High-temperature and high-pressure pump accurate assembly risk identification system based on digital twinning

Abstract

The invention relates to a high-temperature and high-pressure pump precise assembly risk identification method and system based on digital twinning. The method comprises the steps that real-time assembly parameters of all assembly components on a current assembly site are obtained; constructing a first input parameter based on the real-time assembly parameter; constructing a second input parameter based on the digital twinborn model; performing linear quantization on the first input parameter and the second input parameter to respectively obtain a first input vector and a second input vector; carrying out weighted splicing after splicing to obtain a splicing vector, and inputting the splicing vector into a risk identification model; and inputting the first input vector and the second input vector into a risk identification model to obtain a risk identification result. According to the method, real-time evolution and possible risk information can be discovered at the same time, and through reasonable expression, different types of complex information of different components are fused for an artificial intelligence model to mine deep association so as to obtain useful risk identification information, and efficient intelligent risk identification is realized.

Description

Technical Field

[0001] This invention belongs to the field of intelligent identification technology, and in particular relates to a method and system for risk identification in the precise assembly of high-temperature and high-pressure pumps based on digital twins. Background Technology

[0002] High-temperature and high-pressure pumps are specialized pumps designed for handling high-temperature media (such as steam), playing a vital role in key sectors such as energy, chemical engineering, and metallurgy. In recent years, with continuous advancements in industrial technology, special operating conditions involving high temperatures and high pressures are increasingly common in actual production, placing more stringent demands on pumps used in these environments. High-temperature and high-pressure pumps are typically used in extreme environments, such as the petroleum, chemical, and power industries, requiring normal operation under high temperatures (potentially exceeding 600°C) and high pressures (potentially hundreds of atmospheres). In such environments, the pump's sealing performance, material durability, and assembly precision are extremely critical; even slight negligence can lead to leaks, failures, or even accidents. Pumps used in high-pressure differential applications are highly susceptible to severe erosion and cavitation in the pump head, plunger, and other components, accompanied by intense vibration and noise.

[0003] High-temperature, high-pressure pumps have stringent assembly requirements, demanding high reliability and precision. To achieve process consistency and improve pump production efficiency, automated assembly technology is often introduced into the pump production process. However, even with robots in assembly, various assembly parameter deviations cannot be eliminated due to the inconsistent target parameters and low positioning accuracy of manual teaching and visual positioning methods. This is also related to the insufficient precision achievable by automated assembly robotic arms. Therefore, it is necessary to effectively identify potential risks during the assembly process, allowing for proactive adjustments or manual intervention.

[0004] On the one hand, existing risk identification and analysis methods mainly involve constructing risk rule models. This involves using historical data and expert / human experience to select feature combinations sensitive to risk values, and then applying specific rules or models to determine the risk identification situation. However, this traditional approach relies heavily on expert / human experience. High-risk assembly problems are dynamic and evolve in real time. Without evolution, specific rules or models cannot adapt to the forms, occurrences, and information expressions of new high-risk assembly problems. Moreover, human experience is limited and it is difficult to uncover deep-seated correlation information.

[0005] On the other hand, the rapid development of artificial intelligence (AI) technology is continuously driving the vigorous development of computer-related vertical applications. Benefiting from the rapidly increasing amount of data and increasingly powerful computing resources, researchers are gradually transferring the success of AI technology to various domain tasks. The vertical applications of convolutional neural networks (CNNs) are extensive, and the network framework of a CNN determines the learning capacity of the network model, thus determining the performance ceiling of the network model in specific application domains. Training methods ensure the effective learning of massive parameters in the network model, promoting effective convergence during training. The quantity and quality of data drive the network model's representation of data patterns, and the representation of data patterns and the characteristic expression of information have a significant impact on the performance of the neural network model. In the development of neural networks in the field of intelligent manufacturing, the cost of labeling and representing massive amounts of data has limited their widespread application to some extent. To more quickly expand the application scenarios of AI technology, exploring and utilizing big data resources is becoming a hot topic and an inevitable development path. However, neural networks with strong fitting capabilities are extremely sensitive to the noise they contain, easily fitting noisy data, causing a decline in network generalization performance. At the same time, the large amount of data without proper representation leads to a huge computational burden.

[0006] When applying artificial intelligence (AI) technology to risk identification, a key challenge is how to fully utilize on-site information from the assembly process, maximize the discovery of information expressed by assembly parameters, and rationally express these parameters to achieve intelligent risk identification using AI. Based on this challenge, this invention can simultaneously detect real-time evolving and potential risk information. By rationally expressing and integrating complex information from different components and their varying types, it allows AI models to mine deeper relationships and obtain useful risk identification information, thus achieving efficient and intelligent risk identification. Summary of the Invention

[0007] To address the aforementioned problems in the prior art, this invention proposes a method and system for precise assembly risk identification of high-temperature and high-pressure pumps based on digital twins. The method includes: Step S1: Obtain the real-time assembly parameters of each assembly component at the current assembly site; construct the first input parameters based on the real-time assembly parameters; synchronize the digital twin model based on the real-time assembly parameters, and simulate the assembly process based on the digital twin model; obtain the predicted assembly parameters based on the simulation results; construct the second input parameters based on the predicted assembly parameters. The step of obtaining the predicted assembly parameters from the simulation results is as follows: based on the simulation results, the assembly trajectory is obtained, and the assembly parameters corresponding to the next assembly site relative to the current assembly site in the assembly trajectory are used as the predicted assembly parameters. Step S2: Linearly quantize the first input parameter and the second input parameter to obtain the first input vector and the second input vector respectively; weight and concatenate the first input vector and the second input vector to obtain the concatenated vector, which is then input into the risk identification model to obtain the classification result; Step S3: Input the first input vector and the second input vector into the risk identification model to obtain the risk identification result; The first input vector and the second input vector are dynamically weighted. The first weight of the first input vector and the second weight of the second input vector are dynamically assigned according to different assembly stages. Specifically, in the first stage of assembly, the weight of the first input vector is set to be higher than that of the second input vector; in the second stage of assembly, the second weight is set to be higher than that of the first weight; and in the third stage of assembly, the first weight is set to be equal to that of the second weight.

[0008] Furthermore, the step of obtaining the predicted assembly parameters from the simulation results specifically involves: obtaining multiple optional assembly trajectories based on the simulation results; selecting one of the multiple optional assembly trajectories as the predicted assembly trajectory based on the current assembly preference; and using the assembly parameters corresponding to the next assembly site relative to the current assembly site in the predicted assembly trajectory as the predicted assembly parameters.

[0009] Furthermore, assembly preference refers to the current assembly system and / or assembly operators' preference for selecting assembly trajectories.

[0010] Furthermore, a visual view interface is created, and when using real-time assembly parameters to dynamically calibrate the model parameters in the digital twin model, a real-time rendering engine is used to visualize the components to be assembled in the key assembly areas of the assembly process.

[0011] Furthermore, the risk identification model is an artificial intelligence model.

[0012] Furthermore, the artificial intelligence model is a neural network model, a machine learning model, or a temporal graph convolutional network.

[0013] Furthermore, the classification and identification results are coarse classification results, including three types: normal state, slight risk, and severe risk.

[0014] Furthermore, the artificial intelligence model is a graph neural network model based on the graph attention mechanism.

[0015] Furthermore, a 3-layer graph neural network model is used in parallel.

[0016] A risk identification system for precise assembly of high-temperature and high-pressure pumps based on digital twins is provided, wherein the system is used to implement the aforementioned risk identification method for precise assembly of high-temperature and high-pressure pumps based on digital twins.

[0017] The beneficial effects of this invention include: (1) During the assembly of high temperature and high pressure pumps, risk identification is carried out by using real-time parameters and predicted parameters at the same time. Real-time evolution and possible risk information are discovered, forming a complementary value. Through the deep integration of virtual and real data, the comprehensiveness and accuracy of risk judgment are significantly improved, the robustness of risk identification is enhanced, occasional interference is filtered out, and efficient intelligent risk identification is achieved. (2) It can provide a reasonable expression to integrate complex information of different types of components for artificial intelligence models to explore their deep-seated relationships in order to obtain useful risk identification information; furthermore, by using linear quantization, it can quantize different types of complex assembly parameters into the same uniform linear space, which reduces the analytical sensitivity bias caused by the limited numerical expression of some types and the rich expression of some types of numerical parameters. It provides a basis for the organic integration of complex multi-type parameters and enables the assembly parameters of different components to be aligned into the same numerical space at the same time, thereby improving the sensitivity of risk discovery in specific assembly stages using artificial intelligence models. Attached Figure Description

[0018] The accompanying drawings, which are provided to further illustrate the invention and form part of this application, are not intended to unduly limit the invention. In the drawings: Figure 1 A schematic diagram of the risk identification method for precise assembly of high-temperature and high-pressure pumps based on digital twins provided by the present invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The illustrative embodiments and descriptions are only used to explain the present invention and are not intended to limit the present invention.

[0020] This invention proposes a method and system for risk identification in the precise assembly of high-temperature and high-pressure pumps based on digital twins, as shown in the attached figure. Figure 1 As shown, the method includes the following steps: Step S1: Obtain the real-time assembly parameters of each assembly component at the current assembly site; construct the first input parameters based on the real-time assembly parameters; synchronize the digital twin model based on the real-time assembly parameters, and simulate the assembly process based on the digital twin model; obtain the predicted assembly parameters based on the simulation results; construct the second input parameters based on the predicted assembly parameters. The step of obtaining the predicted assembly parameters from the simulation results is as follows: based on the simulation results, the assembly trajectory is obtained, and the assembly parameters corresponding to the next assembly site relative to the current assembly site in the assembly trajectory are used as the predicted assembly parameters. Alternatively: The step of obtaining predicted assembly parameters from simulation results specifically involves: obtaining multiple optional assembly trajectories based on simulation results; selecting one of the multiple optional assembly trajectories as the predicted assembly trajectory based on the current assembly preference; and using the assembly parameters corresponding to the next assembly site relative to the current assembly site in the predicted assembly trajectory as the predicted assembly parameters; wherein: the assembly preference is the selection preference of the current assembly system, assembly operators, etc., for the assembly trajectory. Preferred approach: When simulating the assembly process based on a digital twin model, it is necessary to dynamically calibrate the model parameters in the digital twin model using real-time assembly parameters, correct the finite element material properties, and then use incremental learning to update the model response surface to ensure consistency with the physical entity; calculate the assembly parameters such as stress, strain, and temperature gradient of the model under applied load conditions by calculating the displacement of each node in the finite element model; and obtain the environmental parameters in the current assembly site based on the simulation of environmental parameter changes. Preferred: Create a visual view interface and use a real-time rendering engine to visualize the components to be assembled in the key assembly areas of the assembly process when dynamically calibrating the model parameters in the digital twin model using real-time assembly parameters. Real-time parameters can dynamically reflect the instantaneous state of the assembly site (such as current torque value and temperature fluctuations), providing deterministic information. However, they can only capture static risks at the current moment and may not be able to foresee impending problems. On the other hand, predictive assembly parameters can introduce potential future states (such as stress concentration trends on sealing surfaces and the magnitude of thermal deformation) into the risk identification process, thereby providing probabilistic risk warnings. However, relying solely on predictive parameters may be affected by model errors. Combining the two can compensate for their respective shortcomings. In the assembly process of high-temperature and high-pressure pumps, using both real-time and predictive parameters for risk identification can create complementary value. The deep fusion of virtual and real data can significantly improve the comprehensiveness and accuracy of risk assessment, enhance the robustness of risk identification, and filter out occasional interference. Step S2: Linearly quantize the first input parameter and the second input parameter to obtain the first input vector and the second input vector respectively; linear quantization is to transform the nonlinear and non-uniformly varying assembly parameters into a linearly quantized numerical space; linear quantization is performed across components and across configuration parameter types; Step S2 specifically includes the following steps: Step S21: Obtain historical assembly data; historical assembly data is the assembly parameter information of each assembly component corresponding to the continuously monitored assembly site for the assembly of the high temperature and high pressure pump to be assembled. Preferably, the historical assembly data is big data assembly data or simulation data obtained through simulation; the historical assembly information involves various components and their various types of assembly parameters; Step S22: For each type of assembly parameter, all its historical assembly data are successively distributed into the numerical space of that type, and the effective range is determined based on the distributed numerical space. Preferably, steps S22-25 are performed sequentially for each type of assembly parameter; The determination of the effective interval based on the numerical space it is scattered in specifically means: taking the range of numerical changes of all assembly parameters involved in the historical assembly data as the effective interval; Alternative approach: Extract a portion of the numerical space involved in the assembly parameters from historical assembly data as the effective range; for example, the range of numerical changes in which the frequency of parameter values ​​is negligible can be extracted to improve the effectiveness and efficiency of subsequent nonlinear changes, or a portion of the numerical space corresponding to a specific assembly stage can be extracted to make the training of the risk identification model more targeted. Step S23: Determine the quantization space dimension N of the assembly parameters; calculate the quantization space density. ;in: It is the number of assembly parameters in the historical assembly data within the valid interval; Preferred: N is a preset value, and the quantization space size can be understood as the number of binary bits required to represent this type of assembly parameter; Preferred configuration: N=8, 16; Step S24: Starting from the first assembly parameter in the historical assembly data within the valid interval, construct a quantization interval by taking the cm assembly parameters arranged in the order of assembly parameters. Use the numerical range of the cm assembly parameters contained in this quantization interval as the interval numerical range of this quantization interval, and set the identifier of this quantization interval as the current quantization interval identifier; increment the current quantization interval identifier, and set the (cm+1)th assembly parameter as the first assembly parameter to be set next; repeat this process until all assembly parameters have been processed; linear quantization can uniformly quantize nonlinear, continuous and discrete assembly parameters into a linear space, all parameter types have a certain degree of comparability, and at the same time improve the sensitivity of the risk identification model to dense data; Preferably: the initial value of the first assembly parameter is set to the smallest value among the historical assembly data within the valid interval, and the sorting order is from smallest to largest; the initial value of the quantization interval identifier is set to 0, and the maximum value is... Each increment increases the quantization interval identifier by 1. Step S25: Determine the range of values ​​of the first input parameter, set the value of the first input parameter as the quantization interval identifier corresponding to its range of values; create a first input vector corresponding to the first input parameter using the quantization interval identifier; similarly, set the value of the second input parameter as the quantization interval identifier corresponding to its range of values; create a second input vector corresponding to the second input parameter using the quantization interval identifier. Preferred: When a value is at the boundary of two interval value ranges (e.g., not belonging to either interval value range), calculate the distance between the value and the mean of the cm assembly parameters contained in the two interval value ranges respectively, and take the quantized interval identifier of the interval value range closest to it as its corresponding interval identifier. Preferred: When a value is located at the boundary of two interval value ranges (e.g., not belonging to either interval value range), the quantized interval identifier of the interval value range corresponding to the boundary value of the value range and the interval value range corresponding to the closest one is used as the interval identifier corresponding to the value. Step S3: Input the first input vector and the second input vector into the risk identification model to obtain the risk identification result; Preferably, the risk identification model is a neural network model, a machine learning model, or an artificial intelligence model based on a temporal graph convolutional network; Preferably, the classification and identification results are coarse classification results, including three types: normal state, slight risk, and severe risk; wherein: the normal state indicates that the residual is <5% and the stress distribution is in line with expectations; the slight risk indicates that there is a local interference fit and the pressing speed needs to be adjusted; the severe risk indicates that the probability of sealing surface warping is >90%, triggering an emergency shutdown; Alternative: The classification and identification results are fine classification results, including 6 types: normal state, almost inevitable, relatively likely to occur, rarely occurs, low probability, very low probability, and almost impossible to occur. Preferably, the first input vector and the second input vector have different input weights; the first weight of the first input vector and the second weight of the second input vector are dynamically valued according to different assembly stages; specifically: in the first stage of assembly, the weight value of the first input vector is set higher than that of the second input vector; in the second stage of assembly, the second weight is set higher than the first weight; in the third stage of assembly, the first weight is set equal to the second weight; for example: in the rough assembly stage, the first weight is set higher than the second weight, such as setting the first weight to 0.7 and the second weight to 1 - the first weight = 0.3, so that more attention is paid to the basic positioning accuracy in the first stage; in the precision fitting stage, the second weight is set higher than the first weight value, such as setting the second weight to 0.6 and the first weight to 0.4, taking into account the expected impact of environmental information such as temperature; in the final fastening stage, the first weight is set equal to the second weight for comprehensive evaluation; Preferred approach: In the first to third stages, the first and second weights are dynamically adjusted based on the monitored emergencies. Specifically: When facing emergencies such as increased assembly speed or short-term risks (sudden temperature increases, bolt overload), the first weight is increased to a higher value, for example, 0.8, to address potential emergencies; when the pump enters the high-temperature pressure test, the first weight is increased to a higher value, for example, 0.8; while when the risk decreases (e.g., the impact of short-term fluctuations in the real-time temperature sensor decreases), the first weight is decreased, for example, from 0.5 to 0.35; similarly, when considering the expected increase in risk (e.g., the increased impact of the coefficient of thermal expansion on sealing performance), the second weight is increased, for example, from 0.5 to 0.65. Preferably, the first input vector and the second input vector are concatenated to obtain a concatenated vector, which is then input into the risk identification model to obtain the classification result. Further, the concatenation method can be direct concatenation or weighted concatenation. In direct concatenation, different attention weights are introduced for the first input vector portion and the second input vector portion in the concatenated vector to adjust the distribution of the calculated attention. For example, the attention weight for the first input vector portion is set to be positively correlated with the first weight, while the attention weight for the second input vector portion is positively correlated with the second weight. Preferably, the artificial intelligence model is a graph neural network model based on graph attention mechanism; the nodes in this model are elements in the first input vector and the second input vector; and the edges are the relationships between elements; the relationships can be the numerical linkage relationships between elements or the relationships of jointly experiencing anomalies; it is suitable for graph modeling of component behavior in high temperature and high pressure environments; Preferably, the graph neural network model is a convolutional neural network model; a fully connected layer structure is set to perform convolution operations on the concatenated vector of the first and second input vectors; in this way, during the processing of the input vectors, the information contained in the first and second input vectors is fused and considered, which is different from the method of considering the first and second input vectors separately. The method of considering them separately is no different from inputting them separately, and the impact on the classification result is no different from simple superposition. However, the input vector fusion before processing can discover the risk information expressed by the correlation between the two during the processing, thereby improving the risk identification efficiency; for the risk identification scenario of high temperature and high pressure pump assembly, in conjunction with the fusion dual-input model, through feature deep fusion of the embedding layer, the implicit correlation between real-time assembly parameters and prediction parameters is captured; compared with the traditional dual-branch independent processing structure (such as early fusion / late fusion), an optimized balance can be achieved between feature interaction granularity and computational efficiency; Preferred approach: Use 3 layers of GCN in parallel, using single-hop neighbors to capture local features, two-hop neighbors to capture mid-range features, and full-map neighbors to capture global features respectively; Example 1 Using 12,000 sets of assembly data provided by a pump manufacturer, including 5 types of high-temperature and high-pressure pumps, three types of models were constructed. The samples were divided into a training set of 8,400 sets, a validation set of 2,400 sets, and a test set of 1,200 sets. The test results are shown in the table below.

[0021] Based on the same inventive concept, the present invention also provides a risk identification system for precise assembly of high-temperature and high-pressure pumps based on digital twins. The system is used to complete the above-mentioned risk identification method for precise assembly of high-temperature and high-pressure pumps based on digital twins. The system includes: a data acquisition terminal and a risk identification server; the data acquisition terminal and the risk identification server are connected in communication. Preferably, the communication connection method is a wireless communication connection method; Preferably, there are multiple acquisition terminals used for acquiring assembly parameters; the assembly parameters include one or more of the following: type, size, shape, structure, material, position, pressure, temperature, deformation, contact stress distribution, gap / interference fit, coaxiality, temperature relationship, pressure relationship, vibration spectrum of component gap, etc.; and are acquired through embedded sensors, such as strain gauges, laser displacement sensors, temperature / pressure sensors, ultrasonic testing devices, and vision inspection systems. The system also includes a big data server for storing historical assembly data; the risk identification server obtains historical assembly data by accessing the big data server; and the historical assembly data comes from various types of users, including merchants' self-test data. Based on the same inventive concept, the present invention also provides a risk identification server for precise assembly of high-temperature and high-pressure pumps based on digital twins, the server being used to complete the above-mentioned risk identification method for precise assembly of high-temperature and high-pressure pumps based on digital twins. A computer program (also referred to as a program, software, software application, script, or code) can be written in any form of programming language, including assembly or interpreted languages, declarative or procedural languages, and can be deployed in any form, including as a standalone program or as a module, component, subroutine, object, or other unit suitable for use in a computing environment. A computer program may, but does not necessarily, correspond to a file in a file system. A program can be stored as part of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to said program, or in multiple co-located files (e.g., a file storing one or more modules, subroutines, or code portions). A computer program can be deployed to execute on a single computer or on multiple computers located at a single site or distributed across multiple sites and interconnected by a communications network.

[0022] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0023] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0024] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0025] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0026] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for risk identification in the precise assembly of high-temperature and high-pressure pumps based on digital twins, characterized in that, The method includes: Step S1: Obtain the real-time assembly parameters of each assembly component at the current assembly site; The first input parameters are constructed based on real-time assembly parameters; A digital twin model is used to synchronize real-time assembly parameters, and the assembly process is simulated based on the digital twin model. Predicted assembly parameters are obtained based on the simulation results. A second input parameter is constructed based on the predicted assembly parameters. The step of obtaining the predicted assembly parameters from the simulation results is as follows: based on the simulation results, the assembly trajectory is obtained, and the assembly parameters corresponding to the next assembly site relative to the current assembly site in the assembly trajectory are used as the predicted assembly parameters. Step S2: Linearly quantize the first input parameter and the second input parameter to obtain the first input vector and the second input vector respectively; weight and concatenate the first input vector and the second input vector to obtain the concatenated vector, which is then input into the risk identification model to obtain the classification result; Step S3: Input the first input vector and the second input vector into the risk identification model to obtain the risk identification result; The first input vector and the second input vector are dynamically weighted. The first weight of the first input vector and the second weight of the second input vector are dynamically assigned according to different assembly stages. Specifically, in the first stage of assembly, the weight of the first input vector is set to be higher than that of the second input vector; in the second stage of assembly, the second weight is set to be higher than that of the first weight; and in the third stage of assembly, the first weight is set to be equal to that of the second weight.

2. The method for risk identification in precise assembly of high-temperature and high-pressure pumps based on digital twins according to claim 1, characterized in that, The step of obtaining the predicted assembly parameters from the simulation results involves: obtaining multiple optional assembly trajectories based on the simulation results; selecting one of the multiple optional assembly trajectories as the predicted assembly trajectory based on the current assembly preference; and using the assembly parameters corresponding to the next assembly site relative to the current assembly site in the predicted assembly trajectory as the predicted assembly parameters.

3. The method for risk identification in precise assembly of high-temperature and high-pressure pumps based on digital twins according to claim 2, characterized in that, Assembly preference refers to the current assembly system and / or assembly operators' preference for selecting assembly trajectories.

4. The method for risk identification in precise assembly of high-temperature and high-pressure pumps based on digital twins according to claim 3, characterized in that, Create a visual view interface and use a real-time rendering engine to visualize the components to be assembled in key assembly areas of the assembly process when dynamically calibrating model parameters in the digital twin model using real-time assembly parameters.

5. The method for risk identification in precise assembly of high-temperature and high-pressure pumps based on digital twins according to claim 4, characterized in that, The risk identification model is an artificial intelligence model.

6. The method for risk identification in precise assembly of high-temperature and high-pressure pumps based on digital twins according to claim 5, characterized in that, The artificial intelligence model is a neural network model, a machine learning model, or a temporal graph convolutional network.

7. The method for risk identification in precise assembly of high-temperature and high-pressure pumps based on digital twins according to claim 6, characterized in that, The classification results are coarse classification results, including three categories: normal state, slight risk, and severe risk.

8. The method for risk identification in precise assembly of high-temperature and high-pressure pumps based on digital twins according to claim 7, characterized in that, The artificial intelligence model is a graph neural network model based on the graph attention mechanism.

9. The method for risk identification in precise assembly of high-temperature and high-pressure pumps based on digital twins according to claim 8, characterized in that, A 3-layer graph neural network model is used in parallel.

10. A risk identification system for precise assembly of high-temperature and high-pressure pumps based on digital twins, characterized in that, The digital twin-based high-temperature and high-pressure pump precision assembly risk identification system is used to implement the method described in any one of claims 1-9.

Images