A method and system for irradiation processing multi-tenant data management

Abstract

The application relates to the technical field of data management and discloses a kind of irradiation processing multi-tenant data management method and system, wherein a kind of irradiation processing multi-tenant data management method includes: constructing anti-radiation data acquisition gateway, deploying anti-radiation sensor array to carry out data acquisition;The data collected is processed in stages, and is processed using a corresponding consistency strategy;Deep learning model is converted into edge deployment version by knowledge distillation technology, cloud edge collaborative reasoning is realized;Establish data source credit score, carry out multi-modal data fusion and quality guarantee;Construct equipment knowledge graph, realize fault diagnosis through graph attention network and causal reasoning;Adopt alliance chain technology, realize the trusted management of data full life cycle through smart contract;The application solves the problems of data security isolation, real-time performance and consistency balance, intelligent resource scheduling and edge computing support in the irradiation processing multi-tenant environment, and realizes the efficient and safe management of irradiation processing data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data management, more particularly, it relates to a kind of irradiation processing multi-tenant data management method and system. BACKGROUND

[0002] Irradiation processing is a new green environmental protection industry technology, and is widely used in the fields of medical instrument sterilization, food preservation, polymer material modification and the like. With the expansion of the business scale of irradiation processing enterprises and the increase of the number of customers, the traditional single-tenant data management mode has been unable to meet the data security, isolation and efficiency requirements in the multi-tenant environment. The existing multi-tenant data management technology mainly focuses on cloud computing and SaaS services, and lacks special design for the special needs of irradiation processing industry.

[0003] The multi-tenant data management scheme in the prior art has significant deficiencies: the traditional database partitioning and shared table mode has security risks in physical isolation and logical isolation, especially in the management of irradiation process parameters involving business secrets, and cannot provide sufficient security guarantee; the existing scheme has obvious performance decline when processing real-time data streams, and it is difficult to meet the strict requirements of millisecond-level parameter monitoring in the irradiation process. In addition, the lack of intelligent resource scheduling mechanism leads to low equipment utilization and high energy consumption.

[0004] The present application proposes a comprehensive solution integrating anti-radiation data acquisition, hybrid consistency protocol, edge intelligent algorithm, knowledge graph diagnosis and blockchain storage, aiming to solve the key technical problems in the digital transformation of the industry and improve the data management level and operational efficiency of irradiation processing enterprises. SUMMARY

[0005] The present application provides a kind of irradiation processing multi-tenant data management method and system, solve the technical problems that there are security risks in physical isolation and logical isolation in the related art, obvious performance decline when processing real-time data streams, and it is difficult to meet the strict requirements of millisecond-level parameter monitoring in the irradiation process.

[0006] The present application provides a kind of irradiation processing multi-tenant data management method, comprising the following steps:

[0007] Obtain the original data in the irradiation processing process, the original data includes device data using different communication protocols, and the device data using different communication protocols is uniformly converted into standard data stream using a device driver adapter framework;

[0008] The standard data stream is processed using a hybrid consistency protocol, is divided into four levels according to the influence degree of data on security and quality, and is processed to obtain hierarchical data using a consistency strategy;

[0009] The hierarchical data is processed by a lightweight edge intelligent algorithm, a deep learning model is compressed into an edge deployment version through a knowledge distillation technique, and intelligent processing data is obtained through cloud-edge collaborative reasoning;

[0010] The intelligent processing data is fused by an adaptive quality assurance mechanism, a data source reputation scoring system is established, cross-validation and fusion processing of different modal data are performed to obtain fused data; specifically including:

[0011] Data source reputation evaluation, a reputation scoring system based on historical data quality and equipment state is established;

[0012] Multi-modal data fusion algorithm, cross-validation and fusion processing of sensor data, image data and log data;

[0013] Adaptive data cleaning strategy, dynamically adjusting data cleaning rules and thresholds according to data quality evaluation results;

[0014] Among them, the fusion algorithm is based on the Bayesian inference framework, and the reliability and correlation of each data source are considered comprehensively; the weight calculation method of multi-modal fusion is: multiply the reputation value of each data source with its correlation coefficient, then divide by the sum of the product of the reputation value and the correlation coefficient of all data sources, to obtain the fusion weight of the data source; the correlation coefficient is calculated by the Pearson correlation coefficient method, and the mutual information method is used for different categories of data sources;

[0015] The fused data is analyzed by a knowledge graph driven fault diagnosis method, a knowledge graph containing device entities, fault entities, symptom entities and cause entities is constructed, and a graph neural network is used to realize fault recognition and root cause analysis to obtain a diagnosis result;

[0016] The diagnosis result and the fused data are stored and managed by blockchain technology, an alliance blockchain platform is constructed, and a smart contract is used to realize trusted management of the whole life cycle of data.

[0017] In a preferred embodiment, the acquisition of the original data in the irradiation process includes:

[0018] Deploy an anti-radiation sensor array, use a redundant design and fault detection mechanism to ensure the reliability of data acquisition;

[0019] Implementing heterogeneous device interface adaptation, converting device data of different communication protocols into standard data streams through a device driver adapter framework;

[0020] Adjusting the data acquisition frequency according to the irradiation process stage, increasing the acquisition frequency in the key process stage to obtain more detailed process data.

[0021] In a preferred embodiment, the hierarchical processing of standard data streams using a hybrid consistency protocol includes:

[0022] Data importance grading mechanism, dividing data into four levels of core security, critical quality, important operation, and general statistics according to the degree of influence on security and quality;

[0023] Predictive data synchronization strategy, predicting data synchronization timing based on data access patterns and business requirements;

[0024] Distributed lock mechanism optimization, using hierarchical lock structure to reduce lock competition and deadlock risk.

[0025] In a preferred embodiment, the processing of hierarchical data using lightweight edge intelligent algorithms includes:

[0026] Model compression and knowledge distillation, compressing deep learning models into edge deployment versions, maintaining model accuracy while reducing computational complexity;

[0027] Incremental learning mechanism, supporting continuous learning and parameter updating of models on edge devices;

[0028] Collaborative reasoning architecture, realizing cloud-edge collaborative reasoning, and dynamically allocating computing resources according to task complexity.

[0029] In a preferred embodiment, the analysis of fusion data using a knowledge graph-driven fault diagnosis method includes:

[0030] Knowledge graph construction, establishing a multi-level knowledge graph containing device entities, fault entities, symptom entities, and cause entities;

[0031] Graph neural network fault identification, using graph convolution network to model device state and identify fault patterns;

[0032] Causal reasoning root cause analysis, based on causal graph model for fault root cause analysis and propagation path tracking;

[0033] Dynamic knowledge update mechanism, continuously updating the knowledge graph according to new fault cases and expert knowledge.

[0034] In a preferred embodiment, the use of a device-driven adapter framework to uniformly convert device data of different communication protocols into standard data streams includes:

[0035] Unified interface layer, converting device data of different communication protocols into standard data streams;

[0036] Protocol conversion module, supporting data format conversion of multiple industrial communication protocols;

[0037] Data standardization processing ensures consistent representation of heterogeneous device data.

[0038] In a preferred embodiment, the implementation of cloud-edge collaborative inference to obtain intelligent processing data includes:

[0039] Task allocation strategy dynamically allocates inference tasks according to computational complexity and network delay;

[0040] Edge caching mechanism caches commonly used models and intermediate results on edge devices;

[0041] Collaborative optimization algorithm realizes coordinated configuration of computing resources of cloud and edge devices.

[0042] In a preferred embodiment, the implementation of data lifecycle trusted management through smart contract includes:

[0043] Data access control contract defines data access permissions and operation rules of different tenants;

[0044] Quality evaluation contract automatically performs data quality evaluation and reputation score update;

[0045] Audit traceability contract records data operation history and responsibility traceability information.

[0046] In a preferred embodiment, a radiation processing multi-tenant data management system for executing the above-mentioned radiation processing multi-tenant data management method includes:

[0047] Anti-radiation data acquisition module for acquiring original data in the radiation processing process, the original data including device data in different communication protocols, and the device data in different communication protocols being uniformly converted into standard data stream by using a device driver adapter framework;

[0048] Hybrid consistency protocol module for processing the standard data stream using a hybrid consistency protocol, dividing the data into four levels according to the influence of the data on security and quality, and processing the data using a consistency strategy to obtain classified data;

[0049] Lightweight edge intelligence module for processing the classified data using a lightweight edge intelligence algorithm, compressing a deep learning model into an edge deployment version through a knowledge distillation technology, and realizing cloud-edge collaborative inference to obtain intelligent processing data;

[0050] Adaptive data quality assurance module for processing the intelligent processing data using an adaptive quality assurance mechanism, establishing a data source reputation score system, and cross-verifying and fusing different modal data to obtain fused data;

[0051] The knowledge graph fault diagnosis module is used for analyzing the fusion data by using a knowledge graph driven fault diagnosis method, constructing a knowledge graph comprising device entities, fault entities, symptom entities and cause entities, and realizing fault identification and root cause analysis by a graph neural network to obtain a diagnosis result.

[0052] The blockchain data storage module is used for storing and managing the diagnosis result and the fusion data by using a blockchain technology, constructing a consortium blockchain platform, and realizing trusted management of the whole life cycle of data by an intelligent contract.

[0053] The present application has the following advantages:

[0054] By using the multi-level tenant isolation mechanism and the blockchain data storage technology, the network layer, the computing layer and the storage layer are fully isolated, so that the irradiation data of different tenants are completely independent and cannot be tampered with. The anti-radiation sensor array and the optical fiber network transmission effectively avoid the influence of the radiation environment on data acquisition and transmission, and meet the strict safety requirements of the high supervision industry such as medical equipment.

[0055] The hybrid consistency protocol and the cloud-edge collaborative reasoning architecture effectively balance the real-time and consistency requirements, reduce the data response time, improve the system availability, and improve the data consistency. The intelligent scheduling algorithm optimizes the distribution according to the device load, processing priority, energy cost and other factors, improves the average utilization rate of the device and the fault warning accuracy, and improves the operation efficiency and reliability of the irradiation equipment. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 is a flow chart of a multi-tenant data management method for irradiation processing of the present application;

[0057] Figure 2 is a module diagram of a multi-tenant data management system for irradiation processing of the present application;

[0058] Figure 3 is a comparison chart of system performance monitoring of the present application;

[0059] Figure 4 is a comparison chart of device utilization rate improvement of the present application;

[0060] Figure 5 is a radar chart of data quality comprehensive evaluation of the present application. DETAILED DESCRIPTION

[0061] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that discussions of these implementations are merely provided to enable those skilled in the art to better understand so as to best use the subject matter described herein, and variations of elements can be made dependant on the elements discussed without departing from the scope of the present disclosure. Various examples can omit, substitute, or add various procedures or components as appropriate. Also, it should be understood that instead of being combined, some features of the examples described can be split up into separate examples.

[0062] A method for irradiation processing multi-tenant data management is disclosed in at least one embodiment of the present invention, as shown, comprising the steps of: Figure 1 as shown, comprising the steps of:

[0063] Step 1, obtaining original data in the irradiation processing process, the original data including device data in different communication protocols, and using a device driver adapter framework to uniformly convert the device data in different communication protocols into a standard data stream;

[0064] Specifically, it includes the following contents:

[0065] Step 1.1, heterogeneous device interface adaptation;

[0066] The system first performs interface adaptation on various devices in the irradiation workshop. For different manufacturers' cobalt-60 devices, electron beam devices, X-ray devices, etc., a unified device abstraction layer is established. Through a device driver adapter framework, device data in different communication protocols such as RS485, Ethernet, CAN bus, etc. are uniformly converted into a standard JSON format data stream. Each device adapter contains a device identification module, a protocol conversion module and a data format standardization module to ensure uniform access of heterogeneous device data.

[0067] Step 1.2, anti-radiation sensor array deployment;

[0068] Anti-radiation sensor arrays are deployed at key locations in the irradiation workshop, including irradiation dose sensors, temperature sensors, humidity sensors, pressure sensors, etc. Each sensor is configured with triple redundancy, and reliable measurement data is obtained through a weighted fusion algorithm. The sensor shell is made of lead glass and tantalum metal composite material, which can withstand a cumulative irradiation dose of up to 100 kGy. Sensor data is transmitted through an optical fiber network, avoiding the influence of electromagnetic interference on data transmission.

[0069] Step 1.3, adaptive acquisition frequency adjustment;

[0070] The system dynamically adjusts the data collection frequency according to different stages of the irradiation process. During the start and end stages of irradiation, the collection frequency is set to 100 times per second to ensure accurate capture of key parameter changes. During the stable irradiation stage, the collection frequency is reduced to 10 times per second to reduce data processing load. When abnormal fluctuations are detected, the system automatically increases the collection frequency to 500 times per second to achieve fine-grained monitoring of abnormal events.

[0071] The collection frequency adjustment algorithm dynamically adjusts the collection frequency according to the process stage and abnormality detection results. The algorithm adjusts the base collection frequency by multiplying the process stage adjustment factor and the abnormality detection adjustment factor. The process stage adjustment factor determines the adjustment multiple based on the current irradiation process stage, and the abnormality detection adjustment factor determines the frequency increase multiple based on the degree of abnormality detected by the system.

[0072] Further, an adaptive collection strategy based on deep learning can be used to predict data trends through a long short-term memory network, and dynamically adjust the collection frequency based on the prediction results to further optimize data collection efficiency and quality.

[0073] Step 2, the standard data stream is processed using a hybrid consistency protocol, and is divided into four levels according to the impact of the data on security and quality, and is processed using a consistency strategy to obtain hierarchical data;

[0074] Specifically, the following content is included:

[0075] Step 2.1, data importance classification mechanism;

[0076] The system classifies irradiation data into four levels according to the impact on security and quality: core security data, key quality data, important operation data, and general statistical data. Core security data includes irradiation dose, equipment status, etc., requiring strong consistency guarantee. Key quality data includes process parameters such as temperature, humidity, and pressure, using a quasi-strong consistency strategy. Important operation data such as device load and energy consumption uses weak consistency guarantee. General statistical data such as historical trend analysis data uses eventual consistency strategy.

[0077] The data importance classification mechanism scores data based on business value, timeliness, and integrity. The scoring process consists of three steps: calculating business value score, assigning a score of 1-10 based on the impact of data on production decisions; calculate timeliness score, assign weights based on data freshness and time sensitivity; calculate integrity score, assign appropriate scores based on data missing rate and quality issues.

[0078] The final importance score is calculated using a weighted average, with business value weighted at 0.5, timeliness weighted at 0.3, and completeness weighted at 0.2. Based on the final score, data is categorized into three levels: high importance (8-10 points), medium importance (5-7 points), and low importance (1-4 points). Different synchronization strategies and storage priorities are applied to data of different levels.

[0079] Step 2.2, Predictive data synchronization strategy;

[0080] Based on historical data pattern analysis, the system establishes a data access prediction model to pre-synchronize potentially accessed data to relevant nodes. The prediction model employs time series analysis, combining tenant behavior patterns and business cycle regularities to calculate the probability of data access. When the predicted probability exceeds a set threshold, the system proactively performs data pre-synchronization, significantly reducing data synchronization latency during real-time access.

[0081] Step 2.3, Optimization of distributed lock mechanism;

[0082] To address concurrent access control in a multi-tenant environment, the system implements a distributed lock mechanism based on tenant isolation. Different tenants use independent lock spaces for data access, avoiding lock contention between tenants. An optimistic locking strategy is employed to handle read-heavy, write-light scenarios, while a pessimistic locking strategy handles write-intensive scenarios. The lock strategy is dynamically selected based on the access pattern, thereby improving the system's concurrency handling capabilities.

[0083] Furthermore, a blockchain-based distributed consensus mechanism can be adopted to automatically execute data consistency rules through smart contracts, thereby further enhancing the credibility and transparency of data consistency.

[0084] Step 3: The hierarchical data is processed using a lightweight edge intelligence algorithm. The deep learning model is compressed into an edge deployment version through knowledge distillation technology to achieve cloud-edge collaborative reasoning and intelligent data processing.

[0085] Specifically, it includes the following:

[0086] Step 3.1, Model Compression and Knowledge Distillation;

[0087] The system transforms complex deep learning models into lightweight versions suitable for edge deployment using knowledge distillation. The original teacher model contains millions of parameters, while the distilled student model contains only tens of thousands of parameters, reducing the model size to one-tenth of the original while maintaining over 95% accuracy. The compression process employs a progressive pruning strategy, gradually removing unimportant network connections to ensure a smooth transition in model performance.

[0088] Model compression adopts knowledge distillation technology to transfer the knowledge of large teacher models to lightweight student models. The goal of the compression process is to maximize the reduction of model parameters and computational complexity while maintaining prediction accuracy.

[0089] The compression optimization goal includes three aspects: model accuracy loss minimization, model size minimization, and inference speed maximization. The system balances these three goals through a multi-objective optimization method, where the accuracy loss weight is usually set to 0.6, the model size weight is set to 0.3, and the inference speed weight is set to 0.1. The optimization process uses a gradient descent algorithm for iterative solution until the preset convergence condition or maximum iteration number is reached.

[0090] Step 3.2, incremental learning mechanism;

[0091] The AI model deployed on the edge node has the ability of incremental learning, which can continuously optimize the model performance using locally collected new data. Incremental learning uses elastic weight integration technology to learn new knowledge while maintaining existing knowledge.

[0092] Before incremental learning, the newly collected data needs to be preprocessed. First, normalize the irradiation parameter data to unify different dimensions to the range of 0-1, the normalization formula is: data minus the minimum value divided by the difference between the maximum and minimum values. For categorical device state data, use one-hot encoding to convert to numerical vector. For time series data, use sliding window method to extract fixed length time series fragments as model input.

[0093] The loss function of elastic weight integration is: new task loss plus regularization term, the regularization term controls the degree of preservation of the original knowledge through importance weight, the importance weight is determined by calculating the influence of the parameter on the original task performance, and the parameter with greater influence obtains higher protection weight.

[0094] The system periodically aggregates the model updates from the edge nodes to the cloud to form a globally optimized model version, which is then distributed to each edge node.

[0095] Step 3.3, collaborative inference architecture;

[0096] The system establishes a cloud-edge collaborative inference architecture, simple real-time decisions are completed on the edge node, complex analysis and calculation are performed on the cloud. The edge node is responsible for device state monitoring, anomaly detection, emergency response, and other latency-sensitive tasks. The cloud is responsible for deep data mining, trend prediction, global optimization, and other computationally intensive tasks. Both exchange data through asynchronous message queues to ensure real-time and computational efficiency of the system.

[0097] Further, a federated learning framework can be used, and multiple edge nodes cooperatively train a shared model to achieve knowledge sharing while protecting data privacy, further improving the generalization ability of the model.

[0098] Step 4: The intelligent processing data is fused and processed using an adaptive quality assurance mechanism to establish a data source reputation scoring system, and cross-validation and fusion processing are performed on different modal data to obtain fused data.

[0099] Specifically, the following content is included:

[0100] Step 4.1: Data source reputation evaluation

[0101] The system establishes a reputation scoring mechanism for each data source and dynamically adjusts the data weight according to the historical data quality performance. The reputation score considers multiple dimensions such as data integrity, accuracy, timeliness, etc. The initial reputation value of a newly added data source is set to a medium level, and it is gradually adjusted as the data quality performance accumulates. Data sources with high reputation values have greater weight in data fusion, and data sources with low reputation values trigger additional quality check processes.

[0102] Before calculating the data source reputation value, different types of quality indicators need to be preprocessed: data integrity, accuracy, timeliness, etc. are normalized to 0-1 to ensure that the indicators are evaluated comprehensively in the same dimension. The data source reputation evaluation uses a weighted average method of historical quality scores, combined with a time decay mechanism to ensure that the latest scores have greater influence. The evaluation process first collects the historical quality scores of the data source, then calculates the time decay weight according to the score time, and finally obtains the reputation value by weighted summation.

[0103] The reputation value calculation method is: multiply each quality score by the corresponding weight and time decay factor, then sum to obtain the final reputation value. The quality score is a weighted average of the scores of data integrity, accuracy, and timeliness, with weights of 0.4, 0.4, and 0.2 respectively. The score weight is determined according to the data size at the time of scoring, and the larger the data size, the higher the weight, which is mapped using a logarithmic function. The time decay factor is usually 0.1, so that the newer scores have greater influence.

[0104] Step 4.2: Multi-modal data fusion algorithm

[0105] The system adopts multi-modal data fusion technology to cross-verify and fuse data from different sensors and devices. Before fusion, the data of different modalities need to be standardized and pre-processed. For numerical sensor data, the Z-score standardization method is used to eliminate dimensional differences, and the standardization formula is: data minus mean divided by standard deviation. For image data, pixel value normalization and size unification processing are performed. For text device log data, word vector encoding is used to convert to numerical vector representation.

[0106] The fusion algorithm is based on the Bayesian inference framework, and considers the reliability and correlation of each data source. Numerical data uses Z-score standardization, categorical data uses one-hot encoding conversion, and time series data uses sliding window normalization processing.

[0107] The weight calculation method of multi-modal fusion is: multiply the reputation value of each data source with its correlation coefficient, then divide by the sum of the product of the reputation value and correlation coefficient of all data sources, to obtain the fusion weight of the data source. The correlation coefficient is calculated by the Pearson correlation coefficient method, and the mutual information method is used for different categories of data sources.

[0108] When significant differences are detected between data sources, the system automatically starts the anomaly diagnosis process to determine the authenticity of the data through additional verification means. The chi-square test method is used for statistical analysis to determine whether there are significant differences between data sources by comparing the difference between observed values and expected values.

[0109] Step 4.3, adaptive data cleaning strategy;

[0110] The system automatically selects a cleaning method based on the data type and quality issues. For numerical data, statistical-based outlier detection and smoothing filtering are used. Outlier detection uses a combination of 3σ criterion and boxplot method, and when the data point deviates from the mean by more than three times the standard deviation, it is determined as an outlier.

[0111] For categorical data, rule-based consistency checking is used. For time series data, pattern recognition-based anomaly detection is used, and the score calculation of time series anomaly detection considers deviation degree, trend degree and pattern matching degree.

[0112] The specific calculation method of each component is as follows: the deviation degree is obtained by calculating the standardized deviation of the current data point and the historical mean value; the trend degree is obtained by linear regression analysis of the change trend of the data in the time window, and the absolute value of the slope is normalized; the pattern matching degree is calculated by using the dynamic time warping algorithm to calculate the similarity between the current sequence and the normal pattern library, and the reciprocal of the maximum similarity is taken as the abnormal index. The weight coefficient is determined by grid search optimization according to the historical abnormal detection effect, and the deviation degree weight is usually 0.4, the trend degree weight is 0.3, and the pattern matching degree weight is 0.3.

[0113] The selection of the cleaning strategy is based on data feature analysis and historical cleaning effect evaluation to ensure the effectiveness and adaptability of the cleaning process.

[0114] Further, the data repair technology based on the generative adversarial network can be used to generate reasonable replacement data for missing or damaged data by learning the distribution characteristics of normal data, thereby improving the integrity and availability of the data.

[0115] Step 5, the fusion data is analyzed by using the knowledge graph driven fault diagnosis method, a knowledge graph containing device entities, fault entities, symptom entities and cause entities is constructed, and a diagnosis result is obtained by fault recognition and root cause analysis through a graph neural network;

[0116] Specifically, the following contents are included:

[0117] Step 5.1, construction of the irradiation device knowledge graph;

[0118] The system constructs a knowledge graph containing device entities, fault entities, symptom entities, cause entities and other node types. The device entity contains device model, technical parameters, maintenance records and other attributes; the fault entity describes various possible fault modes and characteristics; the symptom entity records the external manifestations and data anomaly patterns of the fault; the cause entity analyzes the root cause and influencing factors of the fault. Each entity is connected by "causes", "manifests as" and "influences" to form a complex knowledge network.

[0119] Step 5.2, graph neural network fault recognition;

[0120] The system uses a graph attention network to model the knowledge graph and realize intelligent identification of fault modes. Before graph neural network inference, the input fault feature data needs to be preprocessed: the abnormal values of numerical sensors are normalized to the range of 0-1 using the maximum and minimum value normalization method, the fault symptoms of the classification type are converted to integer indexes using label encoding, and the fault evolution process of the time sequence type is extracted using a sliding window to extract feature sequences.

[0121] The network automatically learns the importance of different nodes and relationships through attention mechanisms, capturing complex patterns of fault propagation. When the system detects abnormal data, the graph neural network searches for matching fault patterns in the knowledge graph, outputting possible fault types and confidence levels.

[0122] The attention weight calculation of fault identification adopts the graph attention network mechanism, which determines the importance of different nodes by calculating the attention coefficients between nodes. The calculation process linearly transforms the node feature vector through the weight matrix, then concatenates the transformed feature vector, calculates the attention score through the attention vector, finally uses the leaky linear rectified activation function for nonlinear transformation, and normalizes the final attention weight through softmax.

[0123] The implementation of the leaky linear rectified activation function is: when the input value is greater than 0, the value is directly output, and when the input value is less than or equal to 0, the value is multiplied by a negative slope coefficient (usually 0.01). This can avoid the gradient vanishing problem of traditional ReLU function in the negative value region, and maintain the learning ability of the network.

[0124] Step 5.3, Causal Reasoning Root Cause Analysis;

[0125] The system uses a causal reasoning algorithm for fault root cause analysis, which determines the root cause by analyzing the causal relationship between fault events. The algorithm constructs a causal graph based on historical fault data, identifying the propagation path of direct and indirect causes.

[0126] The root cause probability calculation adopts the Bayesian inference method, which determines the most likely fault cause by calculating the posterior probability of each root cause given the observation of symptoms. The calculation process is: multiply the conditional probability of each root cause given the symptom by the prior probability of the root cause, and divide by the sum of the product of the conditional probability and the prior probability of all possible root causes.

[0127] The system gives priority to root causes with high probability and large impact, and needs to standardize the coefficients before calculating the importance score: the fault severity is normalized to the 0-1 interval after 1-10 level scoring, and the historical occurrence frequency is normalized after logarithmic transformation, ensuring that indicators of different dimensions can be reasonably integrated. The root cause importance score is obtained by multiplying the root cause posterior probability, fault severity coefficient and fault historical occurrence frequency coefficient.

[0128] The fault severity coefficient is determined according to the impact of the fault on production, including factors such as equipment downtime, product quality impact, safety risk level, etc. After expert scoring method determines the 1-10 level, it is normalized. The frequency coefficient is based on historical fault data statistics, calculating the number of occurrences of the root cause in the past year, and normalized to the 0-1 interval after logarithmic transformation.

[0129] The system provides targeted repair recommendations and preventive measures based on the importance score.

[0130] Step 5.4, dynamic knowledge updating mechanism;

[0131] The system establishes a dynamic knowledge graph updating mechanism, which can learn and extract knowledge from new fault cases. When a new fault or repair case occurs, the system automatically analyzes the fault characteristics, causes, and solutions, and updates the nodes and relationships of the knowledge graph. Through continuous learning, the accuracy and coverage of the knowledge graph are continuously improved, and the fault diagnosis capability is continuously enhanced.

[0132] Further, reinforcement learning can be used to optimize fault diagnosis strategies, learn from interactions with maintenance experts, continuously improve diagnosis processes and decision rules, and improve the accuracy and efficiency of fault diagnosis.

[0133] Step 6, use blockchain technology to store and manage diagnostic results and fusion data, build a consortium blockchain platform, and realize trusted management of data throughout its life cycle through smart contracts;

[0134] Specifically, it includes the following content:

[0135] Step 6.1, consortium blockchain architecture design;

[0136] The system uses consortium blockchain technology to build a data storage platform, and the participating nodes include irradiation service enterprises, customer enterprises, regulatory departments, etc. Each participant runs an independent blockchain node, and maintains data consistency and tamper resistance through consensus algorithms. The blockchain adopts a layered architecture, with the consensus layer at the bottom responsible for data consistency, the smart contract layer in the middle responsible for business logic, and the application layer at the top responsible for user interaction.

[0137] Step 6.2, efficient consensus algorithm;

[0138] In view of the performance requirements of the irradiation industry, the system uses an improved practical Byzantine fault tolerance algorithm as the consensus mechanism. The algorithm optimizes the message transmission process under the premise of ensuring security, reducing consensus delay. The consensus process uses a three-phase commit strategy: pre-preparation phase, preparation phase, and submission phase, to ensure data consistency in the case of network partition or node failure.

[0139] The security threshold of the consensus algorithm requires that the total number of nodes must be greater than or equal to three times the maximum number of Byzantine fault tolerance nodes plus one, that is, the total number of nodes must be at least three times the number of Byzantine fault tolerance nodes plus one node, so that in the worst case, it can still reach the correct consensus.

[0140] Step 6.3, automatic execution of smart contracts;

[0141] The system deploys smart contracts to automatically execute data access control and audit rules. The smart contract defines business logic such as data access permissions, audit trigger conditions, compliance check processes, etc. When a data access request triggers the smart contract, the contract automatically verifies the requester's identity, checks access permissions, and records audit logs to ensure data access compliance and traceability.

[0142] The system designs a smart contract template to automatically execute business processes such as data access control, quality inspection, and fee settlement. The contract uses an event-driven mode, automatically triggering corresponding operations when preset conditions are met, reducing manual intervention.

[0143] Step 6.4, cross-chain data sharing;

[0144] The system supports cross-chain data sharing with other blockchain networks, enabling data interoperability between different blockchains through relay chain technology. The cross-chain protocol uses hash locking and time locking mechanisms to ensure the atomicity and security of cross-chain transactions. This enables trusted data exchange between irradiation data and blockchain systems of upstream and downstream enterprises in the supply chain and regulatory authorities.

[0145] Further, zero-knowledge proof technology can be used to prove the authenticity and integrity of data without revealing sensitive information, further enhancing data privacy protection capabilities.

[0146] In one embodiment of the present application, a multi-tenant data management system for irradiation processing is provided, as shown in Figure 2 includes:

[0147] The anti-radiation data acquisition module is used to acquire raw data during the irradiation processing process. The raw data includes device data using different communication protocols. The device driver adapter framework is used to uniformly convert the device data using different communication protocols into standard data streams.

[0148] The hybrid consistency protocol module is used to process the standard data streams using a hybrid consistency protocol. The data is divided into four levels according to the impact of security and quality, and the consistency strategy is used to process the hierarchical data.

[0149] The lightweight edge intelligence module is used to process the hierarchical data using a lightweight edge intelligence algorithm. The knowledge distillation technology is used to compress the deep learning model into an edge deployment version, and the cloud-edge collaborative reasoning is used to obtain intelligent processing data.

[0150] The adaptive data quality assurance module is used to fuse the intelligent processing data using an adaptive quality assurance mechanism. A data source reputation scoring system is established to cross-validate and fuse different modal data to obtain fused data.

[0151] The knowledge graph fault diagnosis module is configured to analyze the fusion data using a knowledge graph-driven fault diagnosis method, construct a knowledge graph containing device entities, fault entities, symptom entities, and cause entities, and realize fault identification and root cause analysis through a graph neural network to obtain a diagnosis result.

[0152] The blockchain data storage module is configured to store and manage the diagnosis result and the fusion data using blockchain technology, construct a consortium blockchain platform, and realize trusted management of the data throughout its life cycle through a smart contract.

[0153] Application effect verification of the embodiment:

[0154] Application example background:

[0155] A medical device irradiation service enterprise before and after implementing the present application. The enterprise has 8 irradiation devices, serves 120 customers, and processes about 200 irradiation batches per day, involving various product types such as medical masks, surgical instruments, and implant materials.

[0156] Data table display:

[0157] The real-time data collection example is shown in Table 1:

[0158] Table 1: Real-time data collection example;

[0159]

[0160] The multi-tenant data processing example is shown in Table 2:

[0161] Table 2: Multi-tenant data processing example;

[0162]

[0163] Chart display and analysis:

[0164] As shown in Figure 3 , the performance monitoring chart shows the changes in key performance indicators within 24 hours before and after the system is implemented. After implementation, the system response time is significantly reduced from a peak of 1500ms to an average of less than 85ms. CPU and memory usage remain within a reasonable range, and network latency is significantly reduced. The resource isolation and intelligent scheduling mechanism of the multi-tenant architecture effectively improves the overall performance of the system, ensuring the quality of service for each tenant.

[0165] As shown in Figure 4As shown in the diagram, the equipment utilization comparison clearly demonstrates the significant effect of the intelligent scheduling algorithm. After implementation, the utilization rates of various irradiation equipment generally improved, with the average utilization rate increasing from 65.8% to 91.2%. The utilization rates of the Cobalt-60 equipment and the electron beam equipment both exceeded 90%, which is mainly attributed to the intelligent scheduling algorithm based on deep reinforcement learning, which can optimize allocation based on multiple factors such as equipment load, processing priority, and energy consumption costs.

[0166] like Figure 5 As shown, the data quality radar chart illustrates the quality improvement effects before and after system implementation across six dimensions. After implementation, the system achieved significant improvements in data integrity, accuracy, timeliness, consistency, access control, and audit traceability. Multimodal data fusion technology and adaptive quality assurance mechanisms ensured high-quality data management, with access control and audit traceability capabilities approaching perfection, meeting the stringent regulatory requirements of the medical device industry.

[0167] Summary of technical effects:

[0168] This invention presents a multi-tenant data management method for irradiation processing, which achieves efficient, safe, and intelligent data management in irradiation processing environments through the organic combination of six core steps. The system has achieved significant improvements in data response speed, equipment utilization, data quality, and fault early warning, providing a complete technical solution for the digital transformation of the irradiation processing industry, and has significant practical value and promotional significance.

[0169] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for irradiation processing multi-tenant data management, characterized in that, The method comprises the following steps: obtaining original data in the irradiation processing process, the original data including device data in different communication protocols, and uniformly converting the device data in different communication protocols into standard data streams by using a device driver adapter framework; processing the standard data streams by using a hybrid consistency protocol, classifying the data into four levels according to the influence degree of the data on safety and quality, and processing the classified data by using a consistency strategy to obtain classified data; processing the classified data by using a lightweight edge intelligent algorithm, compressing a deep learning model into an edge deployment version by using a knowledge distillation technology, and realizing cloud-edge collaborative reasoning to obtain intelligent processing data; fusing the intelligent processing data by using an adaptive quality guarantee mechanism, establishing a data source credit scoring system, cross-verifying and fusing different modal data to obtain fused data; specifically comprising: data source credit evaluation, establishing a credit scoring system based on historical data quality and device state; a multi-modal data fusion algorithm, cross-verifying and fusing sensor data, image data and log data; an adaptive data cleaning strategy, dynamically adjusting data cleaning rules and thresholds according to data quality evaluation results; wherein the fusion algorithm is based on a Bayesian inference framework and comprehensively considers the reliability and correlation of each data source; the weight calculation method of multi-modal fusion is: multiplying the credit value of each data source with its correlation coefficient, then dividing by the sum of the credit value and correlation coefficient product of all data sources to obtain the fusion weight of the data source; the correlation coefficient is calculated by using the Pearson correlation coefficient method, and the mutual information method is used for different categories of data sources; analyzing the fused data by using a knowledge graph driven fault diagnosis method, constructing a knowledge graph including device entities, fault entities, symptom entities and cause entities, and realizing fault identification and root cause analysis by using a graph neural network to obtain a diagnosis result; storing the diagnosis result and the fused data by using a blockchain technology for evidence management, constructing a consortium blockchain platform, and realizing trusted management of the whole life cycle of data by using a smart contract.

2. The multi-tenant data management method for irradiation processing according to claim 1, characterized in that, The original data in the irradiation processing process comprises: deploying an anti-radiation sensor array, using a redundancy design and a fault detection mechanism to ensure the reliability of data acquisition; realizing heterogeneous device interface adaptation, uniformly converting device data in different communication protocols into standard data streams by using a device driver adapter framework; adjusting the data acquisition frequency according to the irradiation process stage, increasing the acquisition frequency in the key process stage to obtain more detailed process data.

3. The method of claim 1, wherein, The hierarchical processing of the standard data streams by using a hybrid consistency protocol comprises: a data importance classification mechanism, classifying the data into four levels of core safety, key quality, important operation and general statistics according to the influence degree of the data on safety and quality; a predictive data synchronization strategy, predicting data synchronization time based on data access mode and business demand; a distributed lock mechanism optimization, using a hierarchical lock structure to reduce lock competition and deadlock risk.

4. The method of claim 1, wherein, The processing of the classified data by using a lightweight edge intelligent algorithm comprises: model compression and knowledge distillation, compressing a deep learning model into an edge deployment version to maintain model accuracy while reducing computational complexity; Incremental learning mechanism to support continuous learning and parameter updating of models on edge devices; Collaborative reasoning architecture to enable cloud-edge collaborative reasoning and dynamic allocation of computing resources based on task complexity.

5. The method of claim 1, wherein, The fusion data is analyzed by a knowledge graph driven fault diagnosis method, which includes: Knowledge graph construction: a multi-level knowledge graph containing device entities, fault entities, symptom entities, and cause entities is established; Graph neural network fault identification: graph convolution network is used to model device status and identify fault patterns; Causal reasoning root cause analysis: fault root cause analysis and propagation path tracking based on causal graph model; Dynamic knowledge updating mechanism: continuously update the knowledge graph based on new fault cases and expert knowledge.

6. The method of claim 1, wherein, The device data of different communication protocols is uniformly converted into standard data streams by using a device driver adapter framework, which includes: Unified interface layer: convert device data of different communication protocols into standard data streams; Protocol conversion module: support data format conversion of multiple industrial communication protocols; Data standardization processing: ensure consistent representation of heterogeneous device data.

7. The method of claim 1, wherein, Intelligent processing data is obtained by implementing cloud-edge collaborative reasoning, which includes: Task allocation strategy: dynamically allocate reasoning tasks based on computational complexity and network delay; Edge caching mechanism: cache common models and intermediate results on edge devices; Collaborative optimization algorithm: coordinate the configuration of computing resources on the cloud and edge devices.

8. The method of claim 1, wherein, The trusted management of data throughout its life cycle is realized through smart contracts, which includes: Data access control contract: define different tenants' data access permissions and operation rules; Quality evaluation contract: automatically perform data quality evaluation and reputation score update; Audit traceability contract: record data operation history and responsibility traceability information.

9. A radiation processing multi-tenant data management system, comprising: A multi-tenant data management method for performing irradiation processing, as claimed in any one of claims 1-8, includes: Anti-radiation data acquisition module: used to obtain original data during irradiation processing, the original data including device data of different communication protocols, and the device data of different communication protocols is uniformly converted into standard data streams by using a device driver adapter framework; Hybrid consistency protocol module: used to process the standard data streams using a hybrid consistency protocol, divide the data into four levels according to the influence of the data on security and quality, and process the data using a consistency strategy to obtain hierarchical data; Lightweight edge intelligence module: used to process the hierarchical data using a lightweight edge intelligence algorithm, compress a deep learning model into an edge deployment version through knowledge distillation technology, and obtain intelligent processing data by implementing cloud-edge collaborative reasoning; Adaptive data quality assurance module: used to fuse the intelligent processing data using an adaptive quality assurance mechanism, establish a data source reputation scoring system, cross-verify and fuse different modal data to obtain fusion data; Knowledge graph fault diagnosis module: used to analyze the fusion data using a knowledge graph driven fault diagnosis method, construct a knowledge graph containing device entities, fault entities, symptom entities, and cause entities, and obtain diagnosis results through graph neural network fault identification and root cause analysis. The blockchain data evidence module is used for storing and managing the diagnosis result and the fusion data by using the blockchain technology, constructing a consortium blockchain platform, and realizing the trusted management of the whole life cycle of data through a smart contract.

Images