A method of industrial mobile equipment maintenance

Abstract

The application discloses an industrial equipment maintenance method, and belongs to the technical field of equipment management. The method comprises the following steps: determining a multi-dimensional representation sub-model containing geometric, physical, data-driven and knowledge models based on obtained multi-source priori and monitoring data; generating a fusion sub-model and a reduced-order solving model by combining the sub-model based on a preset multi-model bidirectional closed-loop fusion mechanism and a multi-physics field collaborative solving and reduction strategy, wherein the fusion mechanism represents cross-domain interaction, mechanism constraint and parameter reverse correction among the sub-models, and the reduction strategy represents differential collaborative calculation and low-dimensional reduction mapping rules of different coupling strength physical fields; constructing a multi-scale equipment maintenance model based on the fusion sub-model and the reduced-order solving model; and inputting real-time monitoring data into the model to obtain a maintenance strategy containing fault root cause positioning and scheme optimization. The method realizes four-layer model deep fusion and efficient reduction simulation, solves the problems of model fragmentation and lack of mechanism constraint, and improves the whole life cycle management and control capability.

Description

Technical Field

[0001] This application relates to the field of equipment management technology, and in particular to methods for maintaining industrial moving equipment. Background Technology

[0002] Industrial moving equipment, such as pumps, fans, motors, and compressors, are core and critical equipment in industrial production processes. Their operational stability and reliability directly determine production efficiency, product quality, and production safety. With the deep advancement of Industry 4.0 and intelligent manufacturing, traditional management models based on manual inspection and periodic maintenance can no longer meet the demands for high precision, high real-time performance, and full lifecycle control. Digital twin technology has become the key to solving these problems.

[0003] However, existing digital twin methods for industrial dynamic equipment have core technological limitations: existing twin models mostly focus on surface-level condition monitoring, lacking in-depth modeling of the equipment's operation and maintenance mechanisms. Because geometry, physics, data-driven approaches, and knowledge models are often isolated, failing to form a two-way closed-loop fusion mechanism, data-driven models lack prior constraints and interpretability based on physical mechanisms, and knowledge models struggle to acquire deep features for reasoning and verification. This fragmentation results in the system "knowing only the result, not the cause," making it difficult to achieve fault root cause location, accurate prediction of remaining lifespan, and in-depth optimization of maintenance plans, thus failing to meet the core requirements of deep collaborative management and control throughout the entire lifecycle of dynamic equipment. Summary of the Invention

[0004] This application provides a method for maintaining industrial moving equipment, which at least solves the problem of difficulty in locating the root cause of failure, accurately predicting the remaining lifespan, and deeply optimizing the maintenance plan, thus failing to meet the requirements of in-depth collaborative management and control of the entire life cycle of moving equipment.

[0005] In a first aspect, this application provides a method for maintaining industrial moving equipment, the method comprising: Based on the acquired multi-source prior and monitoring data, a multi-dimensional representation sub-model is determined. The multi-source prior and monitoring data includes multi-dimensional mechanism structure data and historical operation data of industrial dynamic equipment. The multi-dimensional representation sub-model includes a geometric model, a physical model, a data-driven model, and a knowledge model. Based on the preset multi-model bidirectional closed-loop fusion mechanism and multi-physics field collaborative solution and order reduction strategy, combined with the sub-models of the multi-dimensional representation, a fused sub-model and a reduced-order solution model are generated. Based on the aforementioned fusion sub-model and reduced-order solution model, a multi-scale equipment maintenance model is constructed. The acquired real-time multi-source operation monitoring data is input into a multi-scale equipment maintenance model to obtain a maintenance strategy, which includes fault root cause localization and maintenance scheme optimization.

[0006] The above technical solution constructs a four-layer model encompassing geometry, physics, data-driven approaches, and knowledge. Based on a two-way closed-loop fusion mechanism, it achieves cross-domain information interaction for mechanism constraints, parameter correction, and reasoning verification. Simultaneously, it employs multi-physics collaborative solving and order reduction strategies to balance simulation accuracy and efficiency. Ultimately, it constructs a multi-scale maintenance model that outputs deeply optimized maintenance strategies. Its beneficial effects include: achieving complementary advantages between mechanism and data; solving the problems of data-driven models lacking physical constraints and prone to generalization, and inaccurate physical model parameters; achieving high real-time simulation through order reduction models; and realizing fault root cause localization and closed-loop verification of maintenance solutions based on knowledge reasoning. This enhances the depth, safety, and intelligence level of the entire lifecycle management of industrial dynamic equipment.

[0007] Secondly, this application provides an industrial equipment maintenance system, the system comprising: The model determination module is used to determine multi-dimensional representation sub-models based on the acquired multi-source prior and monitoring data. The multi-source prior and monitoring data include multi-dimensional mechanism structure data and historical operation data of industrial dynamic equipment. The multi-dimensional representation sub-models include geometric models, physical models, data-driven models and knowledge models. The model generation module is used to generate a fused sub-model and a reduced-order solution model based on a preset multi-model bidirectional closed-loop fusion mechanism and a multi-physics field collaborative solution and order reduction strategy, combined with the sub-models represented by the multi-dimensional representation. The model building module is used to construct a multi-scale equipment maintenance model based on the fusion sub-model and the reduced-order solution model; The strategy output module is used to input the acquired real-time multi-source operation monitoring data into the multi-scale equipment maintenance model to obtain the maintenance strategy, which includes fault root cause localization and maintenance scheme optimization.

[0008] Thirdly, this application provides an electronic device including one or more processors and one or more memories, wherein at least one piece of program code is stored in the one or more memories, the program code being loaded and executed by the one or more processors to implement the operations performed by the industrial equipment maintenance method.

[0009] Fourthly, this application also provides a computer-readable storage medium storing at least one piece of program code, which is loaded and executed by a processor to implement the operations performed by the industrial equipment maintenance method.

[0010] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above-described industrial equipment maintenance methods. Attached Figure Description

[0011] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0012] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart illustrating an industrial equipment maintenance method provided in this application embodiment. Figure 1 ; Figure 2 A flowchart illustrating an industrial equipment maintenance method provided in this application embodiment. Figure 2 ; Figure 3 A flowchart illustrating an industrial equipment maintenance method provided in this application embodiment. Figure 3 ; Figure 4 A flowchart illustrating an industrial equipment maintenance method provided in this application embodiment. Figure 4 ; Figure 5 A flowchart illustrating an industrial equipment maintenance method provided in this application embodiment. Figure 5 ; Figure 6 A flowchart illustrating an industrial equipment maintenance method provided in this application embodiment. Figure 6 ; Figure 7 A flowchart illustrating an industrial equipment maintenance method provided in this application embodiment. Figure 7 ; Figure 8 This application provides a schematic diagram of the structure of an industrial dynamic equipment maintenance system. Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0015] It should be noted that, in the description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The terms "first," "second," etc., in this application are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0016] In related technologies, existing industrial equipment maintenance often relies on periodic inspections or threshold alarms based on single data points, lacking insight into the deep operating mechanisms of the equipment. Digital twin technology, introduced in recent years, often uses geometric, physical, data, and knowledge models in isolation, failing to achieve deep fusion and bidirectional feedback of multi-source heterogeneous data. This results in low model prediction accuracy and poor generalization ability. Furthermore, complex multiphysics coupled simulation calculations are too time-consuming, making it difficult to meet the needs of real-time maintenance decision-making. Moreover, the formulation of maintenance plans often lacks simulation-based virtual verification, posing execution risks.

[0017] To address the aforementioned technical challenges, this application proposes a maintenance scheme for industrial dynamic equipment that integrates multi-model bidirectional closed-loop fusion and multi-physics field collaborative order reduction. By constructing a digital twin model that integrates geometry, physics, data-driven approaches, and knowledge, the scheme achieves constraints on data by mechanisms, corrections of mechanisms by data, and verification of reasoning by knowledge. Furthermore, by combining the order reduction solution model with the scheme, the simulation efficiency is improved, enabling high real-time and high-precision fault root cause localization and maintenance scheme optimization.

[0018] The technical solutions provided in this application are mainly applied to the field of industrial dynamic equipment management, which has extremely high requirements for operational continuity, safety, and intelligent collaboration, especially in scenarios that require the integration of multi-source heterogeneous data and complex physical mechanisms for in-depth condition assessment and maintenance decision-making. Specifically, they mainly include the following application scenarios: 1. Predictive maintenance scenarios for core dynamic equipment in the petrochemical industry In the petrochemical industry, compressors, pumps, and other moving equipment operate under harsh conditions such as high temperature, high pressure, and corrosion for extended periods. Traditional periodic maintenance often leads to over-maintenance or neglect. The technical solution proposed in this application can be widely applied in the following areas: In-depth assessment of unit condition: By integrating design drawings, fluid / structure coupling simulation, vibration and temperature monitoring data and expert diagnostic rules, a multi-scale twin model is constructed to accurately assess complex fault conditions such as rotor imbalance, misalignment, and blade wear.

[0019] Virtual verification of maintenance plan: After formulating a plan for compressor rotor replacement or bearing adjustment, the equipment response after maintenance is quickly simulated and deduced through a reduced-order solution model, avoiding secondary damage or start-up failure caused by blind maintenance.

[0020] 2. Intelligent Operation and Maintenance Scenarios for Power and New Energy Generation Equipment In fields such as thermal power generation and wind power generation, the stable operation of equipment such as wind turbines and motors is directly related to the security of the power grid. Wind turbine drivetrain health management: Combining multibody dynamics physical models with SCADA operation data, the model parameters are corrected in real time through bidirectional closed-loop fusion, the remaining life of gearbox and main shaft is predicted, and the root causes such as bearing wear are located based on knowledge models to formulate the optimal lubrication or replacement strategy.

[0021] Steam turbine flow path optimization and maintenance: Based on thermodynamic mechanism-constrained data-driven models, accurately predict steam turbine efficiency decline and blade fouling, and guide decisions on online cleaning or shutdown maintenance.

[0022] 3. High-reliability management scenarios in metallurgy and heavy machinery Heavy machinery such as metallurgical blast furnaces and rolling mills experiences heavy loads and strong impacts, resulting in high maintenance costs. Rolling mill main drive system status monitoring: Using physical models to generate virtual samples of extreme working conditions to make up for data gaps, training a highly robust data-driven model for early fault warning, and using knowledge reasoning to verify and eliminate false alarms, thus achieving precise maintenance.

[0023] After introducing the implementation environment and application scenarios of the embodiments of this application, the technical solutions provided by the embodiments of this application are described below. (See also...) Figure 1 Taking an industrial edge intelligent controller or cloud twin platform as the executing entity, the industrial dynamic equipment maintenance method includes the following steps.

[0024] Step S101: Based on the acquired multi-source prior and monitoring data, determine the multi-dimensional characterization sub-model.

[0025] The multi-source prior and monitoring data includes multi-dimensional mechanistic structure data and historical operation data of industrial dynamic equipment, and the multi-dimensional representation sub-models include geometric models, physical models, data-driven models and knowledge models.

[0026] Multi-source prior and monitoring data are used to comprehensively characterize the static mechanism and dynamic operating characteristics of equipment. Multi-dimensional mechanism structure data can include equipment design drawings, material properties, assembly relationships, multi-physics coupling parameters and boundary conditions, equipment design specifications and fault diagnosis rules, etc.; historical operating data can include historical sensor monitoring data, historical fault records and experience-based maintenance data, etc.

[0027] The multidimensional representation sub-model is used to digitally map industrial moving equipment from different dimensions. The geometric model provides a 1:1 parametric geometric topology and assembly relationship carrier for the equipment; the physical model provides multi-physics coupling simulation and mechanism constraints; the data-driven model provides data-based anomaly identification, fault early warning and remaining life prediction capabilities; and the knowledge model provides fault root cause location and maintenance decision support based on expert rules and experience knowledge.

[0028] Specifically, taking a petrochemical centrifugal compressor as an example, a multi-source data acquisition network is used to acquire CAD design drawings of the compressor, the elastic modulus of rotor materials, bearing assembly clearance and other mechanistic structural data, as well as historical vibration spectrum, bearing temperature, outlet pressure and other operating data. These data are used to drive the construction of a geometric model, a fluid-structure coupled physical model, an anomaly detection model based on an autoencoder, and a knowledge model containing the association rules of "vibration anomaly - insufficient lubrication - bearing wear", forming a set of multi-dimensional characterization sub-models covering the characteristics of the entire life cycle of the equipment.

[0029] Step S102: Based on the preset multi-model bidirectional closed-loop fusion mechanism and multi-physics field collaborative solution and order reduction strategy, and combined with the multi-dimensional representation sub-model, generate the fusion sub-model and the order reduction solution model.

[0030] The multi-model bidirectional closed-loop fusion mechanism is used to characterize the cross-domain information interaction, mechanistic constraints, and reverse parameter correction relationships between sub-models, breaking down model fragmentation. For example, the physical model provides mechanistic prior constraints to the data-driven model, the data-driven model provides dynamic parameter correction to the physical model, the data-driven model performs feature semantic transformation to the knowledge model, and the knowledge model performs rule reasoning verification and label optimization for the data-driven model.

[0031] Multi-physics collaborative solution and order reduction strategies are used to characterize the differentiated collaborative computation logic of physical fields with different coupling strengths and the low-dimensional order reduction mapping rules of the physical model. For strongly coupled fields, fully coupled implicit solution is used to ensure accuracy, while for weakly coupled fields, co-simulation explicit solution is used to improve efficiency. At the same time, order reduction rules are used to map the high-dimensional physical model to a low-dimensional space to achieve real-time simulation.

[0032] Specifically, based on the above mechanisms and strategies, the four-dimensional representation sub-model of the compressor is deeply fused and reconstructed computationally: virtual fault samples generated by the physical model are added to the data-driven model to improve the training effect, while the simulation and measured deviations identified by the data-driven model are used to reversely correct the friction coefficient of the physical model; implicit solutions are used for the fluid-structure strongly coupled field, explicit solutions are used for the heat-structure weakly coupled field, and the million-degree-of-freedom model is reduced to hundreds of dimensions through snapshot matrix and Galerkin projection, generating a fused sub-model with bidirectional data flow capability and a reduced-order solution model with efficient real-time computing capability.

[0033] Step S103: Construct a multi-scale equipment maintenance model based on the fusion sub-model and the reduced-order solution model.

[0034] The multi-scale equipment maintenance model is a comprehensive twin model that integrates a multi-model fusion architecture, a virtual-real calibration mechanism, and a result fusion mechanism. It is used to support the entire process of maintenance management from real-time simulation to decision output.

[0035] Specifically, the reduced-order solution model is embedded into the physical model layer of the fusion sub-model to provide real-time simulation and deduction capabilities. A virtual-real calibration mechanism based on timestamp alignment and logical clock is introduced to ensure the consistency between the model and the physical compressor. A weighted Bayesian network algorithm is used to fuse multi-dimensional outputs such as geometric deviation, physical simulation deviation, and failure probability to calculate the probability of equipment health status. Finally, a complete multi-scale equipment maintenance model for the compressor is constructed.

[0036] Step S104: Input the acquired real-time multi-source operation monitoring data into the multi-scale equipment maintenance model to obtain the maintenance strategy.

[0037] The maintenance strategy includes root cause analysis and maintenance plan optimization. This strategy guides on-site maintenance personnel to perform precise maintenance operations, avoiding over- or under-maintenance.

[0038] Specifically, real-time collected monitoring data on compressor vibration, temperature, and pressure are input into a multi-scale equipment maintenance model. The model is solved in a reduced-order manner to simulate and obtain unmeasurable states such as the internal stress distribution of the rotor. Anomalies are identified by comparing preset thresholds with the baseline of the normal mode. Based on the anomaly-triggered data, the model generates fault warnings and inputs them into a knowledge model to locate the root cause of the fault as "insufficient lubrication leading to bearing wear" through association rule reasoning. The maintenance plan to increase the amount of lubricating oil is formulated based on the remaining life prediction results. Finally, the effectiveness of the plan is verified through virtual debugging, and the optimal maintenance strategy is output.

[0039] This embodiment constructs a four-layer model encompassing geometry, physics, data-driven approaches, and knowledge. Based on a two-way closed-loop fusion mechanism, it achieves cross-domain information interaction for mechanism constraints, parameter correction, and reasoning verification. Simultaneously, it employs multi-physics collaborative solving and order reduction strategies to balance simulation accuracy and efficiency. Ultimately, it constructs a multi-scale maintenance model that outputs deeply optimized maintenance strategies. Its beneficial effects include: achieving complementary advantages between mechanism and data; solving the problems of data-driven models lacking physical constraints and prone to generalization, and inaccurate physical model parameters; achieving high real-time simulation through order reduction models; and realizing fault root cause localization and closed-loop verification of maintenance solutions based on knowledge reasoning. This enhances the depth, safety, and intelligence level of the entire lifecycle management of industrial dynamic equipment.

[0040] It should be noted that the above steps S101-S104 are a simplified description of the embodiments provided in this application.

[0041] To more clearly illustrate the industrial equipment maintenance method provided in the above embodiments of this application, the following detailed description of this application is provided in conjunction with the accompanying drawings and specific embodiments.

[0042] The methods provided in the embodiments of this application will be described in more detail below with some examples. See also... Figure 2 The determination of the multidimensional representation sub-model based on the acquired multi-source prior and monitoring data includes the following steps: Step S201: Determine the geometric model based on the equipment design drawings, material properties and assembly relationships in the acquired multi-source prior and monitoring data.

[0043] Specifically, based on the CAD design drawings of the compressor rotor, a 1:1 high-precision parametric geometric model is constructed using 3D modeling tools. The assembly relationships such as impeller dimensions and bearing span are preserved to form a generalized geometric template, providing a mesh foundation for the physical model.

[0044] Step S202: Determine the physical model based on the multi-physics coupling parameters and boundary conditions in the acquired multi-source prior and monitoring data.

[0045] Specifically, based on rotor dynamics and fluid mechanics theory, a coupled physical model covering multiple physical fields such as mechanical vibration and fluid excitation is constructed to describe the dynamic response characteristics of the compressor under different operating conditions.

[0046] Step S203: Based on the historical operating data and fault records in the acquired multi-source prior and monitoring data, determine the data-driven model.

[0047] Specifically, machine learning algorithms are used to extract time-domain and frequency-domain features based on historical vibration spectrum, temperature trends and other operational data and fault records, and to establish normal operation baseline, fault classification model and remaining life prediction model for unknown fault detection and trend prediction.

[0048] Step S204: Determine the knowledge model based on the equipment design specifications, fault diagnosis rules and experience maintenance data in the acquired multi-source prior and monitoring data.

[0049] Specifically, based on the event-state knowledge graph, compressor design specifications, fault diagnosis rules, and historical maintenance experience are integrated. The fault diagnosis rules, such as the correspondence between spectral characteristics and fault types, are used to construct a human-readable and computer-deployable knowledge model to achieve fault root cause location and maintenance solution recommendation.

[0050] Step S205: By constructing a unified semantic ontology, the heterogeneous parameters and outputs of the geometric model, physical model, data-driven model and knowledge model are uniformly mapped to standard entities and attributes, and the multidimensional representation sub-model is determined.

[0051] A unified semantic ontology is used to address the issues of heterogeneous parameter naming and semantic inconsistencies between different models. Standard entities are used to represent the unified semantic mapping results across models. Core ontology classes include device classes, component classes, parameter classes, fault classes, and maintenance classes.

[0052] Specifically, a core ontology class and object attributes and data attributes are defined to construct a unified semantic ontology for industrial dynamic equipment. An ontology inference engine is used to perform cross-layer semantic matching and consistency verification, and the "component dimensions" of the geometric model, the "node displacements" of the physical model, the "vibration characteristics" of the data-driven model, and the "fault attributes" of the knowledge model are uniformly mapped to standard entities in the ontology, so as to achieve semantic unification of heterogeneous parameters and outputs, thereby determining a deeply integrated multidimensional representation sub-model.

[0053] This embodiment constructs geometric, physical, data-driven, and knowledge models respectively, and maps the four-layer heterogeneous models to standard entities based on a unified semantic ontology, thereby achieving semantic unification and model integration of multi-source heterogeneous data. Its beneficial effect is that by fusing models of different dimensions, it lays the foundation for subsequent bidirectional data flow and closed-loop collaborative computing, and ensures the accuracy and consistency of cross-layer information interaction.

[0054] In some embodiments, see Figure 3 The pre-defined multi-model bidirectional closed-loop fusion mechanism, combined with the multi-dimensional representation sub-model, generates a fusion sub-model, including: Step S301: Based on the preset multi-model bidirectional closed-loop fusion mechanism, determine the linkage relationship, correction relationship and verification relationship.

[0055] Linkage relationships are used to represent the unique mesh mapping and parameterized linkage between the geometric model and the physical model; correction relationships are used to represent the mechanistic prior constraints of the physical model on the data-driven model and the dynamic parameter correction of the physical model by the data-driven model; verification relationships are used to represent the feature semantic transformation of the data-driven model to the knowledge model and the rule reasoning verification and label optimization of the knowledge model on the data-driven model.

[0056] Step S302: Based on the linkage relationship, correction relationship and verification relationship, and combined with the multi-dimensional representation sub-model, perform bidirectional data flow and closed-loop collaborative calculation processing to generate the fusion sub-model.

[0057] Specifically, the geometric model outputs parameterized geometric topology to generate the mesh and boundary conditions of the physical model, and the physical simulation results drive the correction of geometric parameters; the physical model generates virtual samples and mechanistic constraints, which are embedded in the loss function of the data-driven model, and the data-driven model outputs the deviation between simulation and measurement to correct the physical model parameters; the data-driven model outputs fault characteristics and early warning results, which are then converted into inputs to the knowledge model, and the knowledge model uses reasoning to verify false alarms and optimize the labels of the data-driven model. Through the above bidirectional flow and closed-loop collaboration, a deeply integrated fusion sub-model is generated.

[0058] Furthermore, the prior mechanistic constraints of the physical model on the data-driven model include: Using the physical model, virtual samples under different operating conditions and fault levels are generated; The virtual samples are added to the training dataset of the data-driven model; A mechanistic constraint term is embedded in the loss function of the data-driven model. The mechanistic constraint term is used to restrict the prediction results to meet the wear law and fatigue life curve of the equipment.

[0059] The data-driven model's parameter correction of the physical model includes: An autoencoder based on the comparison of virtual and real data is used to construct a parameter deviation identification model; The simulation data of the physical model and the measured data of the physical device are input into the parameter deviation identification model, and the key parameter deviation values ​​of the physical model are output. The gradient descent algorithm is used to correct the parameters of the physical model based on the deviation values ​​of the key parameters.

[0060] This embodiment achieves deep complementarity and closed-loop feedback between geometry, physics, data-driven approaches, and knowledge models by determining the linkage, correction, and verification relationships and performing bidirectional data flow and closed-loop collaborative computation. Its beneficial effects are that physical mechanism constraints improve the generalization and interpretability of the data-driven model, measured data correction improves the accuracy of the physical model, knowledge reasoning verification reduces the false alarm rate of the data-driven model, and label optimization feeds back to improve the training effect of the data-driven model, solving the problem of knowing only the result but not the cause caused by model fragmentation.

[0061] In some embodiments, see Figure 4 The multi-physics collaborative solution and order reduction strategy, combined with the multi-dimensional representation sub-model, generates a reduced-order solution model, including: Step S401: Based on the multi-physics collaborative solution and order reduction strategy, determine the solution logic and order reduction rules.

[0062] The solution logic is used to represent the differentiated collaborative computing method, which adopts a fully coupled implicit solution for strongly coupled physical fields and a joint simulation explicit solution for weakly coupled physical fields based on the differences in physical field coupling strength and time scale. The order reduction rule is used to represent the order reduction mapping method, which extracts the dominant basis vectors based on the singular value decomposition of the snapshot matrix and maps the high-dimensional physical model to the low-dimensional modal coordinate space through Galerkin projection.

[0063] Step S402: Based on the solution logic and order reduction rules, and combined with the sub-model of multidimensional representation, perform partitioned solution and low-dimensional projection order reduction calculation to generate a reduced-order solution model.

[0064] Specifically, for the strongly coupled physical fields of the compressor rotor, such as mechanical-electromagnetic and fluid-structure, a fully coupled implicit solution is adopted to ensure accuracy. For the weakly coupled physical fields, such as thermal-structure and thermal-fluid, a joint simulation explicit solution is adopted to improve efficiency. Cross-field data transfer is achieved through interpolation synchronization. At the same time, a snapshot matrix is ​​collected based on orthogonal experimental design. The snapshot matrix is ​​preprocessed and singular value decomposed. The first k left singular vectors are extracted as dominant basis vectors according to the preset energy criterion. The high-dimensional partial differential control equations with millions of degrees of freedom are reduced to a system of ordinary differential equations with hundreds of dimensions through Galerkin projection, generating a reduced-order solution model.

[0065] This embodiment solves the problem by partitioning based on the difference in coupling strength and time scale, and combines snapshot matrix SVD and Galerkin projection for order reduction mapping to generate a reduced-order solution model. Its beneficial effect is that it solves the engineering contradiction between strong coupling accuracy and weak coupling efficiency, while shortening the calculation time of the original high-fidelity model. Under the premise of ensuring simulation error, it realizes real-time simulation of multi-physics coupling, and provides real-time extrapolation capability for online maintenance decision-making.

[0066] In some embodiments, see Figure 5 The construction of a multi-scale equipment maintenance model based on the fusion sub-model and the reduced-order solution model includes: Step S501: Based on the fusion sub-model and the reduced-order solution model, determine the simulation embedding mechanism, the virtual-real calibration mechanism, and the result fusion mechanism.

[0067] The simulation embedding mechanism is used to indicate that the reduced-order solution model is embedded into the physical model layer of the fusion sub-model; the virtual-real calibration mechanism is used to indicate that the virtual and real data are synchronized according to the timestamp alignment and logical clock, and the parameters of the fusion sub-model and the reduced-order solution model are corrected by using the virtual-real deviation through the gradient descent algorithm; the result fusion mechanism is used to indicate that the multidimensional output of the fusion sub-model is fused using the weighted Bayesian network algorithm to calculate the probability of device health status.

[0068] Step S502: Based on the simulation embedding mechanism, virtual-real calibration mechanism and result fusion mechanism, the fusion sub-model and the reduced-order solution model are integrated and closed-loop connected to construct a multi-scale equipment maintenance model.

[0069] Specifically, the reduced-order solution model is embedded into the fusion architecture as a real-time simulation engine for the physical model layer; timing misalignment caused by network transmission is eliminated by timestamp alignment and logical clock; measured data and simulation data are compared regularly; and key parameters such as friction coefficient and damping coefficient are dynamically corrected using gradient descent algorithm to eliminate virtual-real deviations; at the same time, a weighted Bayesian network is constructed with the equipment health status as the root node and geometric deviation, physical simulation deviation, fault probability and fault confidence as intermediate nodes. Weights are assigned based on historical prediction accuracy, and the equipment health status is comprehensively evaluated through joint probabilistic inference.

[0070] This embodiment constructs a multi-scale equipment maintenance model by determining the simulation embedding, virtual-real calibration, and result fusion mechanism, and integrating and closing-loop connecting the fusion sub-model and the reduced-order solution model. Its beneficial effects are that the embedding of the reduced-order model provides high real-time simulation and deduction capabilities, the virtual-real calibration mechanism ensures the consistency between the twin model and the physical entity, and the weighted Bayesian network integrates multi-source uncertain information, thereby improving the robustness and accuracy of health status assessment.

[0071] In some embodiments, see Figure 6 The process of inputting the acquired real-time multi-source operation monitoring data into a multi-scale equipment maintenance model to obtain maintenance strategies includes: Step S601: Input the acquired real-time multi-source operation monitoring data into the multi-scale equipment maintenance model, and perform real-time multi-physics simulation by solving the model in a reduced order to obtain the real-time simulation status of the equipment.

[0072] The real-time simulation status of the equipment includes the stress distribution, temperature field, and dynamic response of key parts of the equipment.

[0073] Specifically, real-time collected monitoring data such as compressor vibration, temperature, and pressure are input into a multi-scale equipment maintenance model, driving the reduced-order solution model in the multi-scale equipment maintenance model to complete multi-physics coupling calculations in a short time, and obtaining stress, deformation, and other states of key nodes inside the rotor that cannot be directly measured.

[0074] Step S602: Compare the real-time multi-source operation monitoring data with the real-time simulation status of the equipment, and compare the preset threshold with the normal mode baseline to identify abnormal states.

[0075] The preset threshold is used to represent the extreme boundaries of the equipment's operating parameters and the safety alarm limits; the normal mode baseline is used to represent the characteristic distribution and state evolution benchmark of the equipment during healthy operation under multiple operating conditions.

[0076] Specifically, real-time data and simulation status are compared simultaneously with hard thresholds and historical health baselines to identify abnormal operating conditions that deviate from the normal range.

[0077] Step S603: Generate fault warnings based on abnormal states and data-driven models.

[0078] Specifically, when an abnormal state is identified, the autoencoder or isolated forest algorithm in the data-driven model is triggered to evaluate the degree and category of the abnormality and output the fault warning level and the prediction of the occurrence time.

[0079] Step S604: Input the abnormal state and fault warning into the knowledge model, perform knowledge reasoning based on association rules, and determine the root cause of the fault.

[0080] Association rules are used to represent the mapping relationship between fault characteristics, fault modes, and fault causes. Specifically, fault warnings and characteristics are input into the knowledge model, and the model is used for reasoning and verification based on the association rules of "fault mode-fault cause-fault characteristic" to eliminate false alarms and locate the deep-seated root cause of the fault, such as insufficient lubrication leading to bearing wear.

[0081] Step S605: Based on the data-driven model, output the updated remaining lifetime prediction results.

[0082] Specifically, by combining the current abnormal state with the historical degradation trajectory, the remaining service life prediction value of the equipment is dynamically updated using lifetime prediction models such as LSTM.

[0083] Step S606: Based on the updated remaining life prediction results and combined with the root cause location of the failure, formulate a maintenance plan.

[0084] Specifically, based on the remaining service life window and root cause analysis results, a personalized maintenance plan is developed, which includes maintenance time, maintenance content, and spare parts replacement schedule.

[0085] Step S607: Verify the effectiveness of the maintenance plan through virtual debugging to obtain the maintenance strategy for industrial equipment.

[0086] Specifically, the operation instructions in the maintenance plan are mapped to a multi-scale equipment maintenance model. The deduction is performed by solving the model in a reduced order to verify whether the equipment can be restored to the normal baseline after the implementation of the plan, thus ensuring the safety and feasibility of the plan.

[0087] This embodiment obtains the internal state through real-time simulation of a reduced-order solution model, identifies anomalies by comparing thresholds and baselines, combines data-driven model early warning with knowledge model root cause localization, and formulates solutions and conducts virtual debugging verification based on remaining lifetime. Its beneficial effects are that it realizes in-depth analysis from anomaly identification to root cause localization, solves the pain point of "knowing only the result but not the cause", and provides a verification environment for maintenance solutions through virtual debugging verification, avoiding secondary risks caused by improper maintenance solutions, and forming a closed-loop maintenance capability from monitoring, diagnosis to decision verification.

[0088] In some embodiments, see Figure 7 The step of verifying the effectiveness of the maintenance plan through virtual debugging to obtain a maintenance strategy for industrial equipment includes: Step S701: Map the maintenance operation instructions and / or parameter adjustment instructions in the maintenance plan to the multi-scale equipment maintenance model, and perform the post-maintenance simulation deduction by the reduced-order solution model in the multi-scale equipment maintenance model to obtain the deduced state response.

[0089] Specifically, the adjustment instructions in the plan, such as replacing bearings and adjusting rotation speed, are input into the multi-scale equipment maintenance model. The reduced-order solution model is then used to quickly simulate and deduce the operating status of the equipment after the maintenance is performed, such as whether the stress distribution has decreased and whether the vibration has been reduced.

[0090] Step S702: When the simulated state response meets the normal mode baseline and the preset threshold is not triggered, the maintenance plan is determined to be effective and output.

[0091] Specifically, if the simulation results show that the equipment status has returned to the healthy baseline and has not reached the safety alarm limit, it means that the solution can effectively eliminate the fault and restore the equipment performance, and is judged to be effective and output for execution.

[0092] Step S703: When the simulated state response does not meet the normal mode baseline or triggers the preset threshold, adjust the maintenance plan and re-perform the simulation until the validity judgment conditions are met, and output the final maintenance strategy.

[0093] Specifically, if the simulation results show that the parameters are still abnormal or trigger the safety threshold, it indicates that the solution has defects. The solution parameters need to be adjusted automatically or manually and the simulation needs to be repeated until the safety and effectiveness conditions are met, and the optimal maintenance strategy is output.

[0094] This embodiment maps maintenance instructions to a model to deduce the state response from a reduced-order solution model, and determines the validity based on baselines and thresholds. If invalid, the model is adjusted and re-deduced. Its beneficial effect is that it provides a closed-loop verification and iterative optimization mechanism for the formulation of maintenance plans. The effect can be predicted before the actual implementation of the plan, effectively avoiding the risk of secondary equipment damage or production stoppage caused by blind operation, and improving the scientificity and safety of maintenance decisions.

[0095] Figure 8 This is a schematic diagram of the structure of an industrial dynamic equipment maintenance system provided in an embodiment of this application. See also... Figure 8 The industrial dynamic equipment maintenance system 800 includes: a model determination module 801, a model generation module 802, a model construction module 803, and a strategy output module 804.

[0096] The model determination module 801 is used to determine a multi-dimensional representation sub-model based on the acquired multi-source prior and monitoring data. The multi-source prior and monitoring data includes multi-dimensional mechanism structure data and historical operation data of industrial dynamic equipment. The multi-dimensional representation sub-model includes a geometric model, a physical model, a data-driven model, and a knowledge model. The model generation module 802 is used to generate a fused sub-model and a reduced-order solution model based on a preset multi-model bidirectional closed-loop fusion mechanism and a multi-physics field collaborative solution and order reduction strategy, combined with the sub-models represented by the multi-dimensional representation. The model building module 803 is used to build a multi-scale equipment maintenance model based on the fusion sub-model and the reduced-order solution model; The strategy output module 804 is used to input the acquired real-time multi-source operation monitoring data into the multi-scale equipment maintenance model to obtain the maintenance strategy, which includes fault root cause localization and maintenance scheme optimization.

[0097] It should be noted that the industrial moving equipment maintenance system provided in the above embodiments is only illustrated by the division of the above functional modules when maintaining industrial moving equipment. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the equipment can be divided into different functional modules to complete all or part of the functions described above. In addition, the industrial moving equipment maintenance system and the industrial moving equipment maintenance method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process is detailed in the industrial moving equipment maintenance method embodiments, which will not be repeated here.

[0098] To support the operation and spatiotemporal decoupling of the aforementioned system modules, this application provides a hardware and underlying system embodiment based on a cloud-edge collaborative architecture. The schematic diagram of the system's cloud-edge collaborative hardware architecture and memory allocation illustrates this underlying system, which serves as the hardware foundation for the aforementioned maintenance system and specifically includes: The heterogeneous computing unit includes a cloud server cluster and an edge intelligent controller. The cloud server is equipped with a high-performance CPU and GPU cluster, which is dedicated to high-fidelity fully coupled simulation of the physical model, snapshot set generation, and offline training of the POD reduced-order model. The edge intelligent controller integrates a multi-core ARM processor and an NPU / FPGA acceleration chip, which is dedicated to real-time online inference of the reduced-order model, virtual-real synchronization verification, and logic control tasks, realizing hardware-level resource isolation and collaboration between heavy cloud computing and light edge computing.

[0099] The cloud-edge collaborative dual-domain architecture specifically includes: The cloud-based training domain corresponds to the offline modeling stage. It runs a Linux operating system and large-scale simulation software (such as ANSYS), and is responsible for generating massive virtual samples based on the full geometric and physical data of the device, performing SVD decomposition to extract dominant basis vectors, constructing a reduced-order solution model, and distributing it to the edge. The edge inference domain corresponds to the online operation phase. It runs a real-time operating system (RTOS) and a lightweight AI inference engine, responsible for receiving real-time sensor data, performing millisecond-level inference of the reduced-order model, running data-driven models and safety constraint logic, and realizing real-time perception of the field status and rapid issuance of maintenance instructions.

[0100] Cross-domain high-speed data bus: Enables data exchange between the cloud and the edge via 5G URLLC or industrial Ethernet. Data transmission employs a priority sorting and compression mechanism, prioritizing real-time control commands and emergency field data to ensure that hard real-time control and inference tasks at the edge are not interfered with by large data volume synchronization; simultaneously, it uses timestamp alignment and logical clock mechanisms to ensure strict time synchronization between virtual and physical data.

[0101] Hardware implementation of safety constraints and virtual-real verification: Embedded within the edge inference domain, implemented using FPGA hardware circuitry, it is responsible for high-speed comparison and verification of real-time sensor data and reduced-order model simulation results. The verification criteria include: equipment operating safety thresholds, allowable deviations in process mechanisms, etc., leveraging the parallel processing capabilities of the FPGA to achieve microsecond-level dynamic parameter correction and anomaly interception.

[0102] This embodiment achieves physical isolation between the cloud-edge collaborative architecture and the computing tasks. Regardless of the fluctuations in the offline training load in the cloud, the jitter of the real-time control and inference cycle at the edge remains minimal, ensuring the absolute real-time performance of the underlying device. At the same time, the asynchronous collaboration between cloud recalculation and edge push is achieved through 5G URLLC and timestamp alignment mechanism, providing a solid hardware-level spatiotemporal decoupling foundation for the above-mentioned industrial dynamic equipment maintenance method.

[0103] This application also provides an electronic device. Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0104] Typically, an electronic device 900 includes one or more processors 901 and one or more memories 902.

[0105] Processor 901 may include one or more processing cores, such as a quad-core processor, a hexa-core processor, etc. Processor 901 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). Processor 901 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 901 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, processor 901 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0106] The memory 902 may include one or more computer-readable storage media, which may be non-transitory. The memory 902 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory 902 are used to store at least one computer program, which is executed by the processor 901 to implement the industrial equipment maintenance provided in the method embodiments of this application.

[0107] Those skilled in the art will understand that Figure 9 The structure shown does not constitute a limitation on the electronic device 900, and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0108] In addition, the device provided in the embodiments of this application may specifically be a chip, component or module. The chip may include a connected processor and a memory. The memory is used to store instructions. When the processor calls and executes the instructions, the chip can execute an industrial equipment maintenance method provided in the above embodiments.

[0109] This embodiment also provides a computer-readable storage medium storing computer program code. When the computer program code is run on a computer, the computer executes the above-described related method steps to implement the industrial dynamic equipment maintenance method provided in the above embodiment.

[0110] This embodiment also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned related steps to implement the industrial dynamic equipment maintenance method provided in the above embodiment.

[0111] In this embodiment, the device, computer-readable storage medium, computer program product, or chip are all used to execute the corresponding methods provided above. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods provided above, and will not be repeated here.

[0112] Through the above description of the embodiments, those skilled in the art will understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0113] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another apparatus, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0114] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for maintaining industrial moving equipment, characterized in that, The method includes: Based on the acquired multi-source prior and monitoring data, a multi-dimensional representation sub-model is determined. The multi-source prior and monitoring data includes multi-dimensional mechanism structure data and historical operation data of industrial dynamic equipment. The multi-dimensional representation sub-model includes a geometric model, a physical model, a data-driven model, and a knowledge model. Based on the preset multi-model bidirectional closed-loop fusion mechanism and multi-physics field collaborative solution and order reduction strategy, combined with the multi-dimensional characterized sub-model, a fused sub-model and a reduced-order solution model are generated. The preset multi-model bidirectional closed-loop fusion mechanism is used to characterize the cross-domain information interaction, mechanism constraints and parameter reverse correction relationship between the sub-models. The multi-physics field collaborative solution and order reduction strategy is used to characterize the differentiated collaborative calculation logic of physical fields with different coupling strengths and the low-dimensional order reduction mapping rule of the physical model. Based on the aforementioned fusion sub-model and reduced-order solution model, a multi-scale equipment maintenance model is constructed. The acquired real-time multi-source operation monitoring data is input into a multi-scale equipment maintenance model to obtain a maintenance strategy, which includes fault root cause localization and maintenance scheme optimization.

2. The industrial dynamic equipment maintenance method according to claim 1, characterized in that, Based on the acquired multi-source prior and monitoring data, a multi-dimensional representation sub-model is determined. The multi-source prior and monitoring data includes multi-dimensional mechanistic structure data and historical operational data of industrial dynamic equipment. The multi-dimensional representation sub-model includes a geometric model, a physical model, a data-driven model, and a knowledge model, including: Based on the equipment design drawings, material properties and assembly relationships in the acquired multi-source prior and monitoring data, the geometric model is determined. Based on the multi-physics coupling parameters and boundary conditions in the acquired multi-source prior and monitoring data, the physical model is determined. Based on historical operational data and fault records from the acquired multi-source prior and monitoring data, a data-driven model is determined. Based on the equipment design specifications, fault diagnosis rules, and experience maintenance data from the acquired multi-source prior and monitoring data, a knowledge model is determined. By constructing a unified semantic ontology, the heterogeneous parameters and outputs of the geometric model, the physical model, the data-driven model, and the knowledge model are uniformly mapped to standard entities and attributes, thus determining a multidimensional representation sub-model.

3. The industrial dynamic equipment maintenance method according to claim 1, characterized in that, The pre-defined multi-model bidirectional closed-loop fusion mechanism and multi-physics collaborative solution and order reduction strategy, combined with the multi-dimensional represented sub-models, generate a fused sub-model and a reduced-order solution model, including: Based on a pre-defined multi-model bidirectional closed-loop fusion mechanism, linkage relationships, correction relationships, and verification relationships are determined. The linkage relationship is used to represent the unique mesh mapping and parameterized linkage between the geometric model and the physical model. The correction relationship is used to represent the mechanistic prior constraints of the physical model on the data-driven model and the dynamic parameter correction of the physical model by the data-driven model. The verification relationship is used to represent the feature semantic transformation of the data-driven model to the knowledge model and the rule reasoning verification and label optimization of the knowledge model on the data-driven model. Based on the multiphysics collaborative solution and order reduction strategy, the solution logic and order reduction rules are determined. The solution logic represents the differentiated collaborative computation method, which adopts a fully coupled implicit solution for strongly coupled physics fields and a joint simulation explicit solution for weakly coupled physics fields, based on the differences in the coupling strength and time scale of the physics fields. The order reduction rules represent the order reduction mapping method, which extracts the dominant basis vectors based on the singular value decomposition of the snapshot matrix and maps the high-dimensional physical model to the low-dimensional modal coordinate space through Galerkin projection. Based on the aforementioned linkage relationship, correction relationship, and verification relationship, and combined with the sub-model of the multi-dimensional representation, bidirectional data flow and closed-loop collaborative computation processing are performed to generate a fusion sub-model; Based on the solution logic and order reduction rules, and combined with the sub-model of the multidimensional representation, partitioning solution and low-dimensional projection order reduction calculation are performed to generate a reduced-order solution model.

4. The industrial dynamic equipment maintenance method according to claim 1, characterized in that, The construction of a multi-scale equipment maintenance model based on the fusion sub-model and the reduced-order solution model includes: Based on the fusion sub-model and the reduced-order solution model, a simulation embedding mechanism, a virtual-real calibration mechanism, and a result fusion mechanism are determined. The simulation embedding mechanism is used to indicate that the reduced-order solution model is embedded into the physical model layer of the fusion sub-model. The virtual-real calibration mechanism is used to indicate that the virtual and real data timing is synchronized according to the timestamp alignment and logical clock, and the parameters of the fusion sub-model and the reduced-order solution model are corrected by using the virtual-real deviation through the gradient descent algorithm. The result fusion mechanism is used to indicate that the multidimensional output of the fusion sub-model is fused using a weighted Bayesian network algorithm to calculate the probability of device health status. Based on the simulation embedding mechanism, the virtual-real calibration mechanism, and the result fusion mechanism, the fusion sub-model and the reduced-order solution model are integrated and closed-loop connected to construct a multi-scale equipment maintenance model.

5. The industrial dynamic equipment maintenance method according to claim 1, characterized in that, The acquired real-time multi-source operation monitoring data is input into a multi-scale equipment maintenance model to obtain a maintenance strategy. This maintenance strategy includes fault root cause localization and maintenance scheme optimization, including: The acquired real-time multi-source operation monitoring data is input into the multi-scale equipment maintenance model, and real-time multiphysics simulation is performed through the reduced-order solution model to obtain the real-time simulation state of the equipment. The real-time simulation state of the equipment includes the stress distribution, temperature field and dynamic response of key parts of the equipment. The real-time multi-source operation monitoring data and the real-time simulation state of the equipment are compared with a preset threshold and a normal mode baseline to identify abnormal states. The preset threshold is used to represent the limit boundary and safety alarm limit of the equipment operation parameters, and the normal mode baseline is used to represent the characteristic distribution and state evolution benchmark of the equipment during healthy operation under multiple working conditions. Based on the abnormal state and the data-driven model, a fault warning is generated; The abnormal state and the fault warning are input into the knowledge model, and knowledge reasoning is performed based on association rules to determine the root cause of the fault. The association rules are used to represent the mapping relationship between fault characteristics, fault modes and fault causes. Based on the data-driven model, updated remaining lifetime prediction results are output; Based on the updated remaining lifetime prediction results and the root cause location of the failure, a maintenance plan is formulated. The effectiveness of the maintenance scheme is verified through virtual debugging, resulting in a maintenance strategy for industrial equipment.

6. The industrial dynamic equipment maintenance method according to claim 2, characterized in that, The process involves constructing a unified semantic ontology to map the heterogeneous parameters and outputs of the geometric model, the physical model, the data-driven model, and the knowledge model into standard entities and attributes, thereby determining a multidimensional representation sub-model, including: Define a core ontology class, which includes a device class, a component class, a parameter class, a fault class, and a maintenance class; Define the object attributes and data attributes between the core ontology classes. The object attributes include the relationships of inclusion, cause, influence, and correspondence. The data attributes include the parameter values, sampling time, unit, and the fault level and occurrence time. Based on the core ontology class, the object attributes, and the data attributes, a unified semantic ontology for industrial moving equipment is constructed. An ontology inference engine is used to perform cross-layer semantic matching and consistency verification on the unified semantic ontology, transforming heterogeneous parameters and outputs of different dimensions into standard entities, which are used to represent the unified semantic mapping results of the cross-layer model. Based on the standard entity, a multidimensional representation sub-model is determined.

7. The industrial dynamic equipment maintenance method according to claim 3, characterized in that, The physical model imposes prior mechanistic constraints on the data-driven model, including: Using the physical model, virtual samples under different operating conditions and fault levels are generated; The virtual samples are added to the training dataset of the data-driven model; A mechanism constraint term is embedded in the loss function of the data-driven model. The mechanism constraint term is used to restrict the prediction results to meet the wear law and fatigue life curve of the equipment. The data-driven model's parameter correction of the physical model includes: An autoencoder based on the comparison of virtual and real data is used to construct a parameter deviation identification model; The simulation data of the physical model and the measured data of the physical device are input into the parameter deviation identification model, and the key parameter deviation values ​​of the physical model are output. The gradient descent algorithm is used to correct the parameters of the physical model based on the deviation values ​​of the key parameters.

8. The industrial dynamic equipment maintenance method according to claim 3, characterized in that, The process of extracting dominant basis vectors based on the singular value decomposition of the snapshot matrix and mapping the high-dimensional physical model to the low-dimensional modal coordinate space through Galerkin projection includes: Based on the orthogonal experimental design method, sampling is performed in the parameter space and boundary condition space of the physical model to obtain multiple sets of working condition parameters; Under the aforementioned multiple sets of operating parameters, a fully coupled simulation was run to collect time-series response data of each physical field and construct a snapshot matrix. The snapshot matrix is ​​preprocessed and singular value decomposed, and the first left singular vector is extracted as the dominant basis vector according to the preset energy criterion. The original response vector is projected onto the dominant basis vector space to obtain the modal coordinate vector, and the original high-dimensional partial differential control equations are reduced to a set of ordinary differential equations using the Galerkin projection method to generate a reduced-order solution model.

9. The industrial dynamic equipment maintenance method according to claim 4, characterized in that, The method of using a weighted Bayesian network algorithm to fuse the multidimensional outputs of the fusion sub-model to calculate the probability of device health status includes: A Bayesian network structure is constructed with the equipment health status as the root node, geometric deviation, physical simulation deviation, fault probability and fault confidence as intermediate nodes, and the multidimensional output of the fusion sub-model as leaf nodes. Based on historical operation data and maintenance records, the conditional probability table of the Bayesian network is learned using the maximum likelihood estimation method, resulting in a parameterized Bayesian network model. Based on the historical prediction accuracy of the sub-model of the multidimensional representation, differentiated weights are assigned to the nodes of the parameterized Bayesian network model to obtain a weighted Bayesian network model. The real-time output of the fusion sub-model is input into the weighted Bayesian network model, and the probability of the industrial equipment being in each state is obtained through joint probabilistic inference. The state probabilities include the probability of a healthy state, the probability of a sub-healthy state, and the probability of a fault state. When the probability of the fault state exceeds a preset threshold, a fault warning is triggered.

10. The industrial dynamic equipment maintenance method according to claim 5, characterized in that, The process of verifying the effectiveness of the maintenance plan through virtual debugging to obtain a maintenance strategy for industrial equipment includes: The maintenance operation instructions and / or parameter adjustment instructions in the maintenance scheme are mapped to the multi-scale equipment maintenance model. The reduced-order solution model in the multi-scale equipment maintenance model performs the simulation deduction after maintenance to obtain the deduced state response. When the simulation state response meets the normal mode baseline and does not trigger the preset threshold, the maintenance scheme is determined to be valid and output. When the simulated state response fails to meet the normal mode baseline or triggers the preset threshold, the maintenance scheme is adjusted and the simulation is repeated until the validity judgment condition is met, and the final maintenance strategy is output.

Images