A guide rail thermal compensation optimization method and system based on digital twinning

Abstract

This invention relates to the field of intelligent manufacturing technology and discloses a guide rail thermal compensation optimization method and system based on digital twins. The method includes: constructing a unified knowledge representation by mapping digital twins of heterogeneous devices; deploying an intelligent agent network to achieve knowledge exchange based on interpretable homomorphic encryption; designing a hierarchical federated learning framework and applying a heterogeneous model aggregation algorithm to construct a global knowledge model; identifying and processing distributed anomaly patterns based on an encrypted multi-party consensus algorithm; and applying interpretable AI technology to generate knowledge updates and execute personalized deployments. This invention solves the problem of data structure differences between heterogeneous devices, achieves knowledge sharing under the premise of privacy protection, improves thermal compensation accuracy, enhances system decision transparency, and enables the manufacturing system to form a continuous learning and evolution capability, fundamentally improving the quality and efficiency of precision manufacturing.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and more specifically, to a guide rail thermal compensation optimization method and system based on digital twins. Background Technology

[0002] With the rapid development of intelligent manufacturing, thermal deformation compensation technology for precision machining equipment has become a key factor in ensuring machining accuracy. Traditional thermal compensation methods mainly rely on local data acquisition and model building for a single device. Each device independently performs thermal deformation analysis and compensation strategy optimization, which cannot utilize the experience and knowledge of other devices, resulting in low learning efficiency and limited optimization effects.

[0003] In recent years, digital twin technology has provided a new approach to mapping virtual and physical systems in manufacturing. However, existing digital twin applications mainly focus on condition monitoring and predictive maintenance of single devices or homogeneous systems, lacking effective mechanisms for knowledge sharing between heterogeneous devices. Meanwhile, although machine learning technology has made some progress in optimizing heat compensation models, traditional centralized learning methods are difficult to apply directly to heterogeneous data structures and model representation differences between devices from different manufacturers and of different models.

[0004] Furthermore, in industrial environments, the increasing demand from equipment manufacturers and users for the protection of proprietary data and algorithms further limits the sharing of thermal compensation knowledge across devices and vendors. Existing data encryption and privacy protection technologies often reduce data usage efficiency or sacrifice model accuracy, failing to simultaneously meet the dual requirements of privacy protection and efficient knowledge sharing. Therefore, how to achieve efficient sharing and collaborative optimization of thermal compensation knowledge among heterogeneous devices while protecting data privacy has become a key technical challenge that needs to be addressed. Summary of the Invention

[0005] This invention provides a guide rail thermal compensation optimization method and system based on digital twins, which solves the technical problem of how to achieve efficient sharing and collaborative optimization of thermal compensation knowledge among heterogeneous devices while protecting data privacy.

[0006] This invention provides a guide rail thermal compensation optimization method based on digital twins, comprising the following steps:

[0007] Construct digital twin mappings for heterogeneous devices to generate a unified knowledge representation format and a bidirectional mapping function set;

[0008] Based on a unified knowledge representation and mapping function, an intelligent agent network is deployed and an interpretable homomorphic encryption knowledge exchange protocol is implemented to establish a secure communication channel and encrypted knowledge flow;

[0009] By leveraging the established encrypted knowledge flow, a hierarchical federated learning framework comprising a global sharing layer and a local adaptation layer is designed, and a heterogeneous model aggregation algorithm is applied to construct a global knowledge model and a personalized adaptation layer.

[0010] Based on the local observations and global knowledge models of each intelligent agent, an encrypted multi-party consensus algorithm is applied to identify and process distributed anomaly patterns, and to generate a global anomaly assessment and collaborative processing scheme.

[0011] Based on the global knowledge model and local device status, explainable AI technology is applied to generate explainable knowledge update packages and perform personalized deployments.

[0012] In a preferred embodiment, the step of constructing a heterogeneous device digital twin mapping and generating a unified knowledge representation format and a bidirectional mapping function set includes:

[0013] The thermal deformation data structure of each type of equipment is analyzed, and a data structure feature vector is established.

[0014] Analyze the algorithm structure, parameter configuration, and decision logic of the compensation model for each device, and construct the model structure descriptor;

[0015] Semantic alignment algorithms are applied to transform the data structures and model representations of different devices into a unified semantic space;

[0016] Based on the semantic alignment results, a unified knowledge representation format is constructed;

[0017] For each device type, construct a set of bidirectional mapping functions between a unified representation and the original representation.

[0018] In a preferred embodiment, the steps of deploying the agent network and implementing an interpretable homomorphic encryption knowledge exchange protocol include:

[0019] Deploy a corresponding digital twin intelligent agent for each physical device;

[0020] Based on the physical distribution and functional correlation of devices, construct the intelligent agent network topology;

[0021] Design and implement a knowledge exchange protocol based on homomorphic encryption;

[0022] A secure communication channel is established based on a knowledge exchange protocol to enable encrypted knowledge flow exchange between devices;

[0023] Design and implement a knowledge exchange permission management mechanism based on roles and tasks.

[0024] In a preferred embodiment, the steps of designing a hierarchical federated learning framework comprising a global sharing layer and a local adaptation layer include:

[0025] Design a hierarchical federated learning architecture that includes a global sharing layer and a local adaptation layer;

[0026] Design a heterogeneous model aggregation algorithm capable of handling knowledge from different structural models;

[0027] A global knowledge model is constructed based on encrypted knowledge flow and heterogeneous model aggregation algorithms.

[0028] Develop personalized adaptation layers for each device type;

[0029] An incremental learning mechanism is designed to enable the system to continuously learn from new thermal compensation experiences.

[0030] In a preferred embodiment, the step of using an encrypted multi-party consensus algorithm to identify and process distributed anomaly patterns includes:

[0031] Deploy an anomaly detection model on each agent;

[0032] Design a multi-party consensus algorithm based on homomorphic encryption;

[0033] Based on the consensus results from multiple parties, a global anomaly assessment report is generated;

[0034] Based on the global anomaly assessment, a collaborative handling plan is formulated;

[0035] The identified anomaly patterns and their handling experience are stored in the anomaly knowledge base.

[0036] In a preferred embodiment, the step of applying explainable AI technology to generate explainable knowledge update packages and performing personalized deployment includes:

[0037] The application of interpretable AI technology is used to analyze the global knowledge model;

[0038] Based on the extracted decision logic, generate knowledge explanation documents that are understandable to humans;

[0039] Package the model update and the corresponding explanation document into an interpretable update package;

[0040] Generate personalized deployment guidelines based on the status and characteristics of specific devices;

[0041] Design and deploy a feedback mechanism to collect data on the effects of the update package deployment.

[0042] In a preferred embodiment, the digital twin agent includes:

[0043] The local data management module is responsible for the collection, cleaning, and storage of local thermal deformation data;

[0044] The knowledge transformation module applies mapping functions to perform knowledge representation transformation;

[0045] The secure communication module is responsible for encrypted communication and knowledge exchange;

[0046] The decision execution module executes the heat compensation strategy and provides feedback on the results.

[0047] In a preferred embodiment, the hierarchical federated learning architecture includes:

[0048] The global shared layer module is responsible for learning and sharing general thermal compensation knowledge;

[0049] The local adaptation layer module is responsible for personalized adjustments based on the characteristics of specific devices;

[0050] The inter-layer communication module is responsible for information exchange between the global layer and the local layer.

[0051] The security isolation module ensures that proprietary device data is not directly exposed.

[0052] In a preferred embodiment, the encrypted multi-party consensus algorithm includes:

[0053] Local anomaly description: Converts local anomaly detection results into a standard description format;

[0054] Encrypted sharing involves homomorphically encrypting the anomaly description and sharing it with the participants.

[0055] Encrypted voting involves each participant evaluating and voting on abnormal patterns in encrypted form.

[0056] Threshold determination: Determine whether a global anomaly exists based on the voting results and a preset threshold.

[0057] Consensus is formed by generating a consensus description for patterns that are determined to be globally abnormal.

[0058] In a preferred embodiment, a digital twin-based guide rail thermal compensation optimization system is used to execute a digital twin-based guide rail thermal compensation optimization method, comprising:

[0059] The heterogeneous device digital twin mapping module is used to generate a unified knowledge representation format and a bidirectional mapping function set;

[0060] The agent network deployment module is used to deploy digital twin agents and realize knowledge exchange based on interpretable homomorphic encryption;

[0061] The hierarchical federated learning framework module is used to handle the integration of heterogeneous knowledge and the construction of a global knowledge model;

[0062] The distributed anomaly handling module is used to identify and handle anomaly patterns based on a cryptographic multi-party consensus algorithm.

[0063] The Explainable Knowledge Update module is used to generate explainable knowledge update packages and perform customized deployments.

[0064] The beneficial effects of this invention are as follows:

[0065] The unified knowledge representation and bidirectional mapping mechanism solves the problem of differences in data structures and model representations between heterogeneous devices, enabling knowledge exchange between devices from different manufacturers and of different models in a unified semantic space, significantly improving the feasibility and efficiency of cross-device knowledge sharing. This unified representation mechanism allows previously isolated device thermal compensation knowledge to be integrated and utilized, forming a more comprehensive understanding of thermal deformation and compensation strategies.

[0066] The knowledge exchange protocol based on interpretable homomorphic encryption effectively solves the data privacy protection problem, enabling device manufacturers and users to securely share heat compensation knowledge without exposing raw data and proprietary algorithms. This encryption mechanism eliminates privacy concerns in knowledge sharing, significantly increases the willingness and extent of all parties to participate in collaborative optimization, and creates conditions for broader knowledge sharing.

[0067] The hierarchical federated learning framework and heterogeneous model aggregation algorithm organically combine global knowledge with local characteristics, ensuring both the universality of shared knowledge and the adaptability to specific device features. This hierarchical architecture enables global optimization and personalized adaptation to mutually promote each other, forming a virtuous cycle and continuously improving the overall system's thermal compensation performance.

[0068] Explainable AI-generated knowledge interpretations enhance the transparency and credibility of system decisions, enabling engineers to understand and verify automatically generated thermal compensation strategies and promoting human-machine collaborative optimization. This explainability mechanism not only improves system maintainability but also accelerates the absorption and application of new knowledge, enabling the entire manufacturing system to continuously learn and evolve in the field of thermal compensation, fundamentally improving the quality and efficiency of precision manufacturing. Attached Figure Description

[0069] Figure 1 This is a flowchart of a guide rail thermal compensation optimization method based on digital twins according to the present invention;

[0070] Figure 2 This is a bar chart comparing the thermal deformation modes of equipment from different manufacturers according to the present invention;

[0071] Figure 3 This is a network diagram of the digital twin intelligent agent network topology of the present invention;

[0072] Figure 4 This is a line graph showing the iterative optimization effect of hierarchical federated learning in this invention;

[0073] Figure 5This is a radar chart comparing the multi-dimensional performance indicators of different methods of the present invention;

[0074] Figure 6 This is a bar chart of the heterogeneous model aggregation weight factor analysis of the present invention;

[0075] Figure 7 This is a circular diagram showing the distribution of distributed exception mode types in this invention. Detailed Implementation

[0076] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0077] At least one embodiment of the present invention discloses a guide rail thermal compensation optimization method based on digital twin, such as... Figure 1 As shown, it includes the following steps:

[0078] Step 1: Construct a digital twin mapping for heterogeneous devices to generate a unified knowledge representation format and a set of bidirectional mapping functions;

[0079] Specifically, the following steps are included:

[0080] Step 1.1, Analysis of equipment thermal deformation data structure;

[0081] The thermal deformation data structure of each device is analyzed, including but not limited to sensor type, sampling frequency, data format, and preprocessing method, to establish a data structure feature vector. ,in Indicates the device index;

[0082] Data structure feature vector Before construction, normalization processing is required to scale features of different dimensions (such as the number of sensors, sampling frequency, accuracy, etc.) to the [0, 1] interval, and one-hot encoding conversion is performed on categorical features (such as sensor type, data format, etc.).

[0083] like Figure 2The figure shows a comparison of the thermal deformation of three devices from different manufacturers (A, B, and C) at different points along the Z-axis of the guide rail. The grouped bar charts provide a clear comparison of the differences in thermal deformation distribution among the different manufacturers, highlighting the necessity of constructing a digital twin mapping of heterogeneous devices in step 1. The figure reveals significant differences in the thermal deformation patterns of the three manufacturers' devices, with manufacturer B exhibiting larger thermal deformation at most locations. This is a crucial foundation for unified knowledge representation.

[0084] Step 1.2, Compensation model structure mapping;

[0085] Analyze the algorithm structure, parameter configuration, and decision logic of the compensation models for each device, and construct a model structure descriptor. .

[0086] Step 1.3, application of semantic alignment algorithm;

[0087] Design and apply a semantic alignment algorithm, which includes the following processing steps:

[0088] Feature extraction steps: Extract key features and attributes from the data structures of different devices;

[0089] Semantic mapping steps: Constructing mapping relationships between features and establishing semantic connections;

[0090] Transformation verification steps: Ensure semantic consistency through two-way verification;

[0091] Optimization and adjustment steps: Iteratively optimize mapping parameters to improve conversion accuracy.

[0092] Furthermore, the semantic alignment algorithm also includes a terminology standardization step to unify the technical terms and expressions used by different vendors, ensuring consistency in semantic understanding.

[0093] Step 1.4, Generation of unified knowledge representation;

[0094] Based on the semantic alignment results, a unified knowledge representation format is constructed. This format can universally express the thermal deformation characteristics and compensation strategies of various devices;

[0095] The unified knowledge representation process standardizes the thermal deformation characteristic parameters of each device to eliminate the influence of differences in the number and accuracy of sensors between different devices. At the same time, it converts discrete compensation strategy decision parameters into continuous numerical representations to achieve a unified expression of thermal compensation knowledge among heterogeneous devices.

[0096] Step 1.5, construct the bidirectional mapping function;

[0097] For each device type, construct a set of bidirectional mapping functions between the unified representation and the original representation:

[0098] ;

[0099] in, This represents a complete set of bidirectional mapping functions, including mapping functions for all device types. An index identifier representing a specific device; Indicates from device A forward mapping function that converts the original representation format to the unified knowledge representation format; This indicates the conversion from the unified knowledge representation format back to the device. The inverse mapping function of the original representation format; A unified knowledge representation space is a standardized knowledge representation format shared by all devices.

[0100] The mapping function here It is determined by the feature selection function and transformation function Composed of multiple components, including Indicates from device A forward mapping function that converts the original representation format to the unified knowledge representation format; Indicates device The original data or knowledge representation; This represents a function that extracts key features from the raw data, used to filter and retain the feature parameters most valuable for thermal compensation; This represents a function that transforms extracted features into a unified representation space, responsible for standardizing and normalizing features across different devices; Indicates from device raw data The set of key features extracted from it; This represents the result of transforming the extracted features into a unified representation space using a transformation function.

[0101] It is determined by the projection function and refactoring function Composed of multiple components, including This indicates the conversion from the unified knowledge representation format back to the device. The inverse mapping function of the original representation format; Represents data or knowledge representations in a unified knowledge representation space; This indicates that a unified representation will be projected onto the device. A function for a specific feature space, responsible for converting a general representation into a feature form suitable for a specific device; Indicates a reconstruction device based on projection features The raw representation function is responsible for converting features back to a raw format that the device can directly use; This indicates that data in the unified representation space will be used. Projected onto device The result after a specific feature space; This indicates that the projected features are converted back to the device using a reconstruction function. The final result of the original representation format.

[0102] Through the above steps, a unified representation and conversion mechanism for heterogeneous device data and models has been completed, laying the foundation for subsequent security knowledge exchange, sharing, and optimization.

[0103] Step 2: Based on unified knowledge representation and mapping functions, deploy an agent network and implement an interpretable homomorphic encryption knowledge exchange protocol to establish a secure communication channel and encrypted knowledge flow;

[0104] Specifically, the following steps are included:

[0105] Step 2.1, Deployment of the device's digital twin intelligent agent;

[0106] Deploy a corresponding digital twin agent for each physical device. This digital twin agent includes:

[0107] Local data management module: responsible for the collection, cleaning, and storage of local thermal deformation data;

[0108] Knowledge transformation module: The mapping function constructed in step 1 is used to perform knowledge representation transformation;

[0109] Secure communication module: responsible for encrypted communication and knowledge exchange;

[0110] Decision execution module: Executes the heat compensation strategy and provides feedback on the results.

[0111] Step 2.2, Construction of agent network topology;

[0112] Based on the physical distribution and functional correlation of devices, a network topology of intelligent agents is constructed, and the communication relationships and weights between intelligent agents are determined.

[0113] Step 2.3, Application of interpretable homomorphic encryption technology;

[0114] Design and implement a knowledge exchange protocol based on homomorphic encryption, which includes the following processing steps:

[0115] Data preprocessing steps: Convert the knowledge to be shared into a format suitable for homomorphic encryption;

[0116] Encryption steps: Encrypt the data using a homomorphic encryption algorithm;

[0117] Ciphertext computation steps: Perform necessary computational operations while the text is encrypted;

[0118] Proof generation steps: Generate a document proving the legality of the operation;

[0119] Decryption process: The result is decrypted only by the authorized recipient.

[0120] Furthermore, the interpretable homomorphic encryption technology also includes a security level adaptive adjustment step, which dynamically adjusts the encryption strength according to data sensitivity and computational requirements to balance security and computational efficiency.

[0121] Furthermore, the interpretable homomorphic encryption technology also includes a selective knowledge-sharing step, which allows data owners to precisely control the scope, granularity, and recipient range of the shared knowledge.

[0122] Step 2.4, Establish a secure communication channel;

[0123] Establish a secure communication channel based on the above protocol. To enable encrypted knowledge flow between devices exchange.

[0124] Step 2.5, Knowledge Exchange Access Control;

[0125] Design and implement a knowledge exchange permission management mechanism based on roles and tasks to control the scope and level of knowledge sharing among intelligent agents.

[0126] Through the above steps, an intelligent agent network that balances data privacy protection and knowledge sharing has been established, resolving the constraint that raw data cannot be directly shared.

[0127] like Figure 3 The diagram illustrates the network topology of the intelligent agent deployed in step 2, including the communication relationships between the central coordinating node, cluster nodes from various vendors, and specific device nodes. The lines connecting nodes represent communication channels, and the values ​​on the lines represent communication weights. The diagram shows that devices are primarily managed in clusters grouped by vendor, with higher communication weights (above 0.7) between devices from the same vendor and lower weights (below 0.35) between devices from different vendors. This hierarchical network structure helps improve the efficiency of knowledge exchange.

[0128] Step 3: Using the established encrypted knowledge flow, design a hierarchical federated learning framework that includes a global sharing layer and a local adaptation layer, and apply a heterogeneous model aggregation algorithm to construct a global knowledge model and a personalized adaptation layer;

[0129] Specifically, the following steps are included:

[0130] Step 3.1, Design of a hierarchical federated learning architecture;

[0131] The design incorporates a hierarchical federated learning architecture that includes a global sharing layer and a local adaptation layer. This hierarchical federated learning architecture comprises the following component modules:

[0132] Global sharing layer module: responsible for learning and sharing general thermal compensation knowledge, including shared parameter storage unit, general feature extraction unit and global decision unit;

[0133] Local adaptation layer module: responsible for personalized adjustments based on specific device characteristics, including parameter fine-tuning unit, device feature matching unit and feedback optimization unit;

[0134] Inter-layer communication module: responsible for information exchange between the global layer and the local layer, including uplink aggregation unit and downlink distribution unit;

[0135] Security isolation module: Ensures that proprietary device data is not directly exposed and only necessary model updates are shared.

[0136] Furthermore, the hierarchical federated learning architecture also includes a knowledge caching module for temporarily storing frequently used knowledge fragments, thereby improving model response speed.

[0137] like Figure 4 The figure shows the iterative optimization effect of the hierarchical federated learning framework in step 3, tracking the changing trend of thermal deformation compensation deviations of devices from manufacturers A, B, and C during eight iterations. As can be seen from the figure, with the increase in the number of iterations, the thermal compensation accuracy of each manufacturer's device significantly improves, and the compensation deviation gradually decreases and tends to stabilize. In particular, the improvement is most significant for manufacturer B's device, which decreased from an initial 15.3 μm to a final 5.7 μm. This demonstrates that the heterogeneous model aggregation algorithm and the hierarchical federated learning architecture can effectively improve thermal compensation accuracy.

[0138] Step 3.2, Design of heterogeneous model aggregation algorithm;

[0139] Design a heterogeneous model aggregation algorithm capable of handling knowledge from different structural models. This heterogeneous model aggregation algorithm includes the following processing steps:

[0140] Model structure analysis steps: Identify and analyze the structural features and parameter types of different models;

[0141] Knowledge distillation steps: Extract key knowledge and core parameters from complex models, and apply distillation functions:

[0142] ;

[0143] in, Indicates the model The result obtained after knowledge distillation; Indicates the first The original complex model of the device; Indicates from the model The feature extraction function for extracting knowledge is responsible for extracting key features from complex models; This represents a knowledge compression function that transforms extracted features into a compact representation, making them easier to transmit and integrate. Indicates the model Specific parameters are used to adjust the feature extraction and compression intensity during the distillation process.

[0144] Spatial transformation steps: Map the parameters of different models to a unified representation space using a transformation function:

[0145] ;

[0146] in, This indicates the representation of knowledge. The result of converting to a unified representation space; Representation Model The original knowledge representation includes the model's parameters, structure, and decision logic; This represents a function that maps raw knowledge to an intermediate representation, responsible for extracting the essential features of the knowledge and eliminating model-specific representation methods; This represents a function that transforms intermediate representations into a unified representation space, ensuring that knowledge from different sources is comparable and integrateable within the same semantic space.

[0147] Intermediate representation: It is a transitional form that lies between the original knowledge representation and the unified representation space, making it easier to handle complex transformation logic.

[0148] Weight calculation steps: Calculate the contribution weight of each model based on device performance and data quality, using the weighting function:

[0149] ;

[0150] in, Indicates the first The contribution weight of each device model determines the degree of influence of that model in knowledge aggregation; These represent device performance indicators, including hardware performance parameters such as computing power, response speed, and stability. Data quality indicators include parameters that measure the reliability of data, such as data integrity, accuracy, timeliness, and sampling frequency. This represents a historical contribution indicator, reflecting the effectiveness and value of the equipment in providing model updates in the past; This represents the comprehensive evaluation function, an algorithmic function that integrates three indicators into a single weight value.

[0151] Indicators of different dimensions before calculation , , Standardize the data to convert it into a standard distribution with a mean of 0 and a standard deviation of 1, to ensure that different indicators are comparable in weight calculation.

[0152] Weighted aggregation steps: Integrate the knowledge from each model according to its weights to form a unified model, and then use the aggregation function.

[0153] ;

[0154] in, This represents a weighted aggregation function, used to integrate knowledge from multiple models into a unified model; , , Representing the model respectively , , The unified representation of knowledge is the knowledge representation after representation space transformation; , , Representing the model respectively , , The contribution weight determines the degree of influence of the model in the final aggregation result; This represents the total number of models participating in the aggregation. This represents an alignment function that ensures knowledge from different sources is comparable and aggregateable in the same dimension and semantic space. This represents a weighted sum of the knowledge from all models; the higher the weight of a model, the greater its contribution to the final aggregated result.

[0155] Furthermore, the heterogeneous model aggregation algorithm also includes a conflict detection and resolution step, used to identify and process conflicting knowledge from different models to ensure the consistency of the final aggregation result;

[0156] Furthermore, the heterogeneous model aggregation algorithm also includes a model adaptability evaluation step, which evaluates the applicability of the aggregated model on different devices, providing a basis for subsequent personalized adjustments.

[0157] Step 3.3, Global Knowledge Model Construction;

[0158] Based on encrypted knowledge flow Combine heterogeneous model aggregation algorithms to construct a global knowledge model The global knowledge model contains the following component modules:

[0159] Shared knowledge representation module: Stores general heat compensation knowledge parameters and rules;

[0160] Inference Engine Module: Performs hot-compensation strategy inference and decision-making;

[0161] Update management module: handles the integration of new knowledge and the elimination of redundant knowledge;

[0162] Consistency maintenance module: Ensures the consistency and integrity of knowledge.

[0163] Furthermore, the global knowledge model also includes a version control module for tracking and managing the history of knowledge updates, supporting version rollback and difference comparison.

[0164] Step 3.4, Development of the Personalized Adaptation Layer;

[0165] Develop personalized adaptation layers for each device type This personalized adaptation layer includes the following component modules:

[0166] Device characteristics module: Stores and manages the characteristic descriptions of specific devices;

[0167] Parameter fine-tuning module: Adjusts global model parameters according to device characteristics;

[0168] Priority management module: Determines the application priority of global knowledge and local knowledge;

[0169] Performance monitoring module: Monitors the adaptation effect in real time and collects feedback data.

[0170] Furthermore, the personalized adaptation layer also includes a fast response cache module for storing frequently used hot compensation schemes to reduce computational latency.

[0171] like Figure 5 As shown in the figure, the performance of the method in this patent is compared with that of traditional federated learning and distributed learning in six key performance dimensions. As can be seen from the figure, the method has significant advantages in terms of hot compensation accuracy, data privacy protection, heterogeneous model compatibility and decision transparency, especially in heterogeneous model compatibility (91 points) and decision transparency (89 points), which far exceed traditional federated learning (53 points and 36 points respectively).

[0172] Step 3.5, implementation of incremental learning mechanism;

[0173] An incremental learning mechanism is designed to enable the system to continuously learn from new heat compensation experiences and constantly optimize the global knowledge model.

[0174] Through the above steps, a hierarchical federated learning framework capable of handling the integration of heterogeneous knowledge was constructed, overcoming the limitations of traditional federated learning in heterogeneous model scenarios.

[0175] like Figure 6As shown in the figure, the weight calculation factors of the heterogeneous model aggregation algorithm in step 3 are analyzed, and the performance of the three vendor equipment groups (A, B, and C) in three dimensions—data quality score, model performance score, and historical contribution score—is presented. The chart shows that vendor A's equipment group performs best in historical contribution (0.90), vendor B's equipment group leads in model performance (0.92), while vendor C's equipment group has the highest performance in data quality (0.91). This multi-dimensional weight calculation ensures the fairness and effectiveness of heterogeneous model aggregation.

[0176] Step 4: Based on the local observations and global knowledge models of each agent, apply an encrypted multi-party consensus algorithm to identify and process distributed anomaly patterns, and generate a global anomaly assessment and collaborative processing scheme.

[0177] Specifically, the following steps are included:

[0178] Step 4.1, Deploy the local anomaly detection model;

[0179] An anomaly detection model is deployed on each agent. This local anomaly detection model contains the following component modules:

[0180] Data preprocessing module: Cleans and standardizes the raw thermal deformation data;

[0181] Feature extraction module: Extracts key features from thermal deformation data;

[0182] Anomaly scoring module: scores the degree of anomaly of the data. Before scoring the degree of anomaly, the measurement data of each sensor is minimized.

[0183] Maximum normalization ensures that differences in data volume between different sensors do not affect anomaly scoring;

[0184] Threshold adaptive module: dynamically adjusts the anomaly detection threshold based on historical data;

[0185] Report generation module: Generates anomaly detection reports and preliminary analysis results.

[0186] Furthermore, the local anomaly detection model also includes an anomaly classification module, which can classify detected anomalies according to type, severity, and possible causes.

[0187] Step 4.2, Design of encrypted multi-party consensus algorithm;

[0188] Design a multi-party consensus algorithm based on homomorphic encryption. This encrypted multi-party consensus algorithm includes the following processing steps:

[0189] Local anomaly description steps: Convert the local anomaly detection results into a standard description format using the description function:

[0190] ;

[0191] in, This represents a local anomaly description function, used to convert raw anomaly detection results into a standardized description format; Indicates device The original anomaly detection results include various types of indicators, such as temporal fluctuation amplitude, temperature change rate, and location offset value. This represents the anomaly feature extraction function, used to extract key features from the original anomaly detection results; This represents the anomaly severity assessment function, used to evaluate the severity of anomalies; Indicates the index identifier of the device.

[0192] Original anomaly detection results It includes various types of indicators, such as time-series fluctuation amplitude, temperature change rate, and location offset value. Before processing, all numerical indicators are subjected to min-max normalization (scaling the values ​​to the [0, 1] interval), and categorical anomaly types are encoded and converted (converting text categories into numerical representations).

[0193] Encrypted sharing steps: Homomorphically encrypt the exception description and share it with the participants, using the encryption function:

[0194] ;

[0195] in, This represents the overall encryption function, which converts standardized exception descriptions into encrypted form; Indicates device Standardized anomaly descriptions, including anomaly characteristics and severity assessment results; This represents a preprocessing function that performs format conversion and padding on the exception description to make it suitable for homomorphic encryption algorithms. This represents a homomorphic encryption function that supports specific algebraic operations in the ciphertext state; Indicates device The encryption key is used to protect data security; Indicates the index identifier of the device.

[0196] Encrypted voting steps: Each participant evaluates and votes on the abnormal pattern in encrypted form, using an encrypted voting function:

[0197] ;

[0198] in, This represents an encrypted voting function used to aggregate votes from multiple parties in a ciphertext state. Indicates device The encrypted anomaly description includes encrypted anomaly characteristics and evaluation data; , , They represent the first , , A set of encrypted anomaly descriptions for each participating device; Indicates the number of participating devices; Indicates device The voting strategy parameters determine the weight and decision rules of the device in the voting process; This represents the evaluation operation function in the encrypted state, which can evaluate and calculate encrypted data without decryption; Indicates all The evaluation results of each participating device are aggregated and calculated. Indicates the device index participating in the vote, ranging from 1 to .

[0199] Threshold determination steps: Based on the voting results and the preset threshold, determine whether a global anomaly exists using the determination function:

[0200] ;

[0201] in, This represents the threshold determination function used to determine whether a global anomaly exists. It indicates the encrypted voting results, including encrypted assessment data of all devices regarding the anomaly. This represents a preset threshold, which is a predefined judgment standard used to distinguish between normal and abnormal states. This represents the decryption function, used to convert the encrypted voting results back to plaintext. This represents a judgment function that compares the decrypted voting results with a preset threshold and outputs the final anomaly judgment result. Indicates the results of the encrypted vote The plaintext voting data obtained after decryption.

[0202] Consensus formation steps: Generate a consensus description for patterns determined to be globally abnormal, and use a consensus function:

[0203] ;

[0204] in, This represents the consensus-forming function, used to generate a unified description of abnormal patterns; This indicates the anomaly determination result, representing a Boolean value or score indicating whether a global anomaly exists. , , They represent the first , , A set of encrypted anomaly descriptions for each participating device; Indicates the number of participating devices; Indicates device The encrypted anomaly description includes encrypted anomaly characteristics and evaluation data; This represents the decryption function, used to convert encrypted exception descriptions into plaintext. Indicates the equipment The plaintext anomaly data obtained after decrypting the encrypted anomaly description; This represents the set of anomaly descriptions after decryption from all devices; This represents an anomaly description integration function, used to merge anomaly descriptions from multiple devices into a unified consensus description; The device index identifier ranges from 1 to .

[0205] Furthermore, the encrypted multi-party consensus algorithm also includes a dynamic weight adjustment step, which dynamically adjusts the voting weight of the device in the consensus process based on the historical anomaly detection accuracy of the device.

[0206] Furthermore, the encrypted multi-party consensus algorithm also includes a minority anomaly protection step to prevent important but only present anomaly patterns from being ignored.

[0207] Step 4.3, Global Anomaly Assessment Generation;

[0208] Based on the consensus results from multiple parties, a global anomaly assessment report is generated. This includes the type of anomaly, its severity, and possible causes.

[0209] Step 4.4, Development of a collaborative processing plan;

[0210] Based on the global anomaly assessment, a collaborative processing plan is formulated. This includes coordinated and consistent compensation and adjustment measures that should be taken for each piece of equipment.

[0211] Step 4.5, Construction of the anomaly knowledge base;

[0212] The identified anomaly patterns and their handling experience are stored in an anomaly knowledge base to improve the efficiency of future anomaly identification and handling.

[0213] Through the above steps, a mechanism for distributed anomaly identification and handling is realized under the premise of protecting data privacy, thereby improving the system's ability to cope with complex anomalies.

[0214] like Figure 7The figure shows the distribution of distributed anomaly patterns identified in step 4 based on the consensus mechanism. As can be seen from the chart, sensor failure (28%) and sudden environmental changes (23%) are the two main anomaly patterns, followed by uneven thermal distribution (19%) and mechanical collision (15%), while overload operation (9%) and system latency (6%) have relatively smaller proportions. This anomaly pattern analysis helps the system develop more targeted collaborative processing solutions, improving the overall system reliability and stability.

[0215] Step 5: Based on the global knowledge model and local device status, apply explainable AI technology to generate an explainable knowledge update package and perform personalized deployment;

[0216] Specifically, the following steps are included:

[0217] Step 5.1 explains the application of AI technology;

[0218] Applying explainable AI technology to global knowledge models The analysis, and the explainable AI technology, includes the following processing steps:

[0219] Model structure analysis steps: Identify the key structures and decision paths of the model, and use structure analysis functions:

[0220] ;

[0221] in, This represents the overall structural analysis function, used to comprehensively analyze the internal structure of the global knowledge model; This represents a global knowledge model, which includes thermal compensation knowledge integrated from multiple devices; This represents a hierarchical structure recognition function, used to identify hierarchical relationships and organizational structures in a model; This represents the key node identification function, used to locate core nodes in the model that have a significant impact on decision-making. This represents a path tracing function used to analyze the critical decision paths from input to output in a model. It represents the complete result set of the model structure analysis, including hierarchical structure, key nodes, and decision path information.

[0222] Decision rule extraction steps: Extract the decision rules and relationships implicit in the model using the rule extraction function.

[0223] ;

[0224] in, This represents the decision rule extraction function, used to extract understandable decision rules from the model; This represents the results of the model structure analysis, including information on the model's hierarchical structure, key nodes, and decision paths. This represents a neural network to rule set transformation function, which converts a complex neural network structure into a series of logical rules. The function represents the rule optimization and simplification function, which simplifies and optimizes the extracted original rules to improve their understandability; This refers to the process of converting the model structure analysis results into an initial rule set; This represents the final decision rule set obtained after optimizing and simplifying the initial rule set.

[0225] Importance assessment steps: Evaluate the importance weight of each feature and rule in the decision-making process using an importance function.

[0226] ;

[0227] in, This represents the importance evaluation function, used to calculate the importance weights of features and rules; This represents the extracted rule set, which contains all decision rules extracted from the model; This represents the feature set, which contains all the input features used by the model. This represents a correlation analysis function for variables, used to analyze the strength of the correlation between rules and features; This represents the weighting function, which assigns importance weights to each feature and rule based on the correlation analysis results. This represents the results of a correlation analysis between the rule set and the feature set. This represents the final importance weight calculated based on the correlation analysis results. Before feature importance assessment, all numerical features are standardized and scaled to the same order of magnitude, and categorical features are one-hot encoded to ensure the fairness of the importance assessment.

[0228] Visualization transformation steps: Convert complex models into human-understandable visual representations using visualization functions:

[0229] ;

[0230] in, This represents a visualization transformation function used to convert complex models into intuitive visual representations. It represents the model structure and rules, including structural information and decision rules extracted from the model; This indicates the type of target visualization, such as different visualization forms like decision tree diagrams, heatmaps, and network diagrams; The intermediate representation function converts the original model structure into an intermediate format suitable for visualization. This represents a visualization rendering function that converts intermediate representations into the final visual graphics. This represents the result of the process of converting the model structure and rules into an intermediate representation. This represents the final visualization representation generated based on the intermediate representation and the target visualization type.

[0231] Semantic explanation generation steps: Generate text explanations that conform to the understanding habits of domain experts, using semantic generation functions:

[0232] ;

[0233] in, This represents a semantic interpretation generation function, used to generate text interpretations that conform to the understanding habits of domain experts; This represents the model's parsing information, including the structure, rules, and importance assessment results extracted from the model; It indicates the domain context, including the professional terminology, knowledge system, and expression habits of a specific manufacturing field; This represents a semantic mapping function that maps technical model parsing information to concepts and terms familiar to domain experts. This represents a natural language generation function that generates fluent and coherent text interpretations based on the mapped information. This represents the result of the process of semantically mapping the parsed information from the model based on the domain context; This represents the final text interpretation generated based on the semantic mapping result.

[0234] Furthermore, the explainable AI technology also includes a decision tree approximation step, which approximates the decision logic of a complex model into a more easily understood decision tree structure;

[0235] Furthermore, the explainable AI technology also includes a counterfactual explanation step, which enhances the understanding of model behavior by simulating the impact of input changes on the results.

[0236] Step 5.2, knowledge interpretation generation;

[0237] Based on the extracted decision logic, a human-understandable knowledge explanation document is generated, including the principles, basis, and expected effects of the heat compensation strategy.

[0238] Step 5.3, build the interpretable update package;

[0239] Package the model updates and corresponding explanation documents into an interpretable update package. .

[0240] Step 5.4, Personalized deployment guide generated;

[0241] Based on the specific device's status and characteristics, generate personalized deployment guidelines to instruct on how to apply the update package to the specific device.

[0242] Step 5.5, Deployment effect feedback mechanism;

[0243] Design a deployment effect feedback mechanism to collect effect data after the update package is deployed, which is used to evaluate the update quality and guide subsequent optimization.

[0244] Specific application examples of this implementation method:

[0245] Application scenarios:

[0246] This example applies to a large aerospace component manufacturing company that owns 23 high-precision CNC machine tools from three different manufacturers (A, B, and C). These machines are distributed across three different production workshops and are responsible for the precision machining of key components for aero-engines.

[0247] Since thermal deformation-induced processing errors are the main factor affecting product accuracy, and the thermal compensation models and data structures used by different manufacturers' equipment differ significantly, and each manufacturer has strict confidentiality requirements for compensation strategies and data, it is difficult to share thermal compensation experience among equipment.

[0248] Example of core steps implementation:

[0249] Build a digital twin mapping example for heterogeneous devices

[0250] First, the thermal deformation data structure of the equipment from the three manufacturers was analyzed, and a data structure feature vector was established.

[0251] Manufacturer A's equipment uses a thermal deformation prediction model based on a temperature matrix, employs 27 temperature sensors, and has a data sampling frequency of 10Hz.

[0252] Manufacturer B's equipment uses a regional temperature distribution model based on thermal imagers, and is equipped with two infrared thermal imagers with a data sampling frequency of 2Hz;

[0253] Manufacturer C's equipment uses a hybrid sensor network, which includes 15 contact temperature sensors and 4 displacement sensors, with a data sampling frequency of 5Hz.

[0254] After processing with semantic alignment algorithms, a unified knowledge representation format is constructed. The three different thermal deformation data structures were uniformly converted into a representation based on a standardized temperature field and deformation vector field, and a bidirectional mapping function set was constructed. This enables the conversion between different device representations and a unified representation. Table 1 shows examples of feature vectors for thermal deformation data structures from different manufacturers.

[0255] Table 1: Feature vectors of thermal deformation data structure for equipment from different manufacturers;

[0256]

[0257] Examples of intelligent agent networks and secure knowledge exchange:

[0258] A digital twin agent was deployed for each device, forming a network of 23 nodes. Based on the physical distribution and functional relevance of the devices, the network topology was constructed, as shown in Table 2. Devices within the same workshop have higher communication weights, and devices from the same manufacturer have higher model similarity.

[0259] Table 2: Communication Weight Matrix of Agent Network (Partial);

[0260]

[0261] Implement a knowledge exchange protocol based on interpretable homomorphic encryption to encrypt and share parameters of the heat compensation model and establish a secure communication channel. To enable encrypted knowledge flow between devices Exchange. Table 3 shows an example of a knowledge exchange operation log completed under encrypted conditions.

[0262] Table 3: Example of encrypted knowledge exchange operation log;

[0263]

[0264] Example of a hierarchical federated learning framework:

[0265] We design a hierarchical federated learning architecture that includes a global sharing layer and a local adaptation layer, and apply a heterogeneous model aggregation algorithm to construct a global knowledge model. And develop personalized adaptation layers for each device type. The global shared layer is responsible for learning general thermal compensation knowledge, including the general laws governing the thermal deformation of the guide rails due to changes in ambient temperature; the local adaptation layer makes personalized adjustments based on the specific characteristics of the equipment, such as considering differences in equipment structure and workload characteristics. Table 4 shows the parameter contribution weights of different equipment during the federated learning process.

[0266] Table 4: Example of weight allocation in federated learning models;

[0267]

[0268] Technical effectiveness verification:

[0269] In practical applications, this method also implements distributed anomaly pattern recognition and processing based on a cryptographic multi-party consensus algorithm, generating a global anomaly assessment. and collaborative processing solutions Furthermore, it applies explainable AI technology to analyze the global knowledge model, generating an explainable update package that includes model updates and explanatory documents. Execute personalized deployments.

[0270] Improved accuracy of thermal compensation:

[0271] After applying this method, the thermal compensation accuracy of equipment from different manufacturers was significantly improved. Table 5 shows a comparison of the thermal deformation compensation accuracy of equipment from various manufacturers before and after the application under standard test conditions.

[0272] Table 5: Comparison of thermal compensation accuracy (μm);

[0273]

[0274] Data privacy protection and knowledge sharing effects:

[0275] This method achieves effective knowledge sharing while protecting data privacy. Table 6 shows a comparison of the system's privacy protection performance and knowledge sharing efficiency.

[0276] Table 6: Efficiency Indicators of Privacy Protection and Knowledge Sharing;

[0277]

[0278] As can be seen from the above examples, this method not only significantly improves the accuracy of thermal compensation between heterogeneous devices, but also achieves efficient knowledge sharing and collaborative optimization while ensuring data privacy and security.

[0279] 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 guide rail thermal compensation optimization method based on digital twin, characterized in that, Includes the following steps: Constructing a digital twin mapping of heterogeneous devices to generate a unified knowledge representation format and a bidirectional mapping function set includes: analyzing the thermal deformation data structure of each type of device, establishing a data structure feature vector, which needs to be normalized before construction, scaling features of different dimensions to the [0, 1] interval, and performing one-hot encoding conversion on the categorical features; and constructing a bidirectional mapping function set between the unified representation and the original representation for each type of device. Based on a unified knowledge representation and mapping function, an agent network is deployed and an interpretable homomorphic encrypted knowledge exchange protocol is implemented to establish a secure communication channel and encrypted knowledge flow. This includes: deploying a corresponding digital twin agent for each entity device; the digital twin agent includes a local data management module, a knowledge conversion module, a secure communication module, and a decision execution module. The local data management module is responsible for the collection, cleaning, and storage of local thermal deformation data; the knowledge conversion module applies a mapping function to perform knowledge representation conversion; the secure communication module is responsible for encrypted communication and knowledge exchange; and the decision execution module executes a thermal compensation strategy and provides feedback on the results. The interpretable homomorphic encrypted knowledge exchange protocol includes a security level adaptive adjustment step, dynamically adjusting the encryption strength according to data sensitivity and computational requirements to balance security and computational efficiency. By utilizing the established encrypted knowledge flow, a hierarchical federated learning framework comprising a global sharing layer and a local adaptation layer is designed. A heterogeneous model aggregation algorithm is applied to construct a global knowledge model and a personalized adaptation layer. The heterogeneous model aggregation algorithm includes model structure analysis, knowledge distillation, spatial transformation, weight calculation, weighted aggregation, and conflict detection and resolution steps. Based on the local observations and global knowledge models of each intelligent agent, an encrypted multi-party consensus algorithm is applied to identify and process distributed anomaly patterns, and generate a global anomaly assessment and collaborative processing scheme. The encrypted multi-party consensus algorithm includes a local anomaly description step, an encrypted sharing step, an encrypted voting step, a threshold determination step, a consensus formation step, and a weight dynamic adjustment step. Based on the global knowledge model and local device status, explainable AI technology is applied to generate explainable knowledge update packages and perform personalized deployments.

2. The guide rail thermal compensation optimization method based on digital twin according to claim 1, characterized in that, The steps of designing a hierarchical federated learning framework, which includes a global sharing layer and a local adaptation layer, are as follows: Design a hierarchical federated learning architecture that includes a global sharing layer and a local adaptation layer; Design a heterogeneous model aggregation algorithm capable of handling knowledge from different structural models; A global knowledge model is constructed based on encrypted knowledge flow and heterogeneous model aggregation algorithms. Develop personalized adaptation layers for each device type; An incremental learning mechanism is designed to enable the system to continuously learn from new thermal compensation experiences.

3. The guide rail thermal compensation optimization method based on digital twin according to claim 1, characterized in that, The steps for using the encrypted multi-party consensus algorithm to identify and handle distributed anomaly patterns include: Deploy an anomaly detection model on each agent; Design a multi-party consensus algorithm based on homomorphic encryption; Based on the consensus results from multiple parties, a global anomaly assessment report is generated; Based on the global anomaly assessment, a collaborative handling plan is formulated; The identified anomaly patterns and their handling experience are stored in the anomaly knowledge base.

4. The guide rail thermal compensation optimization method based on digital twin according to claim 1, characterized in that, The steps for generating interpretable knowledge update packages and performing personalized deployment using the applied interpretable AI technology include: The application of interpretable AI technology is used to analyze the global knowledge model; Based on the extracted decision logic, generate knowledge explanation documents that are understandable to humans; Package the model update and the corresponding explanation document into an interpretable update package; Generate personalized deployment guidelines based on the status and characteristics of specific devices; Design and deploy a feedback mechanism to collect data on the effects of the update package deployment.

5. The guide rail thermal compensation optimization method based on digital twin according to claim 2, characterized in that, The hierarchical federated learning architecture includes: The global shared layer module is responsible for learning and sharing general thermal compensation knowledge; The local adaptation layer module is responsible for personalized adjustments based on the characteristics of specific devices; The inter-layer communication module is responsible for information exchange between the global layer and the local layer. The security isolation module ensures that proprietary device data is not directly exposed.

6. A guide rail thermal compensation optimization system based on digital twin, used to execute the guide rail thermal compensation optimization method based on digital twin as described in any one of claims 1 to 5, characterized in that, include: The heterogeneous device digital twin mapping module is used to generate a unified knowledge representation format and a bidirectional mapping function set; The agent network deployment module is used to deploy digital twin agents and realize knowledge exchange based on interpretable homomorphic encryption; The hierarchical federated learning framework module is used to handle the integration of heterogeneous knowledge and the construction of a global knowledge model; The distributed anomaly handling module is used to identify and handle anomaly patterns based on a cryptographic multi-party consensus algorithm. The Explainable Knowledge Update module is used to generate explainable knowledge update packages and perform customized deployments.

Images