Digital twin modeling method, device and equipment of multi-stage centrifugal compressor and medium

Abstract

This application discloses a digital twin modeling method, apparatus, device, and medium for a multi-stage centrifugal compressor, relating to the fields of compressor condition monitoring and digital twin technology. By constructing cross-scale data mapping relationships and incorporating these relationships as constraints into the training process of a lightweight simulation model, it fundamentally achieves a dual optimization of computational accuracy and rendering efficiency. The method includes: establishing data mapping relationships based on a multi-condition data matrix obtained from the simulation of the multi-stage centrifugal compressor component; using the multi-condition data matrix as training samples, training the simulation model by incorporating the data mapping relationships as constraints into the interpolation fitting algorithm; deploying the simulation model to the backend server of the digital twin system to output a target strain field result dataset; and using the target strain field result dataset to drive the rendering of a low-precision mesh model of the multi-stage centrifugal compressor component, generating a digital twin model of the multi-stage centrifugal compressor component.

Description

Technical Field

[0001] This application relates to the field of compressor condition monitoring and digital twin technology, and in particular to a digital twin modeling method, device, equipment and medium for a multi-stage centrifugal compressor. Background Technology

[0002] Multistage centrifugal compressors are key energy and power equipment widely used in modern industry, and their continuous and stable operation is directly related to the continuity and safety of industrial production. Traditional monitoring methods for multistage centrifugal compressors suffer from defects such as data fragmentation, low digitization, and delayed maintenance decisions, which can no longer meet the needs of modern industry for refined equipment management. Digital twin technology, as a new generation of digital tools, provides a new approach to the intelligent management of multistage centrifugal compressors by constructing a real-time virtual-real fusion mapping between physical entities and digital models, enabling dynamic two-way interaction and intelligent feedback.

[0003] In existing digital twin applications of multi-stage centrifugal compressors, there are two main modeling approaches: one is to construct a fine mesh based on a high-precision finite element model, perform offline multi-condition calculations, and then display the modeling and simulation results in a static form. While this approach can guarantee a certain level of simulation accuracy, the large number of nodes in the fine mesh results in lengthy simulation calculations, failing to meet the real-time response requirements of digital twin systems. The other approach, to improve display efficiency, maps the modeling and simulation results to a simplified mesh for rendering. Although this improves display efficiency, the lack of a stable and universal high- and low-precision mesh data mapping mechanism easily leads to distortion of the modeling and simulation results, making it impossible to accurately reproduce the stress-strain field distribution of the compressor's key structural responses under complex operating conditions. Furthermore, most systems adopt a front-end and back-end coupled development model, with back-end physical calculations and front-end rendering mutually constraining each other, further exacerbating system response latency and failing to meet the real-time visualization requirements of digital twin models under complex operating conditions. In summary, existing modeling technologies cannot simultaneously meet the high-precision simulation and real-time rendering requirements of multi-stage centrifugal compressor digital twin models, and the system response latency caused by architectural coupling makes it difficult to achieve real-time and accurate mapping of equipment operating status. Summary of the Invention

[0004] In view of this, this application provides a digital twin modeling method, device, equipment and medium for multi-stage centrifugal compressors. The main purpose is to solve the problem that existing modeling techniques cannot meet the high-precision simulation and real-time rendering requirements of digital twin models of multi-stage centrifugal compressors, and that the system response delay caused by architectural coupling makes it difficult to achieve real-time and accurate mapping of equipment operating status.

[0005] The first aspect provides a digital twin modeling method for multi-stage centrifugal compressors, which includes: Based on the multi-condition data matrix obtained from the simulation of multi-stage centrifugal compressor components, a cross-scale data mapping relationship is established between high-precision grid models and low-precision grid models. Using the multi-condition data matrix as training samples, the simulation model is trained by incorporating the data mapping relationship into the interpolation fitting algorithm as a constraint condition, so as to establish a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset. The trained simulation model is deployed to the backend server of the digital twin system, so that the backend server can call the simulation model to predict the strain field and output the target strain field result dataset. The target strain field result dataset is used to drive the rendering of a low-precision mesh model of a multi-stage centrifugal compressor assembly, generating a digital twin model of the multi-stage centrifugal compressor assembly.

[0006] Furthermore, before establishing the cross-scale data mapping relationship between the high-precision mesh model and the low-precision mesh model based on the multi-condition data matrix obtained from the simulation of the multi-stage centrifugal compressor components, the method further includes: Based on the original geometric model of the multi-stage centrifugal compressor components, a high-precision mesh model for finite element calculation and a low-precision mesh model for visualization are pre-built. The high-precision mesh model was used to perform simulation calculations under different operating conditions to obtain a finite element simulation dataset of the multi-stage centrifugal compressor assembly. Based on the finite element simulation dataset of a multi-stage centrifugal compressor component, a multi-condition data matrix is ​​constructed. The multi-condition data matrix contains strain field result datasets corresponding to different condition parameters. The strain field result datasets correspond to the grid node coordinates through node numbers.

[0007] Furthermore, the finite element simulation data initially uses a cylindrical coordinate system to record node positions and deformation information. The finite element simulation dataset based on the multi-stage centrifugal compressor assembly constructs a multi-condition data matrix, including: The cylindrical coordinate system corresponding to the finite element simulation dataset is converted to the position coordinates in the Cartesian coordinate system to obtain the converted intermediate dataset. The intermediate dataset is deduplicated to remove duplicate coordinate records and retain the first occurrence of node data, resulting in a unique set of node coordinates. Establish a continuous index starting from a preset value for the set of unique node coordinates, and construct a mapping table from coordinates to indexes; The intermediate dataset is left-joined with the mapping table to generate a multi-condition data matrix containing standard indexes.

[0008] Furthermore, the establishment of a cross-scale data mapping relationship between the high-precision mesh model and the low-precision mesh model based on the multi-condition data matrix obtained from the simulation of the multi-stage centrifugal compressor components includes: Based on the multi-condition data matrix obtained from simulation, the node coordinates and strain field result dataset of the high-precision mesh model and the node coordinates of the low-precision mesh model are extracted respectively. The strain field result dataset of the high-precision grid model is transferred to the node coordinates of the low-precision grid model through a spatial mapping algorithm of node coordinates, thereby obtaining the cross-scale data mapping relationship between the high-precision grid model and the low-precision grid model.

[0009] Furthermore, the step of transferring the strain field result dataset of the high-precision mesh model to the node coordinates of the low-precision mesh model through a spatial mapping algorithm of node coordinates to obtain the cross-scale data mapping relationship between the high-precision mesh model and the low-precision mesh model includes: For each node in the low-precision mesh model, calculate the distance between that node and all nodes in the high-precision mesh model; Neighboring nodes within a preset neighborhood range are selected as target nodes, and the strain field result datasets of the target nodes are fused using distance weighting to obtain the fused strain field result dataset. The fused strain field result dataset is assigned to the corresponding nodes of the low-precision grid model to obtain the cross-scale data mapping relationship between the high-precision grid model and the low-precision grid model.

[0010] Furthermore, the step of using the multi-condition data matrix as training samples and incorporating the data mapping relationship into the interpolation fitting algorithm as a constraint to train the simulation model, in order to establish a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset, includes: Using the multi-condition data matrix as training samples, a radial basis function of a preset type is selected as the basis function for the interpolation fitting algorithm; Construct a joint objective function, which includes a data fitting term for measuring the deviation between the predicted and the true values, and cross-scale constraints based on the data mapping relationship; Using the minimization of the joint objective function as the optimization objective, the weight coefficients and polynomial coefficients of the radial basis functions are obtained by solving the problem. The obtained weight coefficients and polynomial coefficients are substituted into the radial basis function to complete the training of the simulation model, so as to establish a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset. Accordingly, after using the multi-condition data matrix as training samples and incorporating the data mapping relationship into the interpolation fitting algorithm as a constraint to train the simulation model, thereby establishing a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset, the method further includes: Based on test samples that were not used in the simulation model training, the weight coefficients and polynomial coefficients of the radial basis function are re-solved through cross-validation to retrain the simulation model.

[0011] Furthermore, the backend server adopts a microservice architecture, including a distributed storage module and a lightweight computing module; The distributed storage module is used to store the real-time operating condition database and the prediction result database; The lightweight computing module is used to respond to front-end interaction requests, read the node coordinates and operating parameters of the multi-stage centrifugal compressor component from the real-time operating condition database, call the simulation model to calculate the target strain field result dataset of the multi-stage centrifugal compressor component, and write the target strain field result dataset into the prediction result database for front-end visualization. The process of using the target strain field result dataset to drive the rendering of a low-precision mesh model of a multi-stage centrifugal compressor assembly to generate a digital twin model of the multi-stage centrifugal compressor assembly includes: The target strain field result data is standardized and mapped based on a preset color spectrum to obtain vertex color values; The vertex colors are used to render the low-precision mesh model of the multi-stage centrifugal compressor assembly, generating a digital twin model of the multi-stage centrifugal compressor assembly.

[0012] Secondly, a digital twin modeling apparatus for a multi-stage centrifugal compressor is provided, the apparatus comprising: A cell is established to create a cross-scale data mapping relationship between a high-precision mesh model and a low-precision mesh model based on the multi-condition data matrix obtained from the simulation of a multi-stage centrifugal compressor component. The training unit is used to train the simulation model by incorporating the data mapping relationship into the interpolation fitting algorithm as a constraint condition, using the multi-condition data matrix as training samples, so as to establish a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset. The deployment unit is used to deploy the trained simulation model to the backend server of the digital twin system, so that the simulation model can be called through the backend server to predict the strain field and output the target strain field result dataset. The rendering unit is used to drive the low-precision mesh model of the multi-stage centrifugal compressor component to render using the target strain field result dataset, thereby generating a digital twin model of the multi-stage centrifugal compressor component.

[0013] Furthermore, the device also includes: The first building unit is used to pre-build a high-precision mesh model for finite element calculation and a low-precision mesh model for visualization based on the original geometric model of the multi-stage centrifugal compressor component before establishing the cross-scale data mapping relationship between the high-precision mesh model and the low-precision mesh model based on the multi-condition data matrix obtained from the simulation of the multi-stage centrifugal compressor component. The simulation unit is used to perform simulation calculations under different operating conditions using the high-precision mesh model to obtain the finite element simulation dataset of the multi-stage centrifugal compressor assembly. The second building unit is used to construct a multi-condition data matrix based on the finite element simulation dataset of a multi-stage centrifugal compressor component. The multi-condition data matrix contains strain field result datasets corresponding to different operating condition parameters. The strain field result datasets correspond to the grid node coordinates through node numbers.

[0014] Furthermore, the finite element simulation data initially uses a cylindrical coordinate system to record node positions and deformation information. The second building unit is specifically used for: The cylindrical coordinate system corresponding to the finite element simulation dataset is converted to the position coordinates in the Cartesian coordinate system to obtain the converted intermediate dataset. The intermediate dataset is deduplicated to remove duplicate coordinate records and retain the first occurrence of node data, resulting in a unique set of node coordinates. Establish a continuous index starting from a preset value for the set of unique node coordinates, and construct a mapping table from coordinates to indexes; The intermediate dataset is left-joined with the mapping table to generate a multi-condition data matrix containing standard indexes.

[0015] Furthermore, the establishment unit includes: The extraction module is used to extract the node coordinates and strain field result dataset of the high-precision mesh model and the node coordinates of the low-precision mesh model based on the multi-condition data matrix obtained from the simulation. The mapping module is used to transfer the strain field result dataset of the high-precision grid model to the node coordinates of the low-precision grid model through a spatial mapping algorithm of node coordinates, so as to obtain the cross-scale data mapping relationship between the high-precision grid model and the low-precision grid model.

[0016] Furthermore, the mapping module is specifically used for: For each node in the low-precision mesh model, calculate the distance between that node and all nodes in the high-precision mesh model; Neighboring nodes within a preset neighborhood range are selected as target nodes, and the strain field result datasets of the target nodes are fused using distance weighting to obtain the fused strain field result dataset. The fused strain field result dataset is assigned to the corresponding nodes of the low-precision grid model to obtain the cross-scale data mapping relationship between the high-precision grid model and the low-precision grid model.

[0017] Furthermore, the training unit is specifically used for: Using the multi-condition data matrix as training samples, a radial basis function of a preset type is selected as the basis function for the interpolation fitting algorithm; Construct a joint objective function, which includes a data fitting term for measuring the deviation between the predicted and the true values, and cross-scale constraints based on the data mapping relationship; Using the minimization of the joint objective function as the optimization objective, the weight coefficients and polynomial coefficients of the radial basis functions are obtained by solving the problem. The obtained weight coefficients and polynomial coefficients are substituted into the radial basis function to complete the training of the simulation model, so as to establish a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset. The training unit is further used for: After training the simulation model using the multi-condition data matrix as training samples and incorporating the data mapping relationship into the interpolation fitting algorithm as a constraint to establish a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset, the simulation model is retrained based on test samples that did not participate in the simulation model training by resolving the weight coefficients and polynomial coefficients of the radial basis function through cross-validation.

[0018] Furthermore, the backend server adopts a microservice architecture, including a distributed storage module and a lightweight computing module; The distributed storage module is used to store the real-time operating condition database and the prediction result database; The lightweight computing module is used to respond to front-end interaction requests, read the node coordinates and operating parameters of the multi-stage centrifugal compressor component from the real-time operating condition database, call the simulation model to calculate the target strain field result dataset of the multi-stage centrifugal compressor component, and write the target strain field result dataset into the prediction result database for front-end visualization. The rendering unit is specifically used for: The target strain field result data is standardized and mapped based on a preset color spectrum to obtain vertex color values; The vertex colors are used to render the low-precision mesh model of the multi-stage centrifugal compressor assembly, generating a digital twin model of the multi-stage centrifugal compressor assembly.

[0019] Thirdly, a digital twin modeling device for a multi-stage centrifugal compressor is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor. When the processor executes the program, it implements the aforementioned digital twin modeling method for the multi-stage centrifugal compressor.

[0020] Fourthly, a storage medium is provided on which a computer program is stored, which, when executed by a processor, implements the aforementioned digital twin modeling method for a multi-stage centrifugal compressor.

[0021] By employing the aforementioned technical solution, this application provides a digital twin modeling method, apparatus, device, and medium for multi-stage centrifugal compressors. Compared to current methods that achieve digital twin modeling of multi-stage centrifugal compressors through high-precision finite element modeling simulation results or simplified mesh rendering, this application establishes a cross-scale data mapping relationship between high-precision and low-precision mesh models based on a multi-condition data matrix obtained from multi-stage centrifugal compressor component simulation. Using the multi-condition data matrix as training samples, the simulation model is trained by incorporating the data mapping relationship as a constraint in the interpolation fitting algorithm to establish a nonlinear mapping relationship between multi-condition parameters and strain field result datasets. The trained simulation model is deployed to the backend server of the digital twin system, allowing the backend server to call the simulation model for strain field prediction and output a target strain field result dataset. The target strain field result dataset is then used to drive the rendering of the low-precision mesh model of the multi-stage centrifugal compressor component, generating a digital twin model of the multi-stage centrifugal compressor component. The entire process, by constructing a cross-scale data mapping relationship and incorporating this relationship as a constraint into the training process of the lightweight simulation model, fundamentally achieves a dual optimization of computational accuracy and rendering efficiency. On the one hand, this method utilizes a high-precision mesh model as the data source, fully preserving the high computational accuracy of the finite element model and ensuring the accuracy of the description of the complex physical field of a multi-stage centrifugal compressor. On the other hand, by mapping the high-precision simulation results to a low-precision mesh model, it fully leverages the rendering advantages of the low-precision mesh model, ensuring high frame rate rendering of the front-end interface. More importantly, the simulation model based on offline training decouples the simulation calculation from the 3D rendering architecture, eliminating the need for heavy iterative calculations during system runtime. This effectively balances the dual requirements of high-precision simulation and real-time rendering for the digital twin model of the multi-stage centrifugal compressor, achieving real-time and accurate mapping of the equipment's operating status.

[0022] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating a digital twin modeling method for a multi-stage centrifugal compressor in one embodiment of this application; Figure 2 yes Figure 1 A flowchart illustrating a specific implementation method prior to step 101; Figure 3 yes Figure 2 A flowchart illustrating a specific implementation method for step 203; Figure 4 This is a schematic diagram of the coordinate system transformation process in one embodiment of this application; Figure 5 This is a schematic diagram of the cleaning and indexing process of an intermediate dataset in one embodiment of this application; Figure 6 yes Figure 1 A flowchart illustrating a specific implementation method of step 101; Figure 7 This is a schematic diagram of the process of establishing cross-scale data mapping relationship in one embodiment of this application; Figure 8 yes Figure 1 A flowchart illustrating a specific implementation method for step 102; Figure 9 This is a schematic diagram of the microservice architecture of a multi-stage centrifugal compressor digital twin system in one embodiment of this application; Figure 10 yes Figure 1 A flowchart illustrating a specific implementation method for step 104; Figure 11 This is a schematic diagram of the strain rendering process using a trained simulation model on the front end of a webpage in one embodiment of this application; Figure 12 This is a schematic diagram of the strain rendering process in one embodiment of this application; Figure 13 This is a schematic diagram of the simulation monitoring interface of a digital twin model of a multi-stage centrifugal compressor impeller in one embodiment of this application; Figure 14This is a schematic diagram of the structure of a digital twin modeling device for a multi-stage centrifugal compressor in one embodiment of this application; Figure 15 This is a schematic diagram of the device structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0024] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0025] In existing digital twin applications of multi-stage centrifugal compressors, there are two main modeling approaches: one is to construct a fine mesh based on a high-precision finite element model, perform offline multi-condition calculations, and then display the modeling and simulation results in a static form. While this approach can guarantee a certain level of simulation accuracy, the large number of nodes in the fine mesh results in lengthy simulation calculations, failing to meet the real-time response requirements of digital twin systems. The other approach, to improve display efficiency, maps the modeling and simulation results to a simplified mesh for rendering. Although this improves display efficiency, the lack of a stable and universal high- and low-precision mesh data mapping mechanism easily leads to distortion of the modeling and simulation results, making it impossible to accurately reproduce the stress-strain field distribution of the compressor's key structural responses under complex operating conditions. Furthermore, most systems adopt a front-end and back-end coupled development model, with back-end physical calculations and front-end rendering mutually constraining each other, further exacerbating system response latency and failing to meet the real-time visualization requirements of digital twin models under complex operating conditions. In summary, existing modeling technologies cannot simultaneously meet the high-precision simulation and real-time rendering requirements of multi-stage centrifugal compressor digital twin models, and the system response latency caused by architectural coupling makes it difficult to achieve real-time and accurate mapping of equipment operating status.

[0026] To address this issue, this embodiment provides a digital twin modeling method for multi-stage centrifugal compressors, such as... Figure 1 As shown, it includes the following steps: 101. Based on the multi-condition data matrix obtained from the simulation of multi-stage centrifugal compressor components, establish a cross-scale data mapping relationship between high-precision grid models and low-precision grid models.

[0027] Typically, multistage centrifugal compressors have extremely complex internal structures, including high-speed rotating impellers, diffusers, bends, and other flow channel components. Traditional physical sensors struggle to achieve real-time monitoring of internal stress and strain across the entire flow field without disturbing the flow. High-precision finite element simulation allows for the acquisition of physical field distribution data at any location within the equipment in a virtual environment, compensating for the limitations of traditional physical measurement methods.

[0028] Specifically, in the finite element simulation process, a series of discrete operating parameters can be set for various operating states that a multi-stage centrifugal compressor may encounter throughout its entire life cycle, such as start-up, shutdown, and rated load. A high-precision mesh model is then used to independently solve for each operating parameter using finite element methods, resulting in a multi-condition data matrix covering the entire operating range of the equipment. Specifically, the multi-condition data matrix is ​​a high-dimensional structured data set obtained through standardization. Mathematically, this matrix is ​​typically represented as a two-dimensional array, where the row vectors represent different operating condition samples, with each row corresponding to a specific operating condition parameter vector; the column vectors represent the strain field values ​​of all nodes in the high-precision mesh model under that operating condition, with each column corresponding to the strain value of a specific node in the model.

[0029] In other words, the multi-condition data matrix integrates the strain field results data corresponding to different operating conditions in a standardized form, and the strain field results data correspond to the node coordinates of the high-precision grid model through node numbering, which can completely and accurately reflect the physical field distribution characteristics of the multi-stage centrifugal compressor components under different operating conditions.

[0030] In this embodiment, both the high-precision and low-precision mesh models are pre-constructed based on the original geometric model of the multi-stage centrifugal compressor. For the high-precision finite element calculation requirements, a high-precision mesh model with high node density and fine topology can be constructed, with all node 3D spatial coordinates and node indices pre-calibrated and stored. For the real-time visualization rendering requirements, a low-precision mesh model with low node density and simplified topology can be constructed, with its node 3D spatial coordinates and node numbers also pre-determined and stored independently. Therefore, the node coordinate information of the high-precision and low-precision mesh models is fixed and can be directly retrieved from known data. When establishing cross-scale data mapping relationships between the high-precision and low-precision mesh models, there is no need to construct the model again; only the node coordinate information of the high-precision and low-precision mesh models needs to be directly extracted from the model storage file. Based on this, node spatial correspondence matching is performed, significantly improving the efficiency of establishing cross-scale data mapping relationships while ensuring that the high-precision and low-precision mesh models are consistent with the original geometric model of the multi-stage centrifugal compressor in terms of spatial topology.

[0031] Specifically, the establishment of cross-scale data mapping relationships begins by extracting the node coordinates and strain field results data of the high-precision mesh model from the multi-condition data matrix, while simultaneously retrieving the node coordinates of a pre-constructed low-precision mesh model. Then, using the node coordinates and strain field results data of the high-precision mesh model, along with the node coordinates of the low-precision mesh model, as input, a spatial mapping algorithm is employed to match each node of the low-precision mesh model with its spatial neighboring nodes in the high-precision mesh model, assigning weights to each neighboring node based on spatial distance. Next, combining the strain field results data from the multi-condition data matrix, an interpolation algorithm is used to weightedly fuse the strain field results data of the neighboring nodes, resulting in strain mapping data for all nodes in the low-precision mesh model. Integrating this strain mapping data yields the cross-scale data mapping relationship. Through this data mapping relationship, the accurate strain field data of the high-precision mesh model under different conditions can be transformed into a data format that the low-precision mesh model can recognize and use, preserving the core data value of the high-precision simulation while adapting to the rendering requirements of the low-precision model.

[0032] 102. Using the multi-condition data matrix as training samples, the simulation model is trained by incorporating the data mapping relationship into the interpolation fitting algorithm as a constraint condition, so as to establish a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset.

[0033] In this embodiment, the multi-condition data matrix typically covers various operating condition samples throughout the entire lifecycle of a multi-stage centrifugal compressor, and the data has been precisely bound through unique node indexes. It can serve as a high-quality sample set for training simulation models, allowing for the batch extraction of training data directly from the multi-condition data matrix. Specific training samples may include, but are not limited to: multi-condition parameters corresponding to each operating condition sample, node coordinates of the high-precision mesh model, and strain field results data corresponding to each node of the high-precision mesh model under each set of operating condition parameters.

[0034] Specifically, the multi-condition parameters in the training samples can be used as input features for model training, and the strain field result data corresponding to the nodes in the training samples can be used as output labels for model training to construct the training architecture of the simulation model. Based on the training framework, an interpolation fitting algorithm adapted to strain prediction of multi-stage centrifugal compressors is selected as the computational logic of the simulation model to start the initial training of the model. The training samples are input into the interpolation fitting algorithm, and the simulation model learns the nonlinear mapping relationship between the multi-condition parameters and the strain field result data through the interpolation fitting algorithm. In the calculation process of the interpolation fitting algorithm, the cross-scale data mapping relationship is incorporated as a constraint condition so that the simulation model follows the node mapping rules between the high-precision grid model and the low-precision grid model when learning the nonlinear mapping relationship. Furthermore, the error between the predicted strain field result data output by the interpolation fitting algorithm and the strain field result data in the training samples is used as the optimization target, and the simulation model parameters are iteratively adjusted repeatedly until the fitting error is reduced to a preset threshold and tends to stabilize. At this time, the interpolation fitting algorithm has completed sufficient learning of the training samples, indicating that the simulation model has completed training.

[0035] For example, compressor speed is selected as the main input parameter of the simulation model, and the strain values ​​of key nodes of the compressor impeller are selected as the output parameters of the model. First, operating condition samples are selected from a multi-condition data matrix, with the speed covering the commonly used operating range of the equipment. Simultaneously, the strain values ​​of key impeller nodes obtained from high-precision simulations at each speed are extracted, forming multiple sets of speed-strain value operating condition samples. These operating condition samples are then imported into an interpolation fitting algorithm. Using the error between the fitted value and the true value as the optimization objective, the parameters of the interpolation fitting algorithm are iteratively adjusted to complete the construction of the nonlinear mapping relationship between speed and node strain response.

[0036] 103. Deploy the trained simulation model to the backend server of the digital twin system, so that the backend server can call the simulation model to predict the strain field and output the target strain field result dataset.

[0037] Understandably, in order to provide a stable operating environment for the simulation model and to achieve efficient model access and rapid output of strain field prediction results, the trained simulation model can be deployed to the backend server of the digital twin system by relying on the hardware computing resources and system scheduling capabilities of the backend server.

[0038] Correspondingly, during the actual operation of the digital twin system, the backend server can receive the actual operating condition parameters of the multi-stage centrifugal compressor or the target operating condition parameters set manually in real time. Based on the preset calling logic, it can quickly start the deployed simulation model and carry out strain field prediction calculations. The simulation model does not need to perform cumbersome online finite element iterative solutions. It can directly process the target operating condition parameters based on the established nonlinear mapping relationship and output a target strain field result dataset that accurately matches the operating condition parameters. This target strain field dataset is standardized strain field data adapted to the rendering requirements of low-precision mesh models. It can be directly transmitted to the front end of the digital twin system through the backend server, providing core data support for the real-time visualization mapping of the operating status of the multi-stage centrifugal compressor and the analysis of operating trends, realizing the efficient conversion from operating condition parameter input to strain field result output.

[0039] 104. Using the target strain field result dataset, the low-precision mesh model of the multi-stage centrifugal compressor assembly is rendered to generate a digital twin model of the multi-stage centrifugal compressor assembly.

[0040] Understandably, the target strain field result dataset is standardized data accurately predicted by the simulation model and transformed through cross-scale data mapping relationships. Its data format and node index are completely matched with the pre-built low-precision mesh model, covering the strain values ​​of all nodes of the low-precision mesh model under the target working condition, and can be directly recognized and called by the low-precision mesh model. Among them, the low-precision mesh model is pre-built based on the simplified original geometric model of the multi-stage centrifugal compressor components, retaining the geometric topology of core components such as compressor impellers, shafts, and diffusers, and only simplifying the node density to adapt to the real-time rendering requirements. Its node numbers and spatial coordinates correspond completely with the low-precision nodes in the cross-scale data mapping relationship, providing a reliable model foundation for data-driven rendering.

[0041] Specifically, in the low-precision mesh model-driven rendering process, the target strain field result dataset can be transmitted from the backend server to the frontend rendering module through the frontend-backend data interaction interface of the digital twin system. After receiving the data, the frontend rendering module accurately binds the target strain field result data with the corresponding nodes of the low-precision mesh model according to preset rendering rules. That is, based on the node index in the target strain field result dataset, the strain value of each node is precisely matched to the corresponding node in the low-precision mesh model, realizing a one-to-one correspondence between strain data and virtual model nodes. Subsequently, based on the bound strain field result data and combined with preset visual mapping rules, the rendering module renders the low-precision mesh model to intuitively present the strain distribution state of various parts of the compressor component.

[0042] Accordingly, through the above rendering process, the low-precision mesh model not only presents the complete geometric shape of the multi-stage centrifugal compressor assembly, but also accurately maps the strain field distribution under the target operating condition through visualization, thereby generating a complete digital twin model of the multi-stage centrifugal compressor assembly. This digital twin model not only restores the geometric features of the physical equipment, but also reflects the operating strain state of the equipment under the target operating condition in real time.

[0043] The digital twin modeling method for multi-stage centrifugal compressors provided in this application, compared with current methods that achieve digital twin modeling of multi-stage centrifugal compressors through high-precision finite element modeling simulation results or simplified mesh rendering, establishes a cross-scale data mapping relationship between high-precision and low-precision mesh models based on a multi-condition data matrix obtained from the simulation of multi-stage centrifugal compressor components. Using the multi-condition data matrix as training samples, the simulation model is trained by incorporating the data mapping relationship as a constraint into the interpolation fitting algorithm to establish a nonlinear mapping relationship between multi-condition parameters and strain field result datasets. The trained simulation model is deployed to the backend server of the digital twin system, allowing the backend server to call the simulation model for strain field prediction and output a target strain field result dataset. The target strain field result dataset is then used to drive the rendering of the low-precision mesh model of the multi-stage centrifugal compressor components, generating a digital twin model of the multi-stage centrifugal compressor components. The entire process, by constructing a cross-scale data mapping relationship and incorporating this relationship as a constraint into the training process of the lightweight simulation model, fundamentally achieves a dual optimization of computational accuracy and rendering efficiency. On the one hand, this method utilizes a high-precision mesh model as the data source, fully preserving the high computational accuracy of the finite element model and ensuring the accuracy of the description of the complex physical field of a multi-stage centrifugal compressor. On the other hand, by mapping the high-precision simulation results to a low-precision mesh model, it fully leverages the rendering advantages of the low-precision mesh model, ensuring high frame rate rendering of the front-end interface. More importantly, the simulation model based on offline training decouples the simulation calculation from the 3D rendering architecture, eliminating the need for heavy iterative calculations during system runtime. This effectively balances the dual requirements of high-precision simulation and real-time rendering for the digital twin model of the multi-stage centrifugal compressor, achieving real-time and accurate mapping of the equipment's operating status.

[0044] In practical applications, different operating conditions can cause significant differences in the strain distribution of multi-stage centrifugal compressor components. Simulating the operating conditions of multi-stage centrifugal compressor components can fully cover the actual operating state of the compressor. Furthermore, such as... Figure 2 As shown, prior to step 101, the method further includes the following steps: 201. Based on the original geometric model of the multi-stage centrifugal compressor assembly, a high-precision mesh model for finite element calculation and a low-precision mesh model for visualization are pre-built.

[0045] 202. Using the high-precision mesh model, simulation calculations are performed under different operating conditions to obtain the finite element simulation dataset of the multi-stage centrifugal compressor assembly.

[0046] 203. Based on the finite element simulation dataset of multi-stage centrifugal compressor components, construct a multi-condition data matrix.

[0047] In this embodiment, the model construction of the multi-stage centrifugal compressor assembly is preferably completed offline. Specifically, based on the original geometric model of the multi-stage centrifugal compressor assembly, a high-precision mesh model and a low-precision mesh model can be constructed respectively. The high-precision mesh model uses the original geometric model of the multi-stage centrifugal compressor assembly as a precise benchmark, employs a high-density mesh generation method, refines the geometric details and topology of the compressor assembly, and possesses sufficient node quantity and unit density. The low-precision mesh model is constructed based on the same original geometric model and is consistent with the high-precision mesh model. The difference lies in the reasonable simplification of the mesh structure in the low-precision mesh model, sparsening the number of nodes and simplifying the unit structure, resulting in a lower model data volume. This reduces the computational power consumption of the digital twin system's front-end rendering and adapts to real-time visualization display requirements.

[0048] The aforementioned different operating conditions cover the typical operating range of the compressor in actual operation, including the operating status under single or combined operating parameters such as different speeds, near-inlet pressure, exhaust pressure, and medium flow rates, closely matching the actual operating scenarios throughout the equipment's entire life cycle. During the simulation calculation, relying on the structural characteristics of the high-precision mesh model, the mechanical response and structural characteristics of each mesh node of the compressor components under different operating conditions are accurately obtained. The simulation results cover the strain field results of the components under each operating condition, and the simulation results for all operating conditions are standardized and organized to form a finite element simulation dataset.

[0049] Specifically, in constructing the multi-condition data matrix, finite element simulation data can be classified, organized, and associated according to preset structured rules. This orderly integration of relevant data corresponding to different conditions forms a unified data matrix. Thus, the multi-condition data matrix contains strain field result datasets corresponding to different condition parameters. Each data entry in the matrix corresponds to a specific set of condition parameters and is associated with a complete set of strain field result data, achieving a correspondence between condition parameters and strain field response data. Furthermore, to ensure the matching of strain field result data, the strain field result dataset is associated with grid node coordinates through node numbers. Each node number corresponds to a fixed grid node coordinate in the high-precision grid model, and corresponds to the strain value of that node under the current condition.

[0050] In practical applications, cylindrical coordinates are typically used to record data during the finite element simulation phase. Accordingly, the initial finite element simulation data is represented in cylindrical coordinates to indicate the spatial location of nodes and structural deformation information. However, the characteristics of the cylindrical coordinate system are not suitable for the entire application process of this technical solution. To achieve coordinate system compatibility throughout the entire process, coordinate transformation of the finite element simulation data is necessary. Specifically, such as... Figure 3 As shown, step 203 includes the following steps: 301. Convert the cylindrical coordinate system corresponding to the finite element simulation dataset into position coordinates in the Cartesian coordinate system to obtain the converted intermediate dataset.

[0051] 302. Perform deduplication processing on the intermediate dataset to delete duplicate coordinate records and retain the first occurrence of node data, thereby obtaining a unique set of node coordinates.

[0052] 303. Establish a continuous index starting from a preset value for the set of unique node coordinates, and construct a mapping table from coordinates to indexes.

[0053] 304. Perform a left join between the intermediate dataset and the mapping table to generate a multi-condition data matrix containing standard indexes.

[0054] The specific coordinate system transformation can be based on the standard transformation formula between cylindrical coordinate system and Cartesian coordinate system. The cylindrical coordinates of all nodes in the finite element simulation data are converted one by one into three-dimensional position coordinates in Cartesian coordinate system. At the same time, the deformation information, strain data and working condition parameters of each node are retained to form the intermediate dataset after transformation.

[0055] Since finite element simulation datasets cover parameters under multiple operating conditions, the same node position coordinates may exist under different operating conditions. Duplicate data increases the computational cost of subsequent data processing and may also lead to data association errors. By deduplicating the intermediate dataset, removing duplicate records with completely identical coordinates, and retaining only the complete data corresponding to the first occurrence of each node coordinate, a unique set of node coordinates can be formed. This ensures that each node coordinate exists uniquely in the dataset, simplifying the subsequent data processing flow.

[0056] In the specific process of constructing the coordinate-to-index mapping table, initial index values ​​are preset, and the coordinates are sorted according to the unique node coordinate set. A continuous and unique index value is assigned to each node coordinate, and the node coordinates are bound to the corresponding index values ​​to construct the coordinate-to-index mapping table. For example, the unique node coordinate set contains 3 coordinate nodes, namely... ; Node coordinates are sorted according to the set sorting rules. Assign index value 0, node coordinates Assign index value 1, node coordinates Assign index value 2.

[0057] In this embodiment, the left join ensures that all data in the intermediate dataset is completely preserved. Simultaneously, by precisely matching node coordinates with the mapping table, a corresponding standard index value is added to each record in the intermediate dataset. The resulting multi-condition data matrix not only contains different condition parameters and corresponding strain field result datasets, but also assigns a unique standard index to each node data point, achieving precise binding between the standard index, node coordinates, strain field result data, and condition parameters.

[0058] The coordinate system transformation process described above is illustrated using the impeller of a three-stage centrifugal compressor in a real-world application scenario as an example. See [link to relevant documentation] for details. Figure 4 As shown, the original geometric model of the impeller contains 17 repeating blades. The complete impeller structure is obtained through parametric modeling. The finite element simulation data uses cylindrical coordinates to record node positions and deformation information, while 3D rendering requires vertex coordinates described in Cartesian coordinates. Therefore, a coordinate system transformation is needed to calculate the position coordinates of each node in the Cartesian coordinate system. After the coordinate transformation, an intermediate dataset is obtained, which is then cleaned and indexed. The specific process for cleaning and indexing the intermediate dataset is detailed in [link to documentation]. Figure 5 As shown, firstly, the original dataset A after transformation is deduplicated, removing duplicate coordinate records and retaining only the first occurrence of node data, resulting in a unique coordinate set B, thus avoiding interference from redundant data in rendering. Next, a new continuous index starting from 0 is established for the unique coordinate set B, generating dataset C. Simultaneously, a mapping table from coordinates to indices is constructed, assigning each unique node a normalized identifier. Finally, the original dataset A and the index mapping table C are left-joined, merging to generate a new dataset D containing all the original simulation data and with standard indices. Dataset D is arranged in index order and can be directly loaded and used by the front-end, ensuring data integrity and sequence. This series of transformation and processing steps ensures consistency and compatibility between the simulation data and the front-end rendering module, significantly improving data reading efficiency during subsequent dynamic rendering.

[0059] In practical applications, to ensure both computational accuracy and real-time visualization efficiency of digital twin systems, cross-scale data mapping can accurately map strain data from high-precision mesh models to nodes of low-precision mesh models using spatial topology matching algorithms. This allows the low-precision mesh model to maintain rendering speed while displaying high-precision mechanical responses. Specifically, for example... Figure 6 As shown, step 101 includes the following steps: 401. Based on the multi-condition data matrix obtained from simulation, extract the node coordinates and strain field result dataset of the high-precision mesh model, as well as the node coordinates of the low-precision mesh model.

[0060] 402. The strain field result dataset of the high-precision grid model is transferred to the node coordinates of the low-precision grid model through a spatial mapping algorithm of node coordinates to obtain the cross-scale data mapping relationship between the high-precision grid model and the low-precision grid model.

[0061] This embodiment adopts a double-precision mesh model design, which balances the accuracy of finite element simulation calculation with the real-time visualization performance of the digital twin system front end. It solves the technical problem of high computational power of high-precision models and simulation distortion of low-precision models. High-precision mesh models and low-precision mesh models are pre-constructed. Based on the high-precision mesh models and low-precision mesh models, the node coordinates and strain field result dataset of the high-precision mesh model and the node coordinates of the low-precision mesh model are extracted from the multi-condition data matrix, respectively.

[0062] To achieve accurate transfer of high-precision simulation results to low-precision models and ensure consistency between the front-end displayed data and the high-precision simulation results, this embodiment employs the K-nearest neighbor algorithm to project data between the two types of mesh models, achieving accurate cross-mesh strain field data mapping. Specifically, for each node in the low-precision mesh model, the distance between that node and all nodes in the high-precision mesh model is calculated; neighboring nodes within a preset neighborhood are selected as target nodes, and the strain field result datasets of the target nodes are fused using distance weighting to obtain a fused strain field result dataset; the fused strain field result dataset is then assigned to the corresponding nodes in the low-precision mesh model, resulting in a cross-scale data mapping relationship between the high-precision and low-precision mesh models.

[0063] In practical applications, the process of establishing cross-scale data mapping relationships can be found in [reference needed]. Figure 7 As shown, after the high-precision mesh model completes the offline finite element simulation of a certain working condition, the coordinates of all nodes of the high-precision mesh model and the strain field data corresponding to each node are first extracted from the finite element simulation results. At the same time, the set of all nodes of the low-precision mesh model is obtained. Then, in the feature space, for each node of the low-precision model, the spatial distance between the node and all nodes of the high-precision model is calculated one by one. According to the distance sorting result, the K nearest high-precision mesh model nodes are selected as the neighboring nodes of the low-precision mesh model node. Finally, the distance weighted fusion algorithm is used to perform weighted calculation on the strain values ​​of the selected K neighboring nodes, and the fused interpolation result is assigned to the corresponding low-precision mesh model node to obtain the strain value of the low-precision mesh model node.

[0064] During the dynamic projection process using the K-nearest neighbor algorithm described above, the strain field results obtained from the high-precision mesh model simulation can be transmitted to the corresponding node positions of the low-precision mesh model without distortion. This achieves synergistic optimization of simulation accuracy and front-end visualization performance, effectively ensuring that the strain distribution presented on the front-end model is highly consistent with the real results obtained from the high-precision offline simulation. This not only guarantees the accuracy of data display but also meets the performance requirements of real-time front-end rendering.

[0065] In combination with the above Figure 7 In the process of mapping across medium and large scales, specifically in the extraction of coordinate nodes and strain data from the high-precision mesh model, the node coordinate set can be extracted from the finite element simulation data based on the high-precision mesh model. and strain field results data Then, node coordinates are extracted from the finite element simulation data based on the low-precision mesh model. ,in, .

[0066] A composite characteristic space containing spatial coordinates and physical fields is then established, which can be expressed by the following formula:

[0067]

[0068] Specifically, in the K-nearest neighbor dynamic projection process, for each low-precision grid model node... Perform the following procedure: Calculate the spatial distance matrix:

[0069] Select the set of neighboring nodes:

[0070] Strain value weighted fusion: , in, As weight, This is the normalization factor.

[0071] In practical applications, to achieve lightweight and real-time simulation results, a multi-condition data matrix can be used as training samples to train the simulation model. This allows the simulation model to learn the mapping relationship between condition parameters and strain field results. Specifically, for example... Figure 8 As shown, step 102 includes the following steps: 501. Using the multi-condition data matrix as training samples, select a radial basis function of a preset type as the basis function of the interpolation fitting algorithm.

[0072] 502. Construct a joint objective function.

[0073] 503. Using the minimization of the joint objective function as the optimization objective, the weight coefficients and polynomial coefficients of the radial basis functions are obtained by solving.

[0074] 504. Substitute the obtained weight coefficients and polynomial coefficients into the radial basis function to complete the training of the simulation model, so as to establish a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset.

[0075] In this embodiment, the radial basis function has the characteristics of strong local approximation ability and high fitting accuracy for high-dimensional nonlinear data, and is suitable for the complex nonlinear mapping relationship between working condition parameters and strain field response. By presetting the basis function type, the core calculation logic of the interpolation fitting algorithm can be fixed, ensuring the standardization and reproducibility of the subsequent surrogate model training process.

[0076] Specifically, radial basis functions can be of the types Gaussian functions, quadratic functions, etc. By interpolating the radial basis functions with known data points, the target function can be approximated, enabling smooth prediction of high-dimensional outputs in a low-dimensional input space. Generally speaking, the general form of a radial basis function can be expressed by the following formula:

[0077] in, This refers to the output value of the simulation model, i.e., the target strain field result data; This is the input vector of the simulation model, i.e., the vector of operating parameters; For the first The working condition parameter vector of each training sample ; This represents the total number of training samples; For the first The weight coefficients of each radial basis function are one of the parameters that need to be solved during the training of the simulation model. They are used to measure the contribution of the corresponding basis function in fitting the nonlinear mapping relationship. It is a radial basis function of a preset type, whose input is the Euclidean distance of a vector, used to capture the local nonlinear relationship between the operating parameters and the strain field; Input operating condition parameter vector With the The working condition parameter vector of each training sample The Euclidean distance between them reflects the spatial similarity between the current input and the training samples, and is equivalent to the input vector of the radial basis function; It is a polynomial used to fit the global linear trend of the data, complementing the local approximation ability of the radial basis function.

[0078] The aforementioned joint objective function includes a data fitting term to measure the deviation between predicted and true values, and cross-scale constraints constructed based on the data mapping relationship. The data fitting term ensures the accuracy of the simulation model's predictions; generally, a smaller data fitting term indicates a higher consistency between the simulation model's predictions and the actual simulation results. The cross-scale constraints ensure consistency between the strain distribution displayed on the front end and the high-precision simulation results, and can be accurately transferred to the low-precision visualization model through the cross-scale mapping relationship. Specifically, the joint objective function can be solved using numerical optimization algorithms, such as least squares or gradient descent, to obtain the weight coefficients and polynomial coefficients of the radial basis function. The core of this solution process is to continuously adjust the values ​​of the two types of coefficients to minimize the value of the joint objective function. At this point, the corresponding coefficient combination ensures that the simulation model satisfies both the accuracy requirement of minimizing the deviation between predicted and true values ​​and the visualization adaptation requirement of consistent cross-scale data mapping, thus guaranteeing the prediction accuracy of the simulation model.

[0079] The obtained weight coefficients and polynomial coefficients are then substituted into the preset radial basis functions to complete the training of the simulation model. Since the trained simulation model has accurately learned the nonlinear mapping relationship between the parameters of multiple working conditions and the strain field result dataset, it is no longer necessary to solve the physical equations through high-precision finite element offline simulation. It can directly undertake the real-time strain field prediction task of the digital twin model system.

[0080] In practical applications, the training process of the above simulation model uses a multi-condition data matrix as a single sample. To avoid overfitting of the simulation model to the training samples, after the simulation model training is completed, the weight coefficients and polynomial coefficients of the radial basis function can be recalculated through cross-validation based on the test samples that did not participate in the simulation model training, so as to retrain the simulation model.

[0081] Specifically, the above process uses test samples that were not involved in the initial training of the simulation model as the basis for verification and optimization. These test samples also originate from high-precision finite element simulations of multi-stage centrifugal compressor components, sharing the same source and data structure as the training samples. They contain operating parameters not involved in the initial training and corresponding high-fidelity strain field results, accurately reflecting the model's predictive ability for unknown operating conditions. During the verification and optimization process, cross-validation is conducted based on these test samples. This involves inputting the operating parameters of the test samples into the initially trained model to obtain the model's output strain field predictions. The predicted strain values ​​are then compared with the actual strain values ​​in the test samples to verify the simulation model's prediction deviation for unknown operating conditions. Using the cross-validation results as the optimization guide, the weight coefficients and polynomial coefficients of the radial basis function are recalculated. This recalculation process follows the same logic as the initial training, minimizing the joint objective function as the optimization goal. The only difference is that the data samples are replaced with a combination of test and training samples, and iterative calculations yield more adaptable weight coefficients and polynomial coefficients.

[0082] In practical applications, radial basis functions are selected as the basis functions for training the simulation model in the interpolation fitting algorithm. Specifically, during the training process, impeller strain field data under different speed conditions obtained from finite element simulation are used as training samples, and the corresponding speed and other operating parameters are used as inputs to establish a functional relationship from input to output. Then, the weight coefficients of the radial basis functions and the necessary polynomial coefficients are solved according to the interpolation conditions satisfied by all training sample points, thereby constructing the simulation model. After the simulation model is constructed, its prediction accuracy is verified using test data that was not used in the training. The K value, basis function parameters, or regularization terms are adjusted through cross-validation to prevent overfitting. The trained simulation model has significant advantages: on the one hand, it has high fitting accuracy for the nonlinear distribution of impeller strain and can accurately predict the response of unseen intermediate operating points; on the other hand, the computational cost is much lower than that of the original finite element model, and prediction can be completed in a very short time.

[0083] In practical applications of digital twin systems for multi-stage centrifugal compressors, multiple terminal devices may access and use the system simultaneously. Deploying the trained simulation model to the backend server of the digital twin system can ensure the continuity of data flow and meet high concurrency requirements in the industrial field. In this embodiment, the backend server adopts a microservice architecture, including a distributed storage module and a lightweight computing module. A distributed storage module is used to store the real-time operating condition database and the prediction result database; The lightweight calculation module is used to respond to front-end interaction requests, read the node coordinates and operating parameters of the multi-stage centrifugal compressor components from the real-time operating condition database, call the simulation model to calculate the target strain field result dataset of the multi-stage centrifugal compressor components, and write the target strain field result dataset to the prediction result database for front-end visualization. In specific practical application scenarios, the microservice architecture of a multi-stage centrifugal compressor digital twin system can be found in [reference needed]. Figure 9 As shown, Figure 9 The digital twin system for a multi-stage centrifugal compressor receives Cartesian coordinates and strain data (after coordinate transformation and standardization) as the initial data source through a data input module. This initial data source is uploaded to a database configured on a cloud server for storage, where the database centrally manages the Cartesian coordinates and strain field results. Subsequently, the simulation model module retrieves historical data and newly input operating parameters from the cloud server database to perform strain field prediction calculations. The newly generated strain field results are then sent back to the database, enabling iterative updates and optimizations. During this process, the backend service control module, acting as the system control center, interacts with the simulation model module and the database asynchronously. It sends operating parameter commands to the simulation model module to trigger calculations and asynchronously retrieves updated strain field results from the database, ensuring efficient collaboration between calculations and data retrieval. Finally, the backend service control module transmits the processed strain field results to the front-end page display module. This allows the front-end page display module to drive model rendering using visualization rendering technology, providing a graphical display of the strain field distribution of the multi-stage centrifugal compressor components and presenting users with an intuitive visual interface showcasing the equipment's operating status.

[0084] In practical applications, to enable maintenance personnel to quickly identify the strain status of multi-stage centrifugal compressor components through color distribution, and to achieve accurate mapping from data to visualization, specifically, for example... Figure 10 As shown, step 104 includes the following steps: 601. Based on a preset color spectrum, the target strain field result data is standardized and mapped to obtain vertex color values.

[0085] 602. Render the low-precision mesh model of the multi-stage centrifugal compressor assembly using the vertex colors to generate a digital twin model of the multi-stage centrifugal compressor assembly.

[0086] In this embodiment, the preset color spectrum can be a rainbow color spectrum or a warm / cool color spectrum, which predefines the correspondence between strain value ranges and colors. For example, low strain values ​​are mapped to blue, medium strain values ​​to green, and high strain values ​​to red. Through this mapping, the strain value of each low-precision mesh model node is converted into a unique vertex color value, realizing the quantitative association between strain values ​​and color information, and transforming abstract mechanical data into color features that can be directly perceived visually.

[0087] Subsequently, the obtained vertex color values ​​are bound to the corresponding nodes of the low-precision mesh model of the multi-stage centrifugal compressor component, and the low-precision mesh model is rendered using the 3D rendering engine of the web front-end. After rendering, the surface of the low-precision mesh model will show a continuous color distribution based on the vertex color values. Different color areas intuitively reflect the strain magnitude at various locations of the multi-stage centrifugal compressor component, generating a digital twin model of the multi-stage centrifugal compressor component with strain field visualization effects. This digital twin model retains the real-time rendering efficiency of the low-precision mesh model while restoring the distribution characteristics of the high-precision strain field through color mapping.

[0088] Specifically, in the process of using the trained simulation model for strain rendering on the web front-end, see [link to web page front-end]. Figure 11 As shown, in the offline phase, a neural network-based simulation model is trained using a multi-condition data matrix generated by high-precision finite element simulation as training samples. In the online prediction phase, the system inputs the real-time collected compressor operating parameters into the trained simulation model. Based on the learned mapping relationship, the simulation model quickly outputs the strain field prediction data under the corresponding operating condition. The trained simulation model is then integrated and deployed to the backend server. After receiving the strain field prediction data output by the simulation model, the backend server performs structured processing in conjunction with the real-time operating parameters. The processed strain field data is then transmitted to the web frontend. The web frontend drives a low-precision mesh model through a lightweight 3D rendering engine to achieve real-time 3D visualization of the strain field.

[0089] For details regarding strain rendering, see [link / reference]. Figure 12As shown, the geometric model is first constructed. This process reconstructs the 3D mesh model of the impeller by organizing the pre-imported node data in the Cartesian coordinate system according to the index order, forming triangular patches with every three adjacent points. To improve rendering performance, the mesh vertex set and patch index can be set in batches. Then, vertex coordinates and other attributes are stored using a contiguous memory buffer, while the strain field data is standardized to obtain normalized color values. Next, dynamic color mapping and rendering are performed. This process uses linear interpolation on a preset color gradient based on the color value of each node to obtain the corresponding vertex color value, and assigns this vertex color to the color attribute of the corresponding vertex in the mesh model. Using the shader function, the above color data is passed to the image processor to complete rasterization rendering and generate a color cloud map of the impeller strain field.

[0090] To facilitate front-end interface interaction, the user's browser continuously communicates with the back-end service during rendering to obtain the next frame of data or respond to user interactions. When the user rotates, zooms, or clicks on a component in the front-end interface, the viewpoint can be changed via controls, and the strain value of the selected component can be queried via raycasting. A floating information panel pops up on the graphical interface to display the specific values. See [link to relevant documentation]. Figure 13 The left-hand data panel is shown. Through the above method, the front end achieves immersive real-time monitoring of the impeller strain field: color changes dynamically reflect changes in the equipment's operating status, allowing operators to intuitively observe the stress on key components of the compressor under different operating conditions, achieving a visual perception of the equipment's structural state. When abnormal strain distribution occurs, such as a sudden change in color from blue to red in a certain area, indicating a surge in stress, users can promptly detect this and, combined with backend algorithm analysis, issue an early warning.

[0091] Furthermore, as a specific implementation of the above method, this application embodiment provides a digital twin modeling device for a multi-stage centrifugal compressor, such as... Figure 14 As shown, the device includes: a setup unit 71, a training unit 72, a deployment unit 73, and a rendering unit 74.

[0092] Unit 71 is established to establish a cross-scale data mapping relationship between the high-precision grid model and the low-precision grid model based on the multi-condition data matrix obtained from the simulation of the multi-stage centrifugal compressor components. Training unit 72 is used to train the simulation model by incorporating the data mapping relationship into the interpolation fitting algorithm as a constraint condition, using the multi-condition data matrix as training samples, so as to establish a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset. Deployment unit 73 is used to deploy the trained simulation model to the backend server of the digital twin system, so that the simulation model can be called through the backend server to predict the strain field and output the target strain field result dataset. The rendering unit 74 is used to drive the low-precision mesh model of the multi-stage centrifugal compressor component to render using the target strain field result dataset, thereby generating a digital twin model of the multi-stage centrifugal compressor component.

[0093] The digital twin modeling device for multi-stage centrifugal compressors provided in this invention, compared with current methods that achieve digital twin modeling of multi-stage centrifugal compressors through high-precision finite element modeling simulation results or simplified mesh rendering, establishes a cross-scale data mapping relationship between high-precision and low-precision mesh models based on a multi-condition data matrix obtained from the simulation of multi-stage centrifugal compressor components. Using the multi-condition data matrix as training samples, the simulation model is trained by incorporating the data mapping relationship as a constraint in the interpolation fitting algorithm to establish a nonlinear mapping relationship between multi-condition parameters and the strain field result dataset. The trained simulation model is deployed to the backend server of the digital twin system, allowing the backend server to call the simulation model for strain field prediction and output the target strain field result dataset. The target strain field result dataset is then used to drive the rendering of the low-precision mesh model of the multi-stage centrifugal compressor components, generating a digital twin model of the multi-stage centrifugal compressor components. The entire process, by constructing a cross-scale data mapping relationship and incorporating this relationship as a constraint into the training process of the lightweight simulation model, fundamentally achieves a dual optimization of computational accuracy and rendering efficiency. On the one hand, this method utilizes a high-precision mesh model as the data source, fully preserving the high computational accuracy of the finite element model and ensuring the accuracy of the description of the complex physical field of a multi-stage centrifugal compressor. On the other hand, by mapping the high-precision simulation results to a low-precision mesh model, it fully leverages the rendering advantages of the low-precision mesh model, ensuring high frame rate rendering of the front-end interface. More importantly, the simulation model based on offline training decouples the simulation calculation from the 3D rendering architecture, eliminating the need for heavy iterative calculations during system runtime. This effectively balances the dual requirements of high-precision simulation and real-time rendering for the digital twin model of the multi-stage centrifugal compressor, achieving real-time and accurate mapping of the equipment's operating status.

[0094] In specific application scenarios, the device further includes: The first building unit is used to pre-build a high-precision mesh model for finite element calculation and a low-precision mesh model for visualization based on the original geometric model of the multi-stage centrifugal compressor component before establishing the cross-scale data mapping relationship between the high-precision mesh model and the low-precision mesh model based on the multi-condition data matrix obtained from the simulation of the multi-stage centrifugal compressor component. The simulation unit is used to perform simulation calculations under different operating conditions using the high-precision mesh model to obtain the finite element simulation dataset of the multi-stage centrifugal compressor assembly. The second building unit is used to construct a multi-condition data matrix based on the finite element simulation dataset of a multi-stage centrifugal compressor component. The multi-condition data matrix contains strain field result datasets corresponding to different operating condition parameters. The strain field result datasets correspond to the grid node coordinates through node numbers.

[0095] In specific application scenarios, the finite element simulation data initially uses a cylindrical coordinate system to record node positions and deformation information. The second building unit is specifically used for: The cylindrical coordinate system corresponding to the finite element simulation dataset is converted to the position coordinates in the Cartesian coordinate system to obtain the converted intermediate dataset. The intermediate dataset is deduplicated to remove duplicate coordinate records and retain the first occurrence of node data, resulting in a unique set of node coordinates. Establish a continuous index starting from a preset value for the set of unique node coordinates, and construct a mapping table from coordinates to indexes; The intermediate dataset is left-joined with the mapping table to generate a multi-condition data matrix containing standard indexes.

[0096] In specific application scenarios, the establishment unit includes: The extraction module is used to extract the node coordinates and strain field result dataset of the high-precision mesh model and the node coordinates of the low-precision mesh model based on the multi-condition data matrix obtained from the simulation. The mapping module is used to transfer the strain field result dataset of the high-precision grid model to the node coordinates of the low-precision grid model through a spatial mapping algorithm of node coordinates, so as to obtain the cross-scale data mapping relationship between the high-precision grid model and the low-precision grid model.

[0097] In specific application scenarios, the mapping module is specifically used for: For each node in the low-precision mesh model, calculate the distance between that node and all nodes in the high-precision mesh model; Neighboring nodes within a preset neighborhood range are selected as target nodes, and the strain field result datasets of the target nodes are fused using distance weighting to obtain the fused strain field result dataset. The fused strain field result dataset is assigned to the corresponding nodes of the low-precision grid model to obtain the cross-scale data mapping relationship between the high-precision grid model and the low-precision grid model.

[0098] In specific application scenarios, the training unit is specifically used for: Using the multi-condition data matrix as training samples, a radial basis function of a preset type is selected as the basis function for the interpolation fitting algorithm; Construct a joint objective function, which includes a data fitting term for measuring the deviation between the predicted and the true values, and cross-scale constraints based on the data mapping relationship; Using the minimization of the joint objective function as the optimization objective, the weight coefficients and polynomial coefficients of the radial basis functions are obtained by solving the problem. The obtained weight coefficients and polynomial coefficients are substituted into the radial basis function to complete the training of the simulation model, so as to establish a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset. The training unit is further used for: After training the simulation model using the multi-condition data matrix as training samples and incorporating the data mapping relationship into the interpolation fitting algorithm as a constraint to establish a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset, the simulation model is retrained based on test samples that did not participate in the simulation model training by resolving the weight coefficients and polynomial coefficients of the radial basis function through cross-validation.

[0099] In specific application scenarios, the backend server adopts a microservice architecture, including a distributed storage module and a lightweight computing module; The distributed storage module is used to store the real-time operating condition database and the prediction result database; The lightweight computing module is used to respond to front-end interaction requests, read the node coordinates and operating parameters of the multi-stage centrifugal compressor component from the real-time operating condition database, call the simulation model to calculate the target strain field result dataset of the multi-stage centrifugal compressor component, and write the target strain field result dataset into the prediction result database for front-end visualization. The rendering unit is specifically used for: The target strain field result data is standardized and mapped based on a preset color spectrum to obtain vertex color values; The vertex colors are used to render the low-precision mesh model of the multi-stage centrifugal compressor assembly, generating a digital twin model of the multi-stage centrifugal compressor assembly.

[0100] Based on the above method, the present application also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described digital twin modeling method for a multi-stage centrifugal compressor.

[0101] Based on this understanding, the technical solution of this application can be embodied in the form of a software product. The software product can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, or portable hard drive), and includes several instructions to cause a computer device (such as a personal computer, server, or network device) to execute the methods described in the various implementation scenarios of this application.

[0102] Based on the above method and the virtual device embodiment, in order to achieve the above objectives, this application embodiment also provides a physical device for digital twin modeling of a multi-stage centrifugal compressor, which can be a computer, smartphone, tablet computer, smartwatch, server, or network device, etc. The physical device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the above-mentioned digital twin modeling method for a multi-stage centrifugal compressor.

[0103] Optionally, the physical device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0104] In an exemplary embodiment, see Figure 15 The aforementioned physical device includes a communication bus, a processor, a memory, and a communication interface. It may also include an input / output interface and a display device. The various functional units can communicate with each other via the bus. The memory stores a computer program, and the processor executes the program stored in the memory to perform the digital twin modeling method for the multi-stage centrifugal compressor described in the above embodiments.

[0105] Those skilled in the art will understand that the physical device structure of the digital twin model of the multi-stage centrifugal compressor provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0106] The aforementioned storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the digital twin model of the multi-stage centrifugal compressor, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing entity.

[0107] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented using software plus necessary general-purpose hardware platforms, or it can be implemented in hardware. By applying the technical solution of this application, compared with the existing methods, this application fundamentally achieves dual optimization of computational accuracy and rendering efficiency by constructing cross-scale data mapping relationships and integrating these data mapping relationships as constraints into the training process of the lightweight simulation model. On the one hand, this method uses a high-precision mesh model as the data source, fully preserving the high computational accuracy of the finite element model and ensuring the accuracy of the description of the complex physical field of the multi-stage centrifugal compressor; on the other hand, by mapping the high-precision simulation results to the low-precision mesh model, the rendering advantages of the low-precision mesh model are fully utilized, ensuring high frame rate rendering of the front-end interface. More importantly, the simulation model based on offline training achieves architectural decoupling of simulation calculation and 3D rendering, so that the system does not need heavy iterative calculations during operation, effectively balancing the dual requirements of high-precision simulation and real-time rendering of the digital twin model of the multi-stage centrifugal compressor, and realizing real-time and accurate mapping of the equipment's operating status.

[0108] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0109] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A digital twin modeling method for a multi-stage centrifugal compressor, characterized in that, include: Based on the multi-condition data matrix obtained from the simulation of multi-stage centrifugal compressor components, a cross-scale data mapping relationship is established between high-precision grid models and low-precision grid models. Using the multi-condition data matrix as training samples, the simulation model is trained by incorporating the data mapping relationship into the interpolation fitting algorithm as a constraint condition, so as to establish a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset. The trained simulation model is deployed to the backend server of the digital twin system, so that the backend server can call the simulation model to predict the strain field and output the target strain field result dataset. The target strain field result dataset is used to drive the rendering of a low-precision mesh model of a multi-stage centrifugal compressor assembly, generating a digital twin model of the multi-stage centrifugal compressor assembly.

2. The method according to claim 1, characterized in that, Before establishing the cross-scale data mapping relationship between the high-precision grid model and the low-precision grid model based on the multi-condition data matrix obtained from the simulation of the multi-stage centrifugal compressor components, the method further includes: Based on the original geometric model of the multi-stage centrifugal compressor components, a high-precision mesh model for finite element calculation and a low-precision mesh model for visualization are pre-built. The high-precision mesh model was used to perform simulation calculations under different operating conditions to obtain a finite element simulation dataset of the multi-stage centrifugal compressor assembly. Based on the finite element simulation dataset of a multi-stage centrifugal compressor component, a multi-condition data matrix is ​​constructed. The multi-condition data matrix contains strain field result datasets corresponding to different condition parameters. The strain field result datasets correspond to the grid node coordinates through node numbers.

3. The method according to claim 2, characterized in that, The initial finite element simulation data uses a cylindrical coordinate system to record node positions and deformation information. The finite element simulation dataset based on a multi-stage centrifugal compressor assembly constructs a multi-condition data matrix, including: The cylindrical coordinate system corresponding to the finite element simulation dataset is converted to the position coordinates in the Cartesian coordinate system to obtain the converted intermediate dataset. The intermediate dataset is deduplicated to remove duplicate coordinate records and retain the first occurrence of node data, resulting in a unique set of node coordinates. Establish a continuous index starting from a preset value for the set of unique node coordinates, and construct a mapping table from coordinates to indexes; The intermediate dataset is left-joined with the mapping table to generate a multi-condition data matrix containing standard indexes.

4. The method according to claim 1, characterized in that, The multi-condition data matrix obtained from the simulation of a multi-stage centrifugal compressor component establishes a cross-scale data mapping relationship between the high-precision grid model and the low-precision grid model, including: Based on the multi-condition data matrix obtained from simulation, the node coordinates and strain field result dataset of the high-precision mesh model and the node coordinates of the low-precision mesh model are extracted respectively. The strain field result dataset of the high-precision grid model is transferred to the node coordinates of the low-precision grid model through a spatial mapping algorithm of node coordinates, thereby obtaining the cross-scale data mapping relationship between the high-precision grid model and the low-precision grid model.

5. The method according to claim 4, characterized in that, The step of transferring the strain field result dataset of the high-precision mesh model to the node coordinates of the low-precision mesh model through a spatial mapping algorithm of node coordinates, thereby obtaining the cross-scale data mapping relationship between the high-precision mesh model and the low-precision mesh model, includes: For each node in the low-precision mesh model, calculate the distance between that node and all nodes in the high-precision mesh model; Neighboring nodes within a preset neighborhood range are selected as target nodes, and the strain field result datasets of the target nodes are fused using distance weighting to obtain the fused strain field result dataset. The fused strain field result dataset is assigned to the corresponding nodes of the low-precision grid model to obtain the cross-scale data mapping relationship between the high-precision grid model and the low-precision grid model.

6. The method according to any one of claims 1-5, characterized in that, The step of using the multi-condition data matrix as training samples and incorporating the data mapping relationship into the interpolation fitting algorithm as a constraint to train the simulation model, in order to establish a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset, includes: Using the multi-condition data matrix as training samples, a radial basis function of a preset type is selected as the basis function for the interpolation fitting algorithm; Construct a joint objective function, which includes a data fitting term for measuring the deviation between the predicted and the true values, and cross-scale constraints based on the data mapping relationship; Using the minimization of the joint objective function as the optimization objective, the weight coefficients and polynomial coefficients of the radial basis functions are obtained by solving the problem. The obtained weight coefficients and polynomial coefficients are substituted into the radial basis function to complete the training of the simulation model, so as to establish a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset. Accordingly, after using the multi-condition data matrix as training samples and incorporating the data mapping relationship into the interpolation fitting algorithm as a constraint to train the simulation model, thereby establishing a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset, the method further includes: Based on test samples that were not used in the simulation model training, the weight coefficients and polynomial coefficients of the radial basis function are re-solved through cross-validation to retrain the simulation model.

7. The method according to any one of claims 1-5, characterized in that, The backend server adopts a microservice architecture, including a distributed storage module and a lightweight computing module; The distributed storage module is used to store the real-time operating condition database and the prediction result database; The lightweight computing module is used to respond to front-end interaction requests, read the node coordinates and operating parameters of the multi-stage centrifugal compressor component from the real-time operating condition database, call the simulation model to calculate the target strain field result dataset of the multi-stage centrifugal compressor component, and write the target strain field result dataset into the prediction result database for front-end visualization. The process of using the target strain field result dataset to drive the rendering of a low-precision mesh model of a multi-stage centrifugal compressor assembly to generate a digital twin model of the multi-stage centrifugal compressor assembly includes: The target strain field result data is standardized and mapped based on a preset color spectrum to obtain vertex color values; The vertex colors are used to render the low-precision mesh model of the multi-stage centrifugal compressor assembly, generating a digital twin model of the multi-stage centrifugal compressor assembly.

8. A digital twin modeling device for a multi-stage centrifugal compressor, characterized in that, include: A cell is established to create a cross-scale data mapping relationship between a high-precision mesh model and a low-precision mesh model based on the multi-condition data matrix obtained from the simulation of a multi-stage centrifugal compressor component. The training unit is used to train the simulation model by incorporating the data mapping relationship into the interpolation fitting algorithm as a constraint condition, using the multi-condition data matrix as training samples, so as to establish a nonlinear mapping relationship between the multi-condition parameters and the strain field result dataset. The deployment unit is used to deploy the trained simulation model to the backend server of the digital twin system, so that the simulation model can be called through the backend server to predict the strain field and output the target strain field result dataset. The rendering unit is used to drive the low-precision mesh model of the multi-stage centrifugal compressor component to render using the target strain field result dataset, thereby generating a digital twin model of the multi-stage centrifugal compressor component.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images