Intelligent Maintenance Decision-making Method and System for CNC Machine Tools Based on Digital Twin

Abstract

The present invention provides an intelligent maintenance decision-making method and system for numerically controlled machine tools based on digital twins, which relates to the technical field of equipment maintenance, and includes: obtaining the structural data and simulation accuracy requirements of numerically controlled machine tools, determining the optimal mesh size and distribution parameters of the finite element model, obtaining operating conditions and adding them to the multi-physical field coupling modeling module, generating a machine tool digital twin model and fusing it with the physical field, generating a machine tool operation degradation model, and optimizing it to obtain a self-evolving digital twin model; obtaining machine tool operation parameters, performing preprocessing to obtain multi-modal perception data, extracting multi-perspective latent features and determining latent correlation features to obtain the machine tool health state evaluation result; generating multi-level fault prediction results, determining candidate maintenance plans, adding the candidate maintenance plans and multi-level fault prediction results to the knowledge graph construction module, generating a semantic association network and combining it with a multi-objective optimization algorithm to solve and obtain the optimal maintenance decision.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment maintenance, and in particular, to an intelligent maintenance decision-making method and system for numerically controlled machine tools based on digital twins. Background Art

[0002] With the continuous development of the manufacturing industry, numerically controlled machine tools have become one of the key equipment in modern manufacturing. The reliable operation of numerically controlled machine tools directly affects production efficiency and product quality. However, due to factors such as the complex structure and harsh working environment of numerically controlled machine tools, various faults and performance degradation problems will inevitably occur during their operation;

[0003] Traditional maintenance of numerically controlled machine tools mainly relies on regular maintenance and after-failure repair. This passive maintenance mode is difficult to meet the needs of modern manufacturing. It is difficult to determine the regular maintenance cycle. Being too frequent will increase the maintenance cost, while being too sparse will not be able to detect potential faults in time. After-failure repair will lead to unexpected downtime, affecting production progress and product delivery. In addition, traditional maintenance methods rely heavily on the experience and skills of maintenance personnel, resulting in high labor costs and strong decision-making subjectivity;

[0004] With the development of technologies such as the Internet of Things, big data, and artificial intelligence, intelligent maintenance decision-making methods have received extensive attention. However, most methods only focus on a single data source, making it difficult to comprehensively reflect the health status of machine tools. At the same time, they mainly rely on data-driven modeling and lack consideration of the physical mechanism of machine tools. The interpretability and reliability of prediction results need to be improved;

[0005] Therefore, there is an urgent need for a solution to solve the problems existing in the prior art. Summary of the Invention

[0006] The embodiments of the present invention provide an intelligent maintenance decision-making method and system for numerically controlled machine tools based on digital twins, which can at least solve some problems existing in the prior art.

[0007] In the first aspect of the embodiments of the present invention, an intelligent maintenance decision-making method for numerically controlled machine tools based on digital twins is provided, including:

[0008] Obtain the structural data and simulation accuracy requirements of the numerically controlled machine tool, determine the optimal mesh size and distribution parameters of the finite element model in combination with the adaptive mesh division algorithm, obtain the operating conditions of the numerically controlled machine tool in the finite element model and add them to the multi-physical field coupling modeling module, generate a machine tool digital twin model and fuse it with the physical field, generate a machine tool operation degradation model in combination with the mapping relationship, and continuously optimize it through the pre-obtained machine tool operation big data and the deep transfer learning model to obtain a self-evolving digital twin model;

[0009] Construct a sensing network based on the self-evolving digital twin model, comprehensively sense the CNC machine tool, obtain the operating parameters of the machine tool through distributed adaptive sampling, preprocess the operating parameters of the machine tool, obtain multi-modal sensing data and add it to the multi-view learning fusion model, extract multi-perspective latent features and determine the latent correlation features between different modalities, and combine with the self-evolving digital twin model to obtain the machine tool health status evaluation result;

[0010] Based on the machine tool health status evaluation result, combine with the pre-set multi-time scale fault prediction model to generate multi-level fault prediction results, combine with historical maintenance cases, adaptively determine the candidate maintenance plan through the maintenance strategy optimization algorithm based on deep reinforcement learning, add the candidate maintenance plan and the multi-level fault prediction results to the knowledge graph construction module, generate a semantic association network by associating the machine tool fault causes and failure mechanisms, and combine with the multi-objective optimization algorithm to solve and obtain the optimal maintenance decision.

[0011] In an alternative embodiment,

[0012] Obtain the structural data and simulation accuracy requirements of the CNC machine tool, combine with the adaptive mesh generation algorithm to determine the optimal mesh size and distribution parameters of the finite element model, obtain the operating conditions of the CNC machine tool in the finite element model and add it to the multi-physics field coupling modeling module, generate a machine tool digital twin model and fuse it with the physical field, generate a machine tool operation degradation model in combination with the mapping relationship, and continuously optimize it through the pre-obtained CNC machine tool operation big data and the deep transfer learning model to obtain the self-evolving digital twin model, including:

[0013] Obtain the structural parameters of the CNC machine tool through 3D scanning, where the structural parameters include the dimensions, shapes, and assembly relationships of the CNC machine tool components, and obtain the corresponding simulation accuracy requirements of the CNC machine tool based on the service conditions and fault modes of the CNC machine tool;

[0014] Conduct feature analysis on the CNC machine tool, extract structural stress and hot spots as key features, determine the feature attributes, and determine the mesh scales of different regions in combination with the simulation accuracy requirements. Take the regions with stress concentration and cross-section mutation as weak regions, increase the mesh density in the weak regions, combine with high-order mesh elements, improve the approximation accuracy of the field variables within the elements, complete the mesh generation, and obtain the optimal mesh size and distribution parameters. Build a multi-granularity finite element model of the current CNC machine tool through finite element software;

[0015] Based on the multi-granularity finite element model, obtain the boundary conditions of the CNC machine tool under machining conditions, add the boundary conditions as loads and constraints to the multi-granularity finite element model, combine the physical field corresponding to the CNC machine tool, and perform simulation numerical solution using the physical field coupling solution method to obtain the corresponding physical field behavior. Combine the sensor mapping technology, collect the state data of the CNC machine tool and map it to the corresponding nodes of the virtual model to obtain a multi-physical field digital twin model. Obtain the degradation process of the CNC machine tool, and combine the pre-acquired large data of machine tool operation to extract degradation precursor features to construct a machine tool degradation model;

[0016] Through the deep transfer learning technology, combine the large data of machine tool operation, extract the common features under different working conditions through a deep neural network, determine the degradation law, and combine transfer learning to adaptively adjust the machine tool degradation model under new working conditions to obtain the self-evolving digital twin model.

[0017] In an alternative embodiment,

[0018] The adaptive adjustment of the machine tool degradation model under new working conditions by combining transfer learning is shown in the following formula:

[0019] ;

[0020] where θ t represents the network parameters under the target working condition, L t (θ) represents the loss function under the target working condition, dist(θ s , θ) represents the distance metric between the network parameters of the source working condition and the target working condition, λ1 represents the distance metric weight factor, λ2 represents the regularization term weight factor, and R(θ) represents the regularization term.

[0021] In an alternative embodiment,

[0022] Based on the self-evolving digital twin model, construct a sensing network to comprehensively perceive the CNC machine tool and obtain the machine tool operation parameters through distributed adaptive sampling. Preprocess the machine tool operation parameters to obtain multi-modal perception data and add it to the multi-view learning fusion model. Extract multi-perspective latent features and determine the latent association features between different modalities. Combine the self-evolving digital twin model to obtain the machine tool health state evaluation results, including:

[0023] Based on the self-evolving digital twin model, construct a sensing network covering all components and functional units of the CNC machine tool. Among them, the sensing network includes multi-type sensors corresponding to vibration, temperature, stress, displacement, and current;

[0024] Based on the sensing network, comprehensively perceive the CNC machine tool. Combine the adaptive sampling trigger preset at the edge node, and dynamically adjust the sampling frequency of the sensing network according to the real-time working conditions and degradation status of the CNC machine tool to obtain the operating parameters of the machine tool, and preprocess the operating parameters of the machine tool to obtain the multi-modal perception data;

[0025] Based on the multi-modal perception data, for each modality, set the corresponding feature extraction sub-view. Based on the feature extraction sub-view, extract the trend feature and periodic feature from the data within each modality as the multi-perspective latent features. Based on the multi-perspective latent features, combine the multi-view fusion algorithm, and generate the cross-modal common representation and determine the latent association features by determining the linear correlation and non-linear correlation between different modalities;

[0026] Add the latent association features to the self-evolving digital twin model, and combine the real-time monitoring data to generate the evaluation result of the machine tool health status.

[0027] In an alternative embodiment,

[0028] Based on the feature extraction sub-view, extract the trend feature and periodic feature from the data within each modality as shown in the following formula:

[0029] ;

[0030] where WT(a, b) represents the periodic feature at scale a and position b, a represents the scale parameter for controlling the function width, b represents the translation parameter for determining the position of the function on the time axis, x(t) represents the original signal, and ψ() represents the mother wavelet function;

[0031] ;

[0032] where y(t) represents the trend feature of the data within the current modality, w j represents the weight coefficient of the j-th scale, M j represents the sliding window length of the j-th scale, J represents the total number of scales, and i is an index variable used to represent the time step back from the current time point t.

[0033] In an alternative embodiment,

[0034] Based on the evaluation results of the machine tool health status, combined with a pre-set multi-time scale fault prediction model to generate multi-level fault prediction results, combined with historical maintenance cases, adaptively determine candidate maintenance plans through a maintenance strategy optimization algorithm based on deep reinforcement learning, add the candidate maintenance plans and the multi-level fault prediction results to the knowledge graph construction module, and generate a semantic association network by associating the machine tool fault causes and failure mechanisms, and combine with a multi-objective optimization algorithm to solve and obtain the optimal maintenance decision, including:

[0035] Obtain the evaluation results of the machine tool health status and add them to a pre-set multi-time scale fault prediction model. Generate short-term prediction results through a short-term prediction model based on a gated recurrent unit, determine medium-term prediction results through a medium-term prediction model based on a convolutional long short-term memory network, generate long-term prediction results through a long-term prediction model based on a deep survival model, and combine the predicted results obtained by the combined prediction to obtain the multi-level fault prediction results;

[0036] Based on the multi-level fault prediction results, set a global decision-making layer and a sub-task decision-making layer in combination with the working conditions. Among them, the global decision-making layer takes the health status of the CNC machine tool as the state, the set of candidate maintenance plans as the action, and the maintenance performance index as the reward. The sub-task decision-making layer takes the sub-task state as the state, the sub-task execution plan as the action, and the sub-task performance index as the reward;

[0037] Based on the global decision-making layer and the sub-task decision-making layer, determine the global optimal decision and the local optimal decision in combination with an active exploration mechanism, cooperate and optimize the global optimal decision in combination with the Nash equilibrium strategy to generate a set of global maintenance plans, perform distributed optimization on the sub-task decision-making layer through a multi-agent reinforcement learning algorithm to generate sub-task execution plans, integrate the sub-task execution plans and the set of global maintenance plans to obtain candidate maintenance plans, and repeat the integration to obtain a set of candidate maintenance plans;

[0038] Design an ontology in the field of CNC machine tools, define core concepts and relationship types, extract entities and the corresponding attribute information of the entities from structured data and unstructured data, perform knowledge fusion through methods based on rules and graph embedding to generate a CNC machine tool knowledge graph, add the candidate maintenance plans and the multi-level fault prediction results to the CNC machine tool knowledge graph, and determine the fault causes and failure mechanisms through semantic association analysis, path reasoning and graph embedding reasoning algorithms to generate a semantic association network and the fault propagation path corresponding to the current fault;

[0039] With the goal of minimizing the failure recurrence rate and maintenance cost, the objective function is set by combining a multi-objective optimization algorithm, and multiple elements are randomly selected from the set of candidate maintenance plans to form an initial population. The initial population is subjected to crossover operations and mutation operations to generate an offspring population and perform non-dominated sorting. The crowding degree is calculated and the individuals to be retained are determined in combination with the dominance relationship. The operations are repeated to generate the Pareto front and sorted according to the Pareto solution set, and the individual ranked first is selected as the optimal maintenance decision.

[0040] In an alternative embodiment,

[0041] The loss function corresponding to the long-term prediction model based on the deep survival model is shown in the following formula:

[0042] ;

[0043] where L represents the loss value, δ g represents the indicator function, indicating whether the g-th sample is the time point of the event occurrence, f(x, T) represents the prediction result of the model for the input feature x and time t, T g represents the event occurrence event of the g-th sample, μ represents the regularization coefficient, and W represents the network parameters.

[0044] In the second aspect of the embodiments of the present invention, a digital twin-based intelligent maintenance decision-making system for numerically controlled machine tools is provided, including:

[0045] A first unit for obtaining the structural data and simulation accuracy requirements of the numerically controlled machine tool, determining the optimal mesh size and distribution parameters of the finite element model by combining the adaptive mesh generation algorithm, obtaining the operating conditions of the numerically controlled machine tool in the finite element model and adding them to the multi-physics field coupling modeling module, generating a machine tool digital twin model and fusing it with the physical field, generating a machine tool operation degradation model in combination with the mapping relationship, and continuously optimizing it through the pre-obtained machine tool operation big data and the deep transfer learning model to obtain a self-evolving digital twin model;

[0046] A second unit for constructing a sensing network based on the self-evolving digital twin model, comprehensively perceiving the numerically controlled machine tool and obtaining the machine tool operation parameters through distributed adaptive sampling, preprocessing the machine tool operation parameters to obtain multi-modal perception data and adding them to the multi-view learning fusion model, extracting multi-perspective latent features and determining the latent association features between different modalities, and obtaining the machine tool health status evaluation result in combination with the self-evolving digital twin model;

[0047] A third unit, configured to generate multi-level fault prediction results based on the evaluation results of the machine tool health state, combine with a pre-set multi-time scale fault prediction model, combine historical maintenance cases, adaptively determine candidate maintenance plans through a maintenance strategy optimization algorithm based on deep reinforcement learning, add the candidate maintenance plans and the multi-level fault prediction results to a knowledge graph construction module, generate a semantic association network by associating machine tool fault causes and failure mechanisms, and combine with a multi-objective optimization algorithm to solve for an optimal maintenance decision.

[0048] In a third aspect of the embodiments of the present invention,

[0049] a kind of electronic device is provided, including:

[0050] a processor;

[0051] a memory for storing instructions executable by the processor;

[0052] wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0053] In a fourth aspect of the embodiments of the present invention,

[0054] a computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0055] In the present invention, the finite element model is optimized by an adaptive mesh division algorithm, and combined with multi-physics field coupling modeling to generate a machine tool digital twin model. By combining the machine tool operation big data with a deep transfer learning model, the continuous optimization and self-evolution of the digital twin model are realized, which can accurately reflect the real-time operation state and degradation law of the machine tool, and provide a reliable digital basis for subsequent health assessment and fault prediction. A multi-view learning fusion model is used to extract features and perform correlation analysis on multi-modal perception data to realize information fusion between different modalities, which can comprehensively capture multi-dimensional features of the machine tool operation state and improve the accuracy and robustness of health state assessment. A multi-time scale fault prediction model is introduced to predict machine tool faults from three time scales of short-term, medium-term and long-term, which can capture the dynamic characteristics of fault evolution, provide a comprehensive estimate of the fault occurrence time and severity, and provide a basis for formulating targeted maintenance strategies. A maintenance strategy optimization algorithm based on deep reinforcement learning is used to adaptively determine candidate maintenance plans according to the machine tool health state and fault prediction results, so that the maintenance decision can be continuously improved over time, and the pertinence and effectiveness of maintenance are improved. In summary, the present invention significantly improves the operation and maintenance efficiency and reliability of CNC machine tools, realizes the intelligence of machine tool health state assessment, fault prediction and maintenance decision-making, and provides an innovative solution for realizing the intelligent operation and maintenance of CNC machine tools. Brief Description of the Drawings

[0056] Figure 1 It is a schematic flow chart of the intelligent maintenance decision-making method for a numerically controlled machine tool based on digital twin in an embodiment of the present invention;

[0057] Figure 2 It is a schematic structural diagram of the intelligent maintenance decision-making system for a numerically controlled machine tool based on digital twin in an embodiment of the present invention. Detailed Embodiments

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are only a part rather than all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0059] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments may be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0060] Figure 1 It is a schematic flow chart of the intelligent maintenance decision-making method for a numerically controlled machine tool based on digital twin in an embodiment of the present invention. As Figure 1 shown, the method includes:

[0061] S1. Obtain the structural data and simulation accuracy requirements of the numerically controlled machine tool, determine the optimal mesh size and distribution parameters of the finite element model in combination with the adaptive mesh generation algorithm, obtain the operating conditions of the numerically controlled machine tool in the finite element model and add them to the multi-physics field coupling modeling module, generate a machine tool digital twin model and fuse it with the physical field, generate a machine tool operation degradation model in combination with the mapping relationship, and continuously optimize it through the pre-obtained machine tool operation big data and the deep transfer learning model to obtain a self-evolving digital twin model;

[0062] The simulation accuracy requirement refers to the requirements for the accuracy and precision of the results in the process of computer simulation and numerical simulation. The adaptive mesh generation algorithm is used to automatically adjust the mesh density according to the error of the solution or the change of the gradient during the simulation process. The finite element model is a numerical calculation method that solves complex physical problems by dividing the continuous medium into discrete finite elements. The multi-physics coupling modeling module is used to simulate the interaction between multiple physical fields (such as heat, structure, fluid, electromagnetic, etc.). The deep transfer learning model refers to using the knowledge of a deep learning model trained on a source task to help solve problems on a target task. The self-evolving digital twin model is a digital model that can continuously update and optimize itself as the physical system changes and data accumulates.

[0063] In an optional implementation manner,

[0064] Obtain the structural data and simulation accuracy requirements of the CNC machine tool, determine the optimal mesh size and distribution parameters of the finite element model in combination with the adaptive mesh generation algorithm, obtain the operating conditions of the CNC machine tool in the finite element model and add them to the multi-physics coupling modeling module, generate the digital twin model of the machine tool and fuse it with the physical field, generate the machine tool operation degradation model in combination with the mapping relationship, and continuously optimize it through the pre-obtained large data of the machine tool operation and the deep transfer learning model to obtain the self-evolving digital twin model, including:

[0065] Obtain the structural parameters of the CNC machine tool through 3D scanning. Among them, the structural parameters include the dimensions, shapes, and assembly relationships of the components of the CNC machine tool. Based on the service conditions and failure modes of the CNC machine tool, obtain the corresponding simulation accuracy requirements of the CNC machine tool;

[0066] Conduct feature analysis on the CNC machine tool, extract structural stress and hot spots as key features, determine the feature attributes, and determine the mesh scales of different regions in combination with the simulation accuracy requirements. Take the regions with stress concentration and sudden cross-section change as weak regions, increase the mesh density in the weak regions, combine high-order mesh elements, improve the approximation accuracy of the field variables within the elements, complete the mesh generation, and obtain the optimal mesh size and distribution parameters. Build a multi-granularity finite element model of the current CNC machine tool through finite element software;

[0067] Based on the multi-granularity finite element model, obtain the boundary conditions of the CNC machine tool under processing conditions, add the boundary conditions as loads and constraints to the multi-granularity finite element model, combine with the physical field corresponding to the CNC machine tool, and perform simulation numerical solution using the physical field coupling solution method to obtain the corresponding physical field behavior. Combine with the sensor mapping technology, collect the state data of the CNC machine tool and map it to the corresponding nodes of the virtual model to obtain the multi-physical field digital twin model. Obtain the degradation process of the CNC machine tool, and combine with the pre-acquired large data of machine tool operation to extract degradation precursor features, and construct a machine tool degradation model;

[0068] Through the deep transfer learning technology, combine with the large data of machine tool operation, extract the common features under different working conditions through a deep neural network, determine the degradation law, and adaptively adjust the machine tool degradation model under the new working conditions through transfer learning to obtain the self-evolving digital twin model.

[0069] The fault mode refers to various fault types and manifestation forms that a system or equipment may have. The structural stress refers to the internal stress generated in a structural member under the action of an external load. The approximation accuracy refers to the degree of fitting of the model to the actual system or data during numerical simulation and modeling. The degradation precursor feature refers to the early features or signals shown by a system or equipment during the degradation process. The degradation law refers to the regular change of the performance of a system or equipment gradually decreasing over time during use.

[0070] Obtain the structural parameters of the CNC machine tool through 3D scanning technology, including the dimensions, shapes, and assembly relationships of the CNC machine tool components. Based on the service conditions and fault modes of the CNC machine tool, determine the corresponding simulation accuracy requirements;

[0071] Conduct feature analysis on the CNC machine tool, extract structural stress and hot spots as key features, determine the feature attributes, combine with the simulation accuracy requirements, determine the mesh scales of different regions, regard the regions with stress concentration and sudden cross-section changes as weak regions, increase the mesh density in the weak regions, combine with high-order mesh elements, improve the approximation accuracy of the field variables within the elements, complete the mesh division, obtain the optimal mesh size and distribution parameters, and build a multi-granularity finite element model of the current CNC machine tool through finite element software;

[0072] Add the boundary conditions of the CNC machine tool under actual machining conditions, such as load and constraint conditions, to the multi-granularity finite element model. Considering the multi-physical field coupling of the machine tool, such as the mutual influence between structural stress and thermal field, use the physical field coupling solution method to numerically simulate and solve the multi-granularity finite element model, and obtain the physical field behavior of the CNC machine tool under actual working conditions, such as stress distribution, deformation conditions, and temperature distribution, etc. Combine the sensor mapping technology to map the real-time collected state data of the CNC machine tool to the corresponding nodes of the virtual model, construct a multi-physical field digital twin model, and realize the real-time monitoring and prediction of the working state of the machine tool;

[0073] Use the pre-acquired large data of machine tool operation to extract the degradation process and degradation precursor characteristics of the machine tool, including changes in performance parameters, abnormal vibration spectra, etc., for constructing a machine tool degradation model. Based on the deep transfer learning technology, combined with the large data of machine tool operation, extract the common characteristics under different working conditions through a deep neural network, determine the degradation law, and use the transfer learning method to adaptively adjust the machine tool degradation model under new working conditions to generate a self-evolving digital twin model.

[0074] In this embodiment, accurately obtain the geometric structure and assembly relationship of the CNC machine tool through 3D scanning technology to ensure the authenticity and accuracy of the model, improve the credibility and application value of the simulation results. Through the feature extraction algorithm and structural analysis tool, accurately capture the structural stress and hot spots of the machine tool, determine the weak areas, and improve the approximation accuracy of stress field variables by increasing the mesh density and using high-order mesh elements, which can effectively predict possible structural problems. Based on the deep neural network and transfer learning method, extract the common characteristics under different working conditions, determine the degradation law of the CNC machine tool, and realize the adaptive adjustment of the machine tool degradation model, so that the digital twin model can be continuously optimized and updated, improve the prediction accuracy and fault diagnosis effect. In summary, this embodiment can significantly improve the operation safety, stability and efficiency of the CNC machine tool, effectively extend the service life of the equipment, reduce the maintenance cost, and thus achieve the goals of intelligent manufacturing and service-oriented operation.

[0075] In an alternative embodiment,

[0076] The adaptive adjustment of the machine tool degradation model under new working conditions in combination with transfer learning is shown in the following formula:

[0077] ;

[0078] where θ t represents the network parameters under the target working condition, and L t (θ) represents the loss function under the target working condition, and dist(θ s, θ) represents the distance metric between the network parameters of the source working condition and the target working condition, λ1 represents the distance metric weight factor, λ2 represents the regularization term weight factor, and R(θ) represents the regularization term.

[0079] In this embodiment, by introducing the model parameters trained under the source working condition to measure the difference between the network parameters of the source working condition and the target working condition, and controlling the intensity of the source working condition knowledge transfer through the weight factor, it helps to improve the generalization ability of the model under the target working condition. Especially when the sample data of the target working condition is scarce, by introducing the regularization term, the network parameters under the target working condition are constrained and smoothed to avoid overfitting of the model. By controlling the distance metric weight factor and the regularization term weight factor, the proportion of the source working condition knowledge transfer and the learning of the target working condition's own characteristics can be balanced, which helps to improve the robustness of the model, enabling it to better adapt to the data distribution and noise perturbation under the target working condition. In summary, this embodiment makes full use of the prior knowledge under the source working condition, improves the generalization ability, convergence speed and robustness of the model, realizes the adaptive optimization of the model under the new working condition, and provides strong support for the health management and predictive maintenance of the CNC machine tool.

[0080] S2. Based on the self-evolving digital twin model, construct a sensing network, comprehensively perceive the CNC machine tool, obtain the machine tool operation parameters through distributed adaptive sampling, preprocess the machine tool operation parameters, obtain multi-modal perception data and add it to the multi-view learning fusion model, extract multi-perspective latent features and determine the latent correlation features between different modalities, and combine with the self-evolving digital twin model to obtain the machine tool health status evaluation result;

[0081] The distributed adaptive sampling is a method to cope with the data acquisition requirements in a distributed system. The adaptive sampling strategy can dynamically adjust the sampling frequency and sampling range of the sensors according to the real-time machine tool operation state and environmental conditions to ensure that the collected data is representative and of high quality, while reducing the cost of data transmission and processing. The machine tool operation parameters refer to various parameters that describe the key performance and state of the machine tool during the working process, including but not limited to: cutting speed, feed speed, cutting force, temperature, vibration, machining accuracy, etc. The multi-view learning fusion model combines the data features from different perspectives or modalities to analyze the machine tool operation state and health condition more comprehensively and accurately. The multi-perspective latent features refer to the representative and highly correlated data features extracted from multiple perspectives or modalities. The latent correlation features refer to the correlations and patterns that may exist in the data but are not easily directly observable. The machine tool health status evaluation result is a conclusion drawn based on the multi-view learning fusion model and multi-perspective latent feature analysis, reflecting the overall operation health status of the machine tool at a specific moment or cycle.

[0082] In an alternative embodiment,

[0083] Construct a sensing network based on the self-evolving digital twin model, comprehensively sense the CNC machine tool, obtain the machine tool operation parameters through distributed adaptive sampling, preprocess the machine tool operation parameters, obtain multi-modal sensing data and add it to the multi-view learning fusion model, extract multi-perspective latent features and determine the latent correlation features between different modalities, and combine with the self-evolving digital twin model to obtain the machine tool health status evaluation result, including:

[0084] Based on the self-evolving digital twin model, construct a sensing network covering all components and functional units of the CNC machine tool. Among them, the sensing network includes multi-type sensors corresponding to vibration, temperature, stress, displacement, and current.

[0085] Based on the sensing network, comprehensively sense the CNC machine tool. Combine the adaptive sampling trigger preset on the edge node, and dynamically adjust the sampling frequency of the sensing network according to the real-time working conditions and degradation status of the CNC machine tool to obtain the machine tool operation parameters, and preprocess the machine tool operation parameters to obtain the multi-modal sensing data.

[0086] Based on the multi-modal sensing data, for each modality, set a corresponding feature extraction sub-view. Based on the feature extraction sub-view, extract trend features and periodic features from the data within each modality as multi-perspective latent features. Based on the multi-perspective latent features, combine the multi-view fusion algorithm, and by determining the linear and non-linear correlations between different modalities, generate a cross-modal common representation and combine the multi-view alignment strategy to determine the latent correlation features.

[0087] Add the latent correlation features to the self-evolving digital twin model, and combine with the real-time monitoring data to generate the machine tool health status evaluation result.

[0088] The functional unit refers to the basic component that completes a specific function or task in a system or model. The adaptive sampling trigger is used to automatically trigger data collection according to the system state or specific conditions. The feature extraction sub-view is a module that extracts features of a specific aspect or modality from the data. The cross-modal common representation refers to mapping the information of different modalities or data types into a shared representation space to facilitate comparison and fusion between different data sources. The multi-view alignment strategy refers to the method or technology used to ensure the information consistency and relevance between different views in multi-view learning.

[0089] According to the structural and functional characteristics of the CNC machine tool, deploy multi-type sensors, including vibration sensors, temperature sensors, stress sensors, displacement sensors, current sensors, etc. The arrangement position and quantity of the sensors should fully consider the key components and weak links of the machine tool to ensure a comprehensive perception of the machine tool state.

[0090] An adaptive sampling trigger is set at the edge node. According to the real-time working conditions and degradation states of the CNC machine tools, the sampling frequency of the sensing network is dynamically adjusted. When the machine tool is in a stable working condition, the sampling frequency is reduced to reduce data redundancy. When the working conditions of the machine tool change or signs of degradation appear, the sampling frequency is increased to obtain more refined state data. The collected operating parameters of the machine tool are preprocessed to obtain multi-modal perception data;

[0091] For each data modality, a corresponding feature extraction sub-view is set. In each sub-view, the trend features and periodic features of the data are extracted as multi-view latent features. A multi-view fusion algorithm is adopted. By analyzing the linear and non-linear correlations between different modalities, a cross-modal common representation is generated. Combining with the multi-view alignment strategy, the potential associated features are determined to achieve the effective fusion of multi-modal data;

[0092] The obtained potential associated features are used as new inputs and added to the self-evolving digital twin model. Based on the updated model, combined with the real-time monitoring data collected by the sensing network, the current health state of the CNC machine tool is evaluated. The evaluation results can include the overall health of the machine tool, the health of key components, the prediction of degradation trends, etc., to obtain the evaluation result of the machine tool health state.

[0093] In this embodiment, by arranging multi-type sensors at the key components and weak links of the CNC machine tool, a sensing network covering all components and functional units is constructed, realizing the comprehensive perception of multiple state parameters such as the vibration, temperature, stress, displacement, and current of the machine tool, providing a rich information basis for subsequent data analysis and health assessment. The multi-modal data fusion method makes full use of the complementarity and correlation between different modal data, improves the richness and robustness of feature representation, and provides more comprehensive and accurate information for health state assessment. The digital twin model can dynamically adapt to the degradation process and working condition changes of the CNC machine tool through continuous learning and adaptive adjustment, improving the accuracy and real-time performance of health state assessment. In summary, this embodiment significantly improves the intelligent level of CNC machine tool health management and provides a strong support for realizing predictive maintenance and optimizing operation and maintenance strategies.

[0094] In an alternative embodiment,

[0095] Based on the feature extraction sub-view, the trend features and periodic features of the data within each modality are extracted as shown in the following formula:

[0096] ;

[0097] Among them, WT(a, b) represents the periodic feature at scale a and position b. a represents the scale parameter, which is used to control the function width, b represents the translation parameter, which is used to determine the position of the function on the time axis, x(t) represents the original signal, and ψ() represents the mother wavelet function;

[0098] ;

[0099] Among them, y(t) represents the trend feature of the data within the current mode, w j represents the weight coefficient of the j-th scale, M j represents the sliding window length of the j-th scale, J represents the total number of scales, and i is an index variable, which is used to represent the time step of backward tracing relative to the current time point t.

[0100] In this embodiment, wavelet transform is used to extract periodic features, which can effectively capture the periodic components of the signal at different scales. The moving average method can effectively smooth short-term fluctuations and highlight long-term trends by performing weighted averaging on the data within the sliding windows of different scales. By combining the periodic features and trend features at different scales, the fusion of multi-scale features is achieved. By selecting an appropriate mother wavelet function, high-frequency noise can be effectively suppressed and useful periodic components can be highlighted. In summary, this embodiment can effectively capture the periodic behavior and long-term trends during the operation of the CNC machine tool, and has good noise reduction and feature enhancement capabilities, providing high-quality and highly reliable feature inputs for subsequent health state assessment.

[0101] S3. Based on the machine tool health state assessment result, combine the pre-set multi-time scale fault prediction model to generate a multi-level fault prediction result. Combine historical maintenance cases, and adaptively determine the candidate maintenance plan through the maintenance strategy optimization algorithm based on deep reinforcement learning. Add the candidate maintenance plan and the multi-level fault prediction result to the knowledge graph construction module. By associating the machine tool fault causes and failure mechanisms, generate a semantic association network and combine with the multi-objective optimization algorithm to solve and obtain the optimal maintenance decision.

[0102] The multi-time scale fault prediction model considers the fault prediction of the system or equipment at different time scales. The multi-level fault prediction results refer to the multi-level prediction and analysis of different types of faults that may occur in the system or equipment. The historical maintenance cases refer to the cases and experiences of recording and analyzing the past maintenance and repair of the system or equipment. The maintenance strategy optimization algorithm is for the decision-making problem of equipment maintenance. Based on data analysis and model prediction, it determines the optimal maintenance time, method, and resource allocation through optimization methods. The knowledge graph construction module is a module that uses knowledge graph technology to model and represent the key information, attributes, relationships, and behaviors of the system or equipment in the form of a graph. The failure mechanism refers to the root cause or physical process that causes the failure of the equipment or system. The semantic association network is a network structure based on semantic relationships, used to describe and analyze the relevance between entities.

[0103] In an alternative embodiment,

[0104] Based on the machine tool health status evaluation result, combined with a pre-set multi-time scale fault prediction model to generate multi-level fault prediction results, combined with historical maintenance cases, adaptively determine candidate maintenance plans through a maintenance strategy optimization algorithm based on deep reinforcement learning, add the candidate maintenance plans and the multi-level fault prediction results to the knowledge graph construction module, and generate a semantic association network by associating the machine tool fault causes and failure mechanisms, and combine with a multi-objective optimization algorithm to solve and obtain the optimal maintenance decision, including:

[0105] Obtain the machine tool health status evaluation result and add it to a pre-set multi-time scale fault prediction model. Generate short-term prediction results through a short-term prediction model based on a gated recurrent unit, determine medium-term prediction results through a medium-term prediction model based on a convolutional long short-term memory network, generate long-term prediction results through a long-term prediction model based on a deep survival model, and combine the predicted results obtained by the combined prediction to obtain the multi-level fault prediction results;

[0106] Based on the multi-level fault prediction results, set a global decision layer and a sub-task decision layer in combination with the working conditions. Among them, the global decision layer takes the health status of the CNC machine tool as the state, the set of candidate maintenance plans as the action, and the maintenance performance index as the reward. The sub-task decision layer takes the sub-task status as the state, the sub-task execution plan as the action, and the sub-task performance index as the reward;

[0107] Based on the global decision-making layer and the sub-task decision-making layer, the global optimal decision and the local optimal decision are determined by combining the active exploration mechanism. The global optimal decision is collaboratively optimized by combining the Nash equilibrium strategy to generate a set of global maintenance plans. The sub-task decision-making layer is distributedly optimized by the multi-agent reinforcement learning algorithm to generate sub-task execution plans. The sub-task execution plans and the set of global maintenance plans are integrated to obtain candidate maintenance plans, and the integration is repeated to obtain a set of candidate maintenance plans;

[0108] Design an ontology in the field of CNC machine tools, define core concepts and relationship types, extract entities and the corresponding attribute information of the entities from structured data and unstructured data, perform knowledge fusion through rule-based and graph embedding methods to generate a CNC machine tool knowledge graph, add the candidate maintenance plans and the multi-level fault prediction results to the CNC machine tool knowledge graph, and determine the fault causes and failure mechanisms through semantic association analysis, path reasoning, and graph embedding reasoning algorithms to generate a semantic association network and the fault propagation path corresponding to the current fault;

[0109] With the goal of minimizing the fault recurrence rate and the maintenance cost, set the objective function by combining the multi-objective optimization algorithm and randomly select multiple elements from the set of candidate maintenance plans to form an initial population. Perform crossover operations and mutation operations on the initial population to generate an offspring population and perform non-dominated sorting. Calculate the crowding degree and determine the retained individuals by combining the dominance relationship. Repeat the operations to generate the Pareto front and sort according to the Pareto solution set, and select the individual ranked first as the optimal maintenance decision.

[0110] The global decision-making layer refers to the layer in the entire system or multi-agent system responsible for global task allocation and optimization decisions. The sub-task decision-making layer is the layer in the multi-agent system responsible for executing specific tasks and local decisions. The Nash equilibrium strategy is an important concept in game theory, which means that among multiple decision-makers, each decision-maker selects its best strategy according to the strategies of other decision-makers, so that no decision-maker can improve its own utility by unilaterally changing the strategy. The multi-agent reinforcement learning algorithm is a type of algorithm designed for multi-agent systems, which combines the idea of reinforcement learning to enable multiple agents to learn how to maximize the cumulative reward in the interaction with the environment. Path reasoning refers to inferring possible event sequences or behavior paths in a complex system based on historical data and models. The graph embedding reasoning algorithm is a method of mapping entities and relationships in a knowledge graph to a low-dimensional vector space for easy reasoning and analysis. The fault propagation path describes how a fault propagates and affects through various parts or components of the system after the fault occurs in the system. Non-dominated sorting is a method in multi-objective optimization problems used to compare and sort solutions between multiple objective functions.

[0111] Obtain the evaluation results of the machine tool health status and add them to a pre-set multi-time-scale fault prediction model. Generate short-term prediction results through a short-term prediction model based on gated recurrent units, determine medium-term prediction results through a medium-term prediction model based on convolutional long short-term memory networks, generate long-term prediction results through a long-term prediction model based on deep survival models, and combine the short-term, medium-term, and long-term prediction results to obtain multi-level fault prediction results;

[0112] Based on the multi-level fault prediction results, set up a global decision-making layer and a sub-task decision-making layer in combination with working conditions. The global decision-making layer takes the health status of the CNC machine tool as the state, the set of candidate maintenance plans as the action, and the maintenance performance index as the reward. The sub-task decision-making layer takes the sub-task status as the state, the sub-task execution plan as the action, and the sub-task performance index as the reward. Use the Nash equilibrium strategy to collaboratively optimize the global optimal decision, generate a set of global maintenance plans, perform distributed optimization on the sub-task decision-making layer through a multi-agent reinforcement learning algorithm, generate sub-task execution plans, and integrate the sub-task execution plans and the set of global maintenance plans to obtain candidate maintenance plans. Repeat the integration process to obtain a set of candidate maintenance plans;

[0113] Design an ontology for the CNC machine tool field, define core concepts and relationship types, extract entities and the corresponding attribute information of the entities from structured and unstructured data, perform knowledge fusion through rule-based and graph embedding methods to generate a knowledge graph of the CNC machine tool, add the candidate maintenance plans and multi-level fault prediction results to the knowledge graph of the CNC machine tool, and determine the fault causes and failure mechanisms through semantic association analysis, path reasoning, and graph embedding reasoning algorithms. Generate a semantic association network based on the knowledge graph and the fault propagation path corresponding to the current fault. The semantic association network and the fault propagation path are used to show the root causes and evolution processes of the faults, providing knowledge support for maintenance decision-making;

[0114] Set a multi-objective optimization problem with the goal of minimizing the fault recurrence rate and the maintenance cost. Set up an objective function in combination with a multi-objective optimization algorithm, randomly select multiple elements from the set of candidate maintenance plans to form an initial population, perform crossover operations and mutation operations on the initial population to generate an offspring population, perform non-dominated sorting on the offspring population, calculate the crowding degree of individuals, and determine the retained individuals in combination with the dominance relationship to form a new generation population. Repeat the generation of the offspring population until the preset maximum number of iterations is reached, and finally generate the Pareto front, representing the optimal trade-off solution set between the fault recurrence rate and the maintenance cost;

[0115] Sort according to the Pareto solution set, and select the first individual as the optimal maintenance decision. Among them, the optimal maintenance decision comprehensively considers the fault prediction results, candidate maintenance plans, knowledge graph reasoning, and multi-objective optimization, and can minimize the fault recurrence rate and maintenance cost while ensuring the feasibility and effectiveness of the maintenance plan.

[0116] In this embodiment, a multi-time scale fault prediction model is adopted to predict the faults of the CNC machine tool from three levels: short-term, medium-term, and long-term, providing an important basis for maintenance decisions, helping to improve the forward-looking and pertinence of maintenance strategies. Through the collaborative optimization of the global decision-making layer and the sub-task decision-making layer, a set of candidate maintenance plans is generated, which can continuously learn and optimize in the interaction with the environment, adaptively adjust the maintenance strategy to adapt to the changes in the machine tool state, construct a knowledge graph of the CNC machine tool, integrate structured and unstructured data, form a comprehensive semantic representation of the machine tool faults, provide rich domain knowledge support for maintenance decisions, and help to improve the interpretability and credibility of decisions. Taking the minimization of the fault recurrence rate and the minimization of the maintenance cost as the goals, a multi-objective optimization problem is constructed, which can systematically consider the constraints and conflicts between different maintenance goals, generate the globally optimal maintenance decision, and achieve the balance between the maintenance effect and the maintenance cost. In summary, this embodiment realizes the intelligence and optimization of the CNC machine tool maintenance decision, provides a new solution for the intelligent maintenance of the CNC machine tool, and has significant technical advantages and application value.

[0117] In an alternative embodiment,

[0118] The loss function corresponding to the long-term prediction model based on the deep survival model is shown in the following formula:

[0119] ;

[0120] where L represents the loss value, δ g represents the indicator function, indicating whether the g-th sample is the time point of the event occurrence, f(x, T) represents the prediction result of the model for the input feature x and time t, T g represents the event occurrence time of the g-th sample, μ represents the regularization coefficient, and W represents the network parameters.

[0121] In this embodiment, the time point when an event occurs is incorporated into the loss function, enabling the model to learn the time characteristics of event occurrence, fully utilize the information of the event occurrence time, and improve the accuracy of long-term fault prediction. Through multi-layer non-linear transformation, the model can capture the complex interactions and high-order relationships between features, extract the most effective feature representations for long-term fault prediction. The survival probability estimation provides a quantitative risk assessment basis for formulating long-term maintenance strategies. The regularization constraint can improve the generalization ability of the model, enabling it to make reliable long-term predictions even under unknown machine tool states. The deep survival model adopts an end-to-end training method, directly learning the prediction model from the input features and the event occurrence time through the backpropagation algorithm, avoiding the need for manual feature engineering, simplifying the model construction process, and being able to automatically optimize the entire process of feature extraction and survival probability estimation, improving the overall performance of the model. In summary, this embodiment can effectively process machine tool state data, provide a probability estimate of long-term fault risk, and provide a quantitative decision-making basis for formulating long-term maintenance strategies.

[0122] Figure 2 FIG. is a schematic structural diagram of an intelligent maintenance decision-making system for a numerically controlled machine tool based on digital twin according to an embodiment of the present invention, as Figure 2 shown, the system includes:

[0123] A first unit, configured to obtain the structural data and simulation accuracy requirements of a numerically controlled machine tool, determine the optimal mesh size and distribution parameters of a finite element model in combination with an adaptive mesh division algorithm, obtain the operating conditions of the numerically controlled machine tool in the finite element model and add them to a multi-physical field coupling modeling module, generate a machine tool digital twin model and fuse it with a physical field, generate a machine tool operation degradation model in combination with a mapping relationship, and continuously optimize it through pre-acquired large data of machine tool operation and a deep transfer learning model to obtain a self-evolving digital twin model;

[0124] A second unit, configured to construct a sensing network based on the self-evolving digital twin model, comprehensively sense the numerically controlled machine tool and obtain machine tool operation parameters through distributed adaptive sampling, preprocess the machine tool operation parameters to obtain multi-modal sensing data and add them to a multi-view learning fusion model, extract multi-view latent features and determine the latent correlation features between different modalities, and obtain a machine tool health state assessment result in combination with the self-evolving digital twin model;

[0125] A third unit, configured to generate multi-level fault prediction results based on the evaluation results of the machine tool health state, combine with a pre-set multi-time-scale fault prediction model, combine historical maintenance cases, adaptively determine candidate maintenance plans through a maintenance strategy optimization algorithm based on deep reinforcement learning, add the candidate maintenance plans and the multi-level fault prediction results to a knowledge graph construction module, generate a semantic association network by associating machine tool fault causes and failure mechanisms, and combine with a multi-objective optimization algorithm to solve for an optimal maintenance decision.

[0126] In a third aspect of the embodiments of the present invention,

[0127] a kind of electronic device is provided, including:

[0128] a processor;

[0129] a memory for storing instructions executable by the processor;

[0130] wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0131] In a fourth aspect of the embodiments of the present invention,

[0132] a computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0133] The present invention can be a method, device, system and / or computer program product. The computer program product can include a computer-readable storage medium, on which computer-readable program instructions for executing various aspects of the present invention are uploaded.

[0134] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it is still possible to modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. The intelligent maintenance decision-making method of CNC machine tools based on digital twins is characterized by: include: The structural data and simulation accuracy requirements of the CNC machine tool are obtained, and the optimal grid size and distribution parameters of the finite element model are determined in combination with an adaptive meshing algorithm. The operating conditions of the CNC machine tool in the finite element model are obtained and added to the multi-physics field coupling modeling module, and a machine tool digital twin model is generated and integrated with the physical field. A machine tool operation degradation model is generated in combination with a mapping relationship, and a self-evolving digital twin model is obtained by continuous optimization through pre-acquired machine tool operation big data and a deep transfer learning model, including: based on a pre-constructed multi-granularity finite element model, the boundary conditions of the CNC machine tool under processing conditions are obtained, and the boundary conditions are added to the multi-granularity finite element model as loads and constraints. Combined with the physical field corresponding to the CNC machine tool, a simulation numerical solution is performed in combination with a physical field coupling solution method to obtain the corresponding physical field behavior, and combined with sensor mapping technology, the state data of the CNC machine tool is collected and mapped to the corresponding nodes of the virtual model to obtain a multi-physics field digital twin model, and the degradation process of the CNC machine tool is obtained. In combination with the pre-acquired machine tool operation big data, the degradation precursor features are extracted to construct a machine tool degradation model; Through deep transfer learning technology, combined with the machine tool operation big data, the common features under different working conditions are extracted through deep neural networks, the degradation law is determined, and the machine tool degradation model is adaptively adjusted under the new working condition in combination with transfer learning to obtain the self-evolving digital twin model; A sensor network is constructed based on the self-evolving digital twin model, the CNC machine tool is fully sensed and the machine tool operating parameters are obtained through distributed adaptive sampling, the machine tool operating parameters are preprocessed, multi-modal sensing data is obtained and added to the multi-view learning fusion model, multi-view potential features are extracted and potential correlation features between different modes are determined, and the machine tool health status assessment result is obtained in combination with the self-evolving digital twin model; Based on the machine tool health status assessment result, a multi-level fault prediction result is generated in combination with a pre-set multi-time scale fault prediction model. Combined with historical maintenance cases, candidate maintenance plans are adaptively determined through a maintenance strategy optimization algorithm based on deep reinforcement learning. The candidate maintenance plans and the multi-level fault prediction results are added to the knowledge graph construction module. By associating the machine tool failure causes and failure mechanisms, a semantic association network is generated and combined with a multi-objective optimization algorithm to solve the optimal maintenance decision.

2. The method according to claim 1, characterized in that Obtain the structural data and simulation accuracy requirements of the CNC machine tool, and combine the adaptive meshing algorithm to determine the optimal mesh size and distribution parameters of the finite element model, including: Acquire structural parameters of the CNC machine tool through three-dimensional scanning, wherein the structural parameters include the size, shape and assembly relationship of the parts of the CNC machine tool, and acquire the simulation accuracy requirements corresponding to the CNC machine tool based on the service condition and failure mode of the CNC machine tool; A feature analysis is performed on the CNC machine tool, structural stress and hot spots are extracted as key features, feature attributes are determined, and the grid scales of different areas are determined in combination with simulation accuracy requirements. Areas with concentrated stress and sudden cross-section changes are regarded as weak areas, and the grid density in the weak areas is increased. High-order grid units are combined to improve the approximation accuracy of field variables within the units, complete grid division, and obtain the optimal grid size and distribution parameters. A multi-granularity finite element model of the current CNC machine tool is built using finite element software.

3. The method according to claim 2, characterized in that Combined with transfer learning, the machine tool degradation model is adaptively adjusted under new working conditions as shown in the following formula: Among them, θ t represents the network parameters under the target working condition, L t (θ) represents the loss function under the target working condition, dist(θ s , θ) represents the distance measure between the network parameters of the source condition and the target condition, λ1 represents the distance measure weight factor, λ2 represents the regularization term weight factor, and R(θ) represents the regularization term.

4. The method according to claim 1, characterized in that A sensor network is constructed based on the self-evolving digital twin model, the CNC machine tool is fully sensed and the machine tool operating parameters are obtained through distributed adaptive sampling, the machine tool operating parameters are preprocessed, multi-modal sensing data is obtained and added to the multi-view learning fusion model, multi-view potential features are extracted and potential correlation features between different modes are determined, and the machine tool health status assessment results obtained by combining the self-evolving digital twin model include: Based on the self-evolving digital twin model, a sensor network covering all components and functional units of the CNC machine tool is constructed, wherein the sensor network includes multiple types of sensors corresponding to vibration, temperature, stress, displacement and current; Based on the sensor network, the CNC machine tool is fully sensed, combined with the adaptive sampling trigger pre-set at the edge node, the sampling frequency of the sensor network is dynamically adjusted according to the real-time working condition and degradation state of the CNC machine tool, the machine tool operation parameters are obtained, and the machine tool operation parameters are pre-processed to obtain the multimodal perception data; Based on the multimodal perception data, for each modality, a corresponding feature extraction subview is set, and based on the feature extraction subview, trend features and period features are extracted from the data in each modality as multi-view potential features, and based on the multi-view potential features, a multi-view fusion algorithm is combined to generate a cross-modal common representation by determining linear correlations and nonlinear correlations between different modalities, and a multi-view alignment strategy is combined to determine potential correlation features; The potential correlation features are added to the self-evolving digital twin model, and combined with real-time monitoring data to generate the machine tool health status assessment result.

5. The method according to claim 4, characterized in that Based on the feature extraction subview, the trend features and period features of the data in each mode are extracted as shown in the following formula: Where WT(a, b) represents the periodic characteristics at scale a and position b, a represents the scale parameter, which is used to control the width of the function, b represents the translation parameter, which is used to determine the position of the function on the time axis, x(t) represents the original signal, and ψ() represents the mother wavelet function; Among them, y(t) represents the trend characteristics of the data in the current mode, w j represents the weight coefficient of the jth scale, M j represents the length of the sliding window of the jth scale, J represents the total number of scales, and i is an index variable used to represent the time step back relative to the current time point t.

6. The method according to claim 1, characterized in that Based on the machine tool health status assessment result, a multi-level fault prediction result is generated in combination with a pre-set multi-time scale fault prediction model. In combination with historical maintenance cases, a candidate maintenance plan is adaptively determined through a maintenance strategy optimization algorithm based on deep reinforcement learning. The candidate maintenance plan and the multi-level fault prediction result are added to the knowledge graph construction module. By associating the cause of the machine tool failure and the failure mechanism, a semantic association network is generated and combined with a multi-objective optimization algorithm to solve the optimal maintenance decision, including: Obtain the machine tool health status assessment result and add it to a pre-set multi-time scale fault prediction model, generate a short-term prediction result through a short-term prediction model based on a gated recurrent unit, determine a medium-term prediction result through a medium-term prediction model based on a convolutional long short-term memory network, generate a long-term prediction result through a long-term prediction model based on a deep survival model, and combine the prediction results to obtain the multi-level fault prediction result; Based on the multi-level fault prediction results, a global decision layer and a subtask decision layer are set in combination with working conditions, wherein the global decision layer takes the health state of the CNC machine tool as the state, the candidate maintenance plan set as the action, and the maintenance performance index as the reward, and the subtask decision layer takes the subtask state as the state, the subtask execution plan as the action, and the subtask performance index as the reward; Based on the global decision layer and the subtask decision layer, the global optimal decision and the local optimal decision are determined in combination with the active exploration mechanism, the global optimal decision is collaboratively optimized in combination with the Nash equilibrium strategy, and a global maintenance solution set is generated. The subtask decision layer is distributedly optimized through a multi-agent reinforcement learning algorithm to generate a subtask execution solution, and the subtask execution solution and the global maintenance solution set are integrated to obtain a candidate maintenance solution, and the integration is repeated to obtain a candidate maintenance solution set; Design the CNC machine tool domain ontology, define core concepts and relationship types, extract entities and attribute information corresponding to the entities from structured data and unstructured data, perform knowledge fusion based on rule and graph embedding methods, generate a CNC machine tool knowledge graph, add the candidate maintenance plan and the multi-level fault prediction results to the CNC machine tool knowledge graph, determine the fault cause and failure mechanism through semantic association analysis and path reasoning and graph embedding reasoning algorithm, generate a semantic association network and the fault propagation path corresponding to the current fault; With the goal of minimizing the failure recurrence rate and the maintenance cost, the objective function is set in combination with a multi-objective optimization algorithm and multiple elements are randomly selected from the candidate maintenance plan set to form an initial population. The initial population is subjected to crossover and mutation operations to generate a child population and perform non-dominated sorting. The congestion degree is calculated and the retained individuals are determined in combination with the dominance relationship. The operation is repeated to generate a Pareto frontier and sort them according to the Pareto solution set, and the individual ranked first is selected as the optimal maintenance decision.

7. The method according to claim 6, characterized in that The loss function corresponding to the long-term prediction model based on the deep survival model is shown in the following formula: Among them, L represents the loss value, δ g represents the indicative function, indicating whether the g-th sample is the time point when the event occurs. f(x, T) represents the prediction result of the model for the input feature x and time t. g represents the event occurrence of the g-th sample, μ represents the regularization coefficient, and W represents the network parameter.

8. An intelligent maintenance decision system for CNC machine tools based on digital twins, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to obtain the structural data and simulation accuracy requirements of the CNC machine tool, determine the optimal mesh size and distribution parameters of the finite element model in combination with the adaptive meshing algorithm, obtain the operating conditions of the CNC machine tool in the finite element model and add them to the multi-physics field coupling modeling module, generate a machine tool digital twin model and integrate it with the physical field, generate a machine tool operation degradation model in combination with the mapping relationship, and continuously optimize through the pre-acquired machine tool operation big data and the deep transfer learning model to obtain a self-evolving digital twin model; The second unit is used to build a sensor network based on the self-evolving digital twin model, comprehensively perceive the CNC machine tool and obtain the machine tool operating parameters through distributed adaptive sampling, pre-process the machine tool operating parameters, obtain multi-modal perception data and add it to the multi-view learning fusion model, extract multi-perspective potential features and determine the potential correlation features between different modes, and obtain the machine tool health status assessment result in combination with the self-evolving digital twin model; The third unit is used to generate a multi-level fault prediction result based on the machine tool health status assessment result in combination with a pre-set multi-time scale fault prediction model, and in combination with historical maintenance cases, adaptively determine candidate maintenance plans through a maintenance strategy optimization algorithm based on deep reinforcement learning, add the candidate maintenance plans and the multi-level fault prediction results to the knowledge graph construction module, and generate a semantic association network by associating the cause of machine tool failure and the failure mechanism, and combine it with a multi-objective optimization algorithm to solve and obtain the optimal maintenance decision.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images