deviation information; finally, the association graph, deduction data, and deviation information are fused, and a time-series data semantic embedding model is used for fuzzy retrieval and intelligent attribution, outputting the end-to-end traceability results. Throughout the process, homomorphic encryption technology is integrated to protect privacy data, and edge computing power allocation is optimized based on a dynamic scheduling mechanism. Correspondingly, the system includes a reliability processing module for sequential data connection, an association construction module, a deduction and reverse inference module, an application tracing module, and a security adaptation module that integrates security and computing power adaptation functions. Each module works together through a data bus and message middleware interface, and is uniformly scheduled by the collaborative control module to form the hardware and software entities that implement the above methods.

[0058] Example 1

[0059] This embodiment provides a method and system for tracing data during the separation process in radiopharmaceutical preparation, for the overall system architecture and collaborative data flow implementation. Specific implementation details include:

[0060] Purpose of implementation:

[0061] This embodiment aims to fully demonstrate how the sequentially executed steps with a closed-loop data flow, as well as the system's modular structure and collaborative control mechanism, can be physically implemented and executed in software within a specific industrial environment.

[0062] Implementation System:

[0063] This system adopts an edge-server hybrid architecture, as implemented below:

[0064] Reliability processing module: Deployed as a standalone microservice on an edge industrial computer located next to the radioactivity separation equipment. This computer is directly connected to temperature sensors, pressure transmitters, and radioactivity detectors via analog input cards and digital I / O cards.

[0065] The association construction module and the inference and reverse inference module are two core computing services deployed on the central server in the workshop. The association construction service is responsible for running Bayesian network inference; the inference and reverse inference service integrates a symbolic mathematics engine and a numerical computation library.

[0066] Application tracing module: Deployed as a web service on the same central server, providing APIs and a front-end visual interface.

[0067] Security adaptation module: Its functions are split into two parts: the encryption component is embedded in all the above modules as a library file; the scheduling component runs as an independent resource management service.

[0068] Collaborative Control Module: Deployed on a central server as a central scheduling service program.

[0069] System connection:

[0070] Data bus: Built using standard industrial communication protocols. Edge computers act as data servers, publishing real-time data nodes, while modules on the central server subscribe to these nodes as clients to obtain real-time repair data streams.

[0071] Message middleware interface: Built using a distributed message queue system. The system defines a series of event topics, and all modules act as either producers or consumers, coordinating through event-driven processes via publish / subscribe messages.

[0072] Implementation steps:

[0073] Taking a column chromatography purification batch of a labeled drug as an example, the system collaboratively performs the following steps:

[0074] S1: Data Reliability Processing: The reliability processing module on the edge computer reads raw sensor data in real time. It loads a dedicated dictionary file pre-trained for a specific purification process and performs sparse coding reconstruction on the raw time-series data. Simultaneously, it calls an interference model library to identify typical interference patterns, replaces the identified outliers with reconstructed values, and interpolates short-term packet loss to ultimately generate repaired data.

[0075] S2: Association Graph Construction: Repair data is transmitted to the central server in real time via the data bus. After receiving the batch start message from the message middleware, the association construction module begins caching the repair data for that batch. When the batch ends, the Bayesian network training process is initiated to generate or update the association graph describing the relationship between equipment parameters, material properties, and environmental variables. The graph is then stored in the graph database, and an update message is subsequently published.

[0076] S3: Process Deduction and Reverse Estimation: The deduction and reverse estimation module subscribes to chromatogram update topics. Upon receiving a message, it reads the latest correlation chromatogram and the remediation data for this batch. It calls the built-in adsorption kinetic equations (a set of partial differential equations), using relevant nodes in the chromatogram as equation parameters and key variables in the remediation data as boundary conditions, to perform numerical solutions, outputting deduced data that cannot be directly measured within the chromatographic column (such as spatial concentration distribution). Simultaneously, it uses product quality test results as targets to deduce the actual values ​​of key process parameters for this batch and compares them with standard values ​​to generate parameter deviation information. This deviation information is written back to the correlation chromatogram as an "abnormal event" node.

[0077] S4: Intelligent Traceability Application: The application traceability module monitors abnormal event topics. This module is triggered when the inference and reverse inference module writes a parameter deviation event, or when an operator submits a query. It first uses a temporal semantic embedding model to convert the query or event into a feature vector. Then, it queries in parallel: 1) the association graph in the graph database to find all nodes and paths strongly correlated with the deviation parameter; 2) the inference data in the time series database to locate the time period in which the anomaly occurred; and 3) the repair logs in the file server. Finally, it integrates all the information to generate a full-link traceability result and visualization report containing the causal chain.

[0078] Throughout the S1 to S4 process, the encryption component of the security adaptation module performs homomorphic encryption on all messages containing process parameter values ​​transmitted through the message middleware. Simultaneously, its scheduling component listens to load status messages published by each module and dynamically adjusts the computing resource limits of each service container according to pre-defined policies.

[0079] Implementation results:

[0080] This embodiment fully and specifically implements the system architecture and method steps by specifying a concrete communication architecture, coordination mechanism and execution process, proving that the entire solution is deployable, runnable and verifiable.

[0081] Example 2

[0082] This embodiment provides a method and system for tracing data during the separation process in radiopharmaceutical preparation, and details the implementation of data reliability processing (S1). The specific implementation includes:

[0083] Purpose of implementation:

[0084] This embodiment demonstrates the specific algorithm, data structure, and output of "interference prevention repair and verification," ensuring that the generation process of "repair data" and "repair operation log" is fully disclosed.

[0085] Implementation System:

[0086] The implementation system in this embodiment is consistent with the relevant parts in Embodiment 1. Specifically, a reliability processing module deployed on an edge industrial computer is enabled. This module receives raw timing data through a data bus and publishes its status events through a message middleware interface.

[0087] Implementation steps:

[0088] S1: Focuses on repairing the "pre-column pressure" signal.

[0089] a) Dedicated dictionary construction (offline): Collect a large number of normal batches of pressure signal segments with a fixed sampling rate. Use an online dictionary learning algorithm to learn an overcomplete dictionary matrix that can sparsely represent normal pressure fluctuation patterns.

[0090] b) Real-time repair process:

[0091] 1. For sliding window data arriving in real time.

[0092] 2. Sparse Coding: Solving an optimization problem that includes an L1 regularization term to obtain sparse representation coefficients of the data in a dictionary. An iterative shrinking thresholding algorithm is used for this purpose.

[0093] 3. Reconstruction: Calculate the reconstructed signal using a dictionary and sparse coefficients.

[0094] 4. Anomaly Detection: Calculate the residuals between the original signal and the reconstructed signal. If the absolute value of the residual of a data point exceeds a specified multiple (e.g., 3 times) of the standard deviation of the historical residuals, it is marked as a candidate anomaly.

[0095] 5. Interference Model Verification: Data segments from candidate anomalies and their neighboring points are dynamically time-warped and their distances are calculated against predefined interference templates such as the "water hammer effect." If the distance is less than a set threshold, the anomaly is confirmed as being caused by a specific interference.

[0096] 6. Repair and Output: For confirmed outliers, replace them with the corresponding values ​​from the reconstructed signal. For consecutive missing segments identified as packet loss, use the reconstructed signal to perform cubic spline interpolation to complete the segment. Finally, output the repaired stress sequence as the repaired data.

[0097] c) Log Recording: Generate a detailed structured log record, which includes a timestamp, sensor identifier, operation type (replacement or interpolation), index position of the data points involved, original value, repaired value, matched interference cause identifier, and confidence level of this repair operation. This log file is organized and stored in batches and associated with the corresponding repaired data segments through metadata.

[0098] Implementation results:

[0099] This embodiment fully and clearly discloses the technical features of "removing abnormal data", "completing lost packet data" and "recording repair trajectory" by providing specific algorithms, mathematical models, judgment thresholds and structured log information.

[0100] Example 3

[0101] This embodiment provides a method and system for tracing data on the separation process in radiopharmaceutical preparation, used for association map construction (S2) and Bayesian network training. Specific implementation details include:

[0102] Purpose of implementation:

[0103] This embodiment details how "implicit semantic association mining" and "incremental adaptive training" are implemented through specific Bayesian network models.

[0104] Implementation System:

[0105] The implementation system in this embodiment is consistent with the relevant parts in Embodiment 1. Specifically, the association building module deployed on the central server is enabled. This module receives the repair data stream from the reliability processing module through the data bus and coordinates events with other modules through the message middleware interface.

[0106] Implementation steps:

[0107] S2:

[0108] a) Data preparation and node definition: The module extracts multiple key variables from the repair data as network nodes, such as column temperature, flow rate, back pressure, inlet activity, cooling water temperature, and final product purity.

[0109] b) Network structure and initialization: The initial network structure incorporates some causal edges determined by domain knowledge, while the existence and direction of the remaining edges allow for learning from the data.

[0110] c) Incremental Adaptive Training: Training is performed on a mini-batch basis. Each production batch consists of time-aligned multivariate samples forming a mini-batch sequence. The module executes the following process:

[0111] 1. Read in a new mini-batch sample sequence.

[0112] 2. Using the conditional probability table parameters of the current Bayesian network, calculate the log-likelihood of the new data sequence.

[0113] 3. The stochastic gradient descent algorithm is used to update the network's conditional probability table parameters with the goal of maximizing the log-likelihood. The learning rate and momentum parameters have explicitly set values.

[0114] 4. Through this process, the network continuously adjusts its probability estimates of the dependencies between variables. For example, it may become more certain that "cooling water temperature" has a strong influence on "column temperature".

[0115] d) Graph Generation: After training, edges whose conditional probability estimates change significantly (e.g., changes exceeding 0.1) are selected, and standardized semantic description labels are generated for them. For example, an edge might be labeled "Strong Influence: Cooling Water Temperature → Column Temperature". Finally, all variable nodes and semantically labeled associated edges are imported into the graph database to form a queryable association graph.

[0116] Implementation results:

[0117] This embodiment clarifies the specific input variables, training algorithm, update triggering conditions, and label generation rules of the Bayesian network.

[0118] Example 4

[0119] This embodiment provides a method and system for tracing data of the separation process in radiopharmaceutical preparation, and for the numerical implementation of process deduction and reverse inference (S3). Specific implementation details include:

[0120] Purpose of implementation:

[0121] This embodiment demonstrates how "incremental solution of symbolic differential equations" and "parameter back-calculation" are accomplished through numerical computation.

[0122] Implementation System:

[0123] The implementation system in this embodiment is consistent with the relevant parts in Embodiment 1. Specifically, the inference and reverse inference module deployed on the central server is enabled. This module receives correlation graph update events through a message middleware interface and obtains the required repair data through a data bus.

[0124] Implementation steps:

[0125] S3: Taking the separation process of ion exchange column as an example.

[0126] a) Model Binding: The kinetics of this separation process are described by a set of convection-diffusion-reaction partial differential equations containing convection, diffusion, and adsorption terms. The module binds the symbolic parameters in the equations (such as diffusion coefficient and mass transfer coefficient) to the actual process parameters represented by nodes such as "resin type" and "flow rate" in the correlation graph.

[0127] b) Data extrapolation: The repair data obtained from S1, such as the inlet concentration time series and flow rate, are used as the boundary conditions or initial conditions for the partial differential equations. Using a finite element method solver, numerical discretization and solution are performed in the spatial domain (column length) and time domain (batch duration) to obtain the detailed distribution of the target concentration in the entire separation column as a function of time and space. This is the extrapolation data.

[0128] c) Parameter Back-calculation: When the final product quality (e.g., chemical purity) is unqualified, the optimization objective is set as minimizing the difference between the simulated outlet concentration curve and the measured outlet concentration curve. The Levenburg-Marquardt optimization algorithm is used to automatically adjust the values ​​of specific key parameters (e.g., mass transfer coefficient) in the kinetic equation. Through iterative calculation, the percentage deviation of the actual value of this parameter from the standard value for this batch is derived, generating parameter deviation information. This information is added as an event node to the correlation graph and connected to relevant process condition nodes.

[0129] Implementation results:

[0130] By providing specific equation forms, solution methods, and optimization algorithms, this embodiment gives clear and operable engineering meanings to terms such as "symbolic differential equations," "numerical solutions," and "inverse problem solutions."

[0131] Example 5

[0132] This embodiment provides a method and system for tracing data on the separation process in radiopharmaceutical preparation, used for fusion retrieval and attribution in intelligent traceability applications (S4). Specific implementation details include:

[0133] Purpose of implementation:

[0134] This embodiment illustrates how "fuzzy retrieval" and "intelligent attribution" are achieved through multi-source information fusion, and clearly explains the role of the temporal semantic embedding model.

[0135] Implementation System:

[0136] The implementation system in this embodiment is consistent with the relevant parts in Embodiment 1. Specifically, the application tracing module deployed on the central server is enabled. This module listens for various events through a message middleware interface and accesses the correlation graph, inference data, and repair logs through a data bus and a dedicated query interface.

[0137] Implementation steps:

[0138] S4: Assume that the key quality attributes of a certain batch of products do not meet the standards.

[0139] a) Triggering and Vectorization: The system automatically triggers the attribution analysis process. The temporal semantic embedding model within the module (e.g., a model pre-trained and fine-tuned based on the Transformer architecture) converts the natural language query for the abnormal event name or operator into a high-dimensional feature vector. This model can map natural language text or structured labels describing the separation process events into vector representations in a high-dimensional semantic space, thereby achieving cross-modal retrieval based on semantic similarity.

[0140] b) Cross-source parallel retrieval:

[0141] In the correlation graph: Use graph query language to find all nodes and correlation paths connected to the quality problem node within a few steps. Return potentially related process parameters, material properties, and historical event nodes.

[0142] In the extrapolated data: In the high-dimensional vector space, the extrapolated data segment with the highest vector cosine similarity is retrieved and queried (these data segments have been pre-converted into vectors by the same model). This allows the specific time period during which abnormal state changes occurred.

[0143] In the repair log: In the log database, retrieve all interference events with a warning level or higher recorded within the same time period mentioned above.

[0144] c) Association Analysis and Attribution: The module performs spatiotemporal alignment and logical association analysis on the search results from three sources. Through cross-validation, it infers the root cause chain leading to quality anomalies. For example, sensor drift causes deviations in actual process parameters, which in turn triggers changes in internal mass transfer efficiency (as revealed by the simulation data), ultimately resulting in substandard final product quality. The system automatically traverses the association graph in reverse to clearly display this causal chain.

[0145] d) Report generation: Automatically generate a visual attribution report that integrates time series plots, causal relationship diagrams, key data comparisons, and root cause conclusions.

[0146] Implementation results:

[0147] This embodiment clearly demonstrates the specific implementation methods of "fuzzy search" and "intelligent attribution" functions by specifying the retrieval method, analysis logic, and report content, and clarifying the technical role of the temporal semantic embedding model.

[0148] Comparative Example 1

[0149] The comparison model reportedly provides unified data monitoring and offline analysis modes, including:

[0150] Implementation System:

[0151] A separate programmable logic controller / data acquisition and monitoring control system is used for real-time monitoring and alarms. Process data and quality data are stored separately in different historical databases and laboratory information management systems. Problem analysis relies entirely on engineers performing it manually using spreadsheets and business analytics software.

[0152] Implementation steps and effects:

[0153] Facing the same quality anomaly problem as in Example 5:

[0154] Step 1: Monitoring systems typically only alarm when process parameters exceed hard thresholds. They are difficult to detect slow drifts or coupling effects of parameters within the range, and therefore may not trigger alarms.

[0155] Step 2: Engineers need to manually export massive amounts of time-series data and quality inspection results from different systems, and perform tedious data cleaning and time alignment work, which takes up to several hours.

[0156] Step 3: The analytical process is highly dependent on personal experience. An engineer may notice a slight trend in a parameter (such as pH) in a spreadsheet, but it is difficult to quantify its correlation with the final quality, let alone understand how it works by affecting internal processes (such as adsorption efficiency).

[0157] Step 4: Due to the lack of ability to extrapolate the internal state of the process, the analysis remains at the level of surface parameters and cannot pinpoint the deep-seated process root causes such as "decreased mass transfer coefficient".

[0158] Step 5: The entire investigation and analysis process is slow, usually taking half a day to several days. The conclusions are subject to considerable uncertainty, and the analytical logic and knowledge are difficult to solidify and standardize for reuse.

[0159] Compared to Examples 1-5 and Comparative Example 1, this invention constructs a collaborative and intelligent data traceability system, fundamentally contrasting with the traditional discrete monitoring and manual analysis mode represented by the comparative example. In terms of technical architecture and working mechanism, Example 1 demonstrates a modular, closed-loop system tightly coupled with a data bus and message middleware, achieving fully automated flow from data acquisition to intelligent applications. In contrast, the traditional mode relies on isolated subsystems, with data fragmented and processes dependent on manual connection. A deeper difference lies in core capabilities: Example 2 achieves data-level intelligent repair and self-verification through dynamic dictionary learning and sparse coding; Example 3 autonomously mines and solidifies explicit and implicit associations from the data stream using incremental Bayesian networks, forming a dynamically evolving knowledge graph; Example 4 achieves quantitative deduction and parameter inversion of unmeasurable internal process states by fusing symbolic dynamic equations with real-time data and the knowledge graph; and Example 5 achieves accurate and rapid root cause localization through semantic embedding and multi-source fusion retrieval. In contrast, traditional methods can only perform threshold alarms and data storage. Their analysis relies entirely on engineers making passive, empirical guesses about correlations between discrete data points. They cannot build deep understanding of correlations or gain insight into the internal state of the process "black box".

[0160] The two methods demonstrate a qualitative difference in their final results, fully proving the significant progress of this invention. In terms of efficiency and accuracy, this invention can automatically complete the entire process from anomaly detection to evidence chain generation within minutes (Example 5), while traditional manual analysis takes several days and yields vague conclusions (Comparative Example). Regarding analytical depth, this invention can quantitatively reveal the complete causal chain of "parameter deviation → internal state change → quality defect" (Example 4), while traditional analysis stops at observing the correlation of surface parameters. In terms of knowledge iteration and system adaptability, the correlation graph and model of this invention possess continuous autonomous learning and evolution capabilities (Example 3), while the analytical experience of traditional methods is lost with personnel changes, and the system itself lacks evolutionary capabilities. Furthermore, this invention achieves a balance between security and efficiency by integrating homomorphic encryption and dynamic computing power scheduling (throughout each step), something traditional methods often struggle to achieve simultaneously. In summary, this invention, through multi-technology collaboration, achieves a paradigm shift from "discrete, passive, shallow, and reliant on manual methods" to "closed-loop, proactive, deep, automated, and intelligent" methods, solving the traceability challenges in demanding industrial scenarios.

[0161] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of protection of the claims.

[0162] In conclusion, the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method and system for tracing data on the separation process in the preparation of radiopharmaceuticals, characterized in that, Includes the following steps: S1, Data Reliability Processing: Perform anti-interference repair and verification on the raw time-series acquisition data output by the radiopharmaceutical separation equipment, and output repaired data with high confidence in the time-series data after repair and verification; S2, Association Graph Construction: Based on the repair data output from step S1, association analysis is performed on the multi-source heterogeneous data in the separation process to generate an association graph defined as data entities as nodes, relationships between entities as edges, and with standardized semantic labels. S3, Process Deduction and Back-calculation: Using the correlation map generated in step S2 as data and relational constraints, the process data is deduced and key process parameters are back-calculated by solving the symbolic differential equations describing the dynamics of the separation process, and the deduction data and parameter deviation information are output. S4, Intelligent Traceability Application: Based on the correlation map generated in step S2, the inference data and parameter deviation information output in step S3, data retrieval and attribution analysis are performed on the separation process, and the full-link traceability results are output. During the execution of steps S1 to S4, security and computing power adaptation operations are integrated: The privacy data involving core processes and drug quality in steps S1 to S4 are processed using homomorphic encryption technology, and the encryption process supports the association, deduction and retrieval calculations required in steps S2 to S4 directly in the encrypted state. Based on the real-time stage of different computing tasks in steps S1 to S4 and the preset urgency tags of the traceability tasks, edge computing resources are allocated through a dynamic scheduling mechanism.

2. The method for tracing data of the separation process in the preparation of radiopharmaceuticals as described in claim 1, characterized in that, Step S1 specifically includes: Construct a dedicated dictionary for matching the spectral features of time-series data during the separation process; Based on the interference-distortion mapping relationship established by historical data, abnormal data points caused by radiation fluctuations or equipment noise in the original time-series acquisition data are identified and removed, and missing data segments caused by communication packet loss are filled in. Record the repair operation logs for abnormal data points and missing data segments, along with the corresponding interference identifiers. The logs and identifiers serve as traceable metadata associated with the output repair data.

3. The method for tracing data of the separation process in the preparation of radiopharmaceuticals as described in claim 1, characterized in that, Step S2 specifically includes: A Bayesian network model was used to calculate the association probability of each data entity in the repaired data. The Bayesian network model automatically learns and outputs the conditional probability relationships between three types of data entities: separation equipment parameters, material properties, and environmental variables, based on the input repair data. Associations with conditional probabilities greater than a set threshold are transformed into standardized semantic tags that are readable by machines, and the association graph is dynamically constructed and updated based on these tags.

4. The method for tracing data of the separation process in the preparation of radiopharmaceuticals as described in claim 1, characterized in that, The Bayesian network model is incrementally adaptively trained by acquiring a small batch of sample sequences composed of the repair data output from step S1 online, so as to dynamically optimize its network parameters.

5. The method for tracing data of the separation process in the preparation of radiopharmaceuticals as described in claim 1, characterized in that, Step S3 specifically includes: The ordinary differential or partial differential equations describing the separation kinetics of radiopharmaceuticals are symbolically bound to the node data in the correlation graph to form a unified computational model. Using key process variables in the repair data as initial or boundary conditions, the bound symbolic differential equations are numerically solved to obtain a sequence of intermediate state variables that cannot be directly measured during the separation process, which serves as inference data. Using offline or online detection data of the separated products as constraints, the inverse problem of specific process parameters in the separation kinetic equation is solved to obtain parameter deviation information, which is then mapped to the corresponding parameter nodes in the correlation graph.

6. The method for tracing data of the separation process in the preparation of radiopharmaceuticals as described in claim 1, characterized in that, Step S4 specifically includes: Using a temporal data semantic embedding model, user requests for natural language or structured queries about the separation process are converted into high-dimensional feature vectors; In the feature vector space, standardized semantic labels, inferred data and data repair operation logs from step S1 are retrieved in parallel from the association graph, and the full-link tracing results containing the complete evolution path of the data are returned in order of timestamp and causal logic. When the system receives an alarm about abnormal quality of the separated product or detects that a key parameter exceeds the limit, it automatically triggers the attribution analysis process: based on the parameter deviation information, it locates the root cause parameter of the abnormality, and reverses the correlation graph to trace back to the core process link, equipment sensor node and original interference event that caused the abnormality, and generates a structured and visualized attribution report.

7. The method for tracing data of the separation process in the preparation of radiopharmaceuticals as described in claim 1, characterized in that, Homomorphic encryption technology specifically employs a linear transformation homomorphic encryption algorithm to ensure that after data is encrypted, the calculation result on the ciphertext is consistent with the result of direct calculation on the plaintext after decryption. The method also includes: recording all access, query, and calculation operations on encrypted data, as well as the operation subjects and timestamps, through an audit log module, and using such audit logs as an immutable traceability evidence chain, which is then linked to the final end-to-end traceability result.

8. The method for tracing data of the separation process in the preparation of radiopharmaceuticals as described in claim 1, characterized in that, The dynamic scheduling mechanism specifically includes: Real-time monitoring of the execution status of each subtask in steps S1 to S4, including ready, running, blocked, and completed; Query the urgency tags associated with the current traceability task. The tags are set according to the production batch priority or anomaly level. Based on a pre-defined resource allocation strategy model that takes task status and urgency labels as input parameters, the CPU and memory resource quotas and execution priorities of each subtask on the edge computing node are dynamically adjusted. The resource allocation strategy model is optimized based on historical task execution time and resource utilization data.

9. The system for data traceability in the separation process of radiopharmaceutical preparation as described in any one of claims 1-8, characterized in that, include: The reliability processing module is configured to execute step S1, and its output is connected to the input of the associated construction module via a data bus. The association construction module is configured to execute step S2, and its output association graph is pushed to the inference and reverse inference module and the application tracing module through the message middleware interface. The deduction and reverse deduction module is configured to execute step S3, and its output is connected to the input of the application tracing module via a data bus. The application traceability module is configured to execute step S4, which is used to perform data retrieval and attribution analysis on the separation process based on the correlation map, inference data and parameter deviation information, and output the traceability results. The security adaptation module, whose functions are integrated into the above modules in the form of service components, is configured to perform security and computing power adaptation operations; The reliability processing module, the association construction module, the deduction and reverse inference module, and the application tracing module constitute the main data processing link in sequence. The output of each module serves as the direct input or key decision constraint for the downstream module, forming a closed loop.

10. The separation process data traceability system for radiopharmaceutical preparation as claimed in claim 9, characterized in that, It also includes a collaborative control module; The collaborative control module is specifically a central scheduling service program deployed on the edge gateway of the separation equipment or the server of the preparation workshop. It subscribes to the running status messages of each module in real time through the message middleware interface and sends control commands to each module through the data bus. The collaborative control module has a pre-set task priority strategy table, which is used to generate specific computing power scheduling instructions based on the running status messages and externally input production scheduling instructions, and send them to the executor of the security adaptation module to drive the dynamic allocation of resources.

Images