Apparatus for predicting transient state of pipe of nuclear reactor based on machine learning, method thereof and computer-readable recording medium
The machine learning-based reactor piping transient state prediction device addresses the inefficiencies of traditional finite element analysis by using actual and historical data to rapidly predict piping temperature and stress, enhancing the evaluation of nuclear power system safety and efficiency.
Patent Information
- Application Number
- KR1020250058100
- Authority / Receiving Office
- KR · KR
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-05-02
- Publication Date
- 2026-11-10
AI Technical Summary
Existing finite element analysis techniques for evaluating reactor piping integrity during transient states in nuclear power systems are time-consuming and labor-intensive, failing to comprehensively assess variables like piping system layout, flow rate, and support locations.
A machine learning-based reactor piping transient state prediction device that utilizes actual data and past finite element analysis data to predict piping temperature distribution and thermal stress, incorporating a data collection unit, storage unit, and a machine learning model learning unit to generate rapid predictions.
This approach significantly reduces the time and labor required for analysis by omitting dynamic calculations and preprocessing tasks, enabling efficient and timely evaluation of piping integrity.
Smart Images

Figure PAT00004_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to a machine learning-based reactor piping transient state prediction device, a method thereof, and a computer-readable recording medium. More specifically, it relates to a machine learning-based reactor piping transient state prediction device, a method thereof, and a computer-readable recording medium related to a machine learning algorithm that predicts piping temperature distribution and stress state based on actual data in order to design to prevent damage to equipment during reactor operation and to respond quickly to functional failure of equipment. Background Technology
[0002] During the operation of a nuclear power generation system, a turbulent infiltration phenomenon occurs in which high-temperature fluid flowing in through the main pipe penetrates into the branch pipe during the transient state before reaching the steady state. During this transient state of operation, there is a possibility of inducing thermal loads not considered in the design, such as thermal stratification loads, which could cause deformation of the pipes and damage to the support structures.
[0003] Consequently, finite element analysis techniques have traditionally been applied to evaluate the integrity of piping. A procedure was required to assess damage by visualizing the flow field within the pipes using computational fluid dynamics and calculating the pipe stress states caused by thermal flow through fluid-structure coupling analysis techniques. However, as a numerical method, it has been problematic to require significant time and labor resources to comprehensively evaluate all variables, such as the piping system layout, flow rate, diameter, and support locations. The problem to be solved
[0004] The problem that the present invention aims to solve is to provide a reactor piping transient state prediction device, a method, and a computer-readable recording medium based on machine learning that predicts piping temperature distribution and thermal stress based on existing finite element analysis data and actual measurement data in order to improve the economic efficiency of the piping integrity evaluation and safety design process. means of solving the problem
[0005] According to one embodiment of the present invention for solving the above technical problem, a machine learning-based reactor piping transient state prediction device comprises: a data collection unit that receives actual data and past finite element analysis data related to the operation of a nuclear power generation system; a storage unit that stores the received actual data and past finite element analysis data; and a machine learning model learning unit that learns a machine learning model for predicting pipe temperature distribution and thermal stress based on the received actual data and past finite element analysis data to generate pipe temperature distribution and thermal stress prediction data.
[0006] It is preferable that the above measured data include at least two of the following: pipe temperature and pressure data measured during inspection while the nuclear power plant is in operation, flow rate and fluid temperature which are operating conditions of the nuclear power plant, pipe shape information such as pipe diameter, pipe layout shape, direction of branch pipes, and convective heat transfer coefficient of the insulation material.
[0007] According to another embodiment of the present invention for solving the above technical problem, a method for predicting the transient state of a reactor piping based on machine learning comprises: receiving actual data and past finite element analysis data related to the operation of a nuclear power generation system; storing the received actual data and past finite element analysis data; and training a machine learning model for predicting pipe temperature distribution and thermal stress based on the received actual data and past finite element analysis data to generate pipe temperature distribution and thermal stress prediction data.
[0008] According to another embodiment of the present invention for solving the above technical problem, a machine learning-based method for predicting a transient state of a reactor piping in a computer-readable recording medium comprising a program for executing a machine learning-based method for predicting a transient state of a reactor piping in a computer-readable recording medium comprises: a step of receiving actual data and past finite element analysis data related to the operation of a nuclear power generation system; a step of storing the received actual data and past finite element analysis data; and a step of generating pipe temperature distribution and thermal stress prediction data by training a machine learning model for predicting pipe temperature distribution and thermal stress based on the received actual data and past finite element analysis data. Effects of the invention
[0009] Since dynamic analysis, such as finite element analysis or computational fluid analysis, requires a vast amount of time to calculate governing equations depending on the number of meshes and time increment settings, omitting this calculation process according to the present invention allows for the rapid acquisition of a state and the saving of time resources.
[0010] In addition, according to the present invention, analysis preprocessing tasks such as modeling by pipe layout shape and generating a grid can be omitted, thereby providing a labor-saving effect. Brief explanation of the drawing
[0011] FIG. 1 is a block diagram illustrating a machine learning-based reactor piping transient state prediction device according to one embodiment of the present invention. FIG. 2 is a diagram showing a flowchart regarding a method for constructing a finite element analysis database according to an embodiment of the present invention. FIG. 3 is a diagram showing a schematic diagram of the operation of a recurrent neural network used in the present invention. FIG. 4 is a flowchart illustrating a method for predicting a reactor piping transient state based on machine learning according to an embodiment of the present invention. Specific details for implementing the invention
[0012] Preferred embodiments of the present invention will be described in detail with reference to the attached drawings.
[0013] In describing the embodiments of this specification, if it is determined that a detailed description of related known configurations or functions could obscure the gist of this specification, such detailed description is omitted.
[0014] In the present invention, the description of a specific configuration as "including" does not exclude configurations other than said configuration, and means that additional configurations may be included within the scope of the practice of the present invention or the technical concept of the present invention.
[0015] Furthermore, the components shown in the embodiments of the present invention are depicted independently to represent different characteristic functions and do not imply that each component consists of separate hardware or a single software unit. That is, each component is listed and included as a separate component for convenience of explanation; however, at least two of the components may be combined to form a single component, or a single component may be divided into multiple components to perform a function, and such integrated and separated embodiments of each component are included within the scope of the present invention as long as they do not deviate from the essence of the invention.
[0016] Furthermore, some components may not be essential components performing an essential function in the present invention, but merely optional components for enhancing performance. The present invention may be implemented by including only the components essential for realizing the essence of the present invention, excluding components used solely for performance enhancement, and a structure including only the essential components, excluding optional components used solely for performance enhancement, is also included within the scope of the rights of the present invention.
[0017] In addition, components expressed as '~part (unit)', 'module', etc. in this specification may consist of two or more components combined into a single component, or a single component may be divided into two or more components according to more detailed functions. Furthermore, each component described below may additionally perform some or all of the functions performed by other components in addition to the primary function it is responsible for, and it goes without saying that some of the primary functions performed by each component may be exclusively performed by other components.
[0018] In addition, detailed descriptions of known functions and configurations that may obscure the essence of the invention will be omitted in this specification.
[0019] In this specification, machine learning, machine learning, artificial intelligence (AI), and deep learning are treated as the same concept.
[0020] FIG. 1 is a block diagram showing a reactor piping transient state prediction device based on machine learning according to one embodiment of the present invention (hereinafter referred to as the "piping transient state prediction device").
[0021] Referring to FIG. 1, the pipe transient state prediction device (100) includes a data collection unit (110), a preprocessing unit (120), a storage unit (130), a machine learning model learning unit (140), a control unit (150), and a finite element analysis database (160).
[0022] The data collection unit (110) receives actual data related to the operation of the nuclear power generation system. The actual data includes at least two of the following: pipe temperature and pressure data measured during inspection while the nuclear power plant is in operation, flow rate and fluid temperature which are operating conditions of the nuclear power plant, pipe shape information such as pipe diameter, pipe arrangement shape, direction of branch pipes, and convective heat transfer coefficient of the insulation material.
[0023] In nuclear power generation, insulation is a critical component primarily associated with the reactor. It plays a role in maintaining heat generated in high-temperature environments, enhancing reactor efficiency, and preventing heat from escaping to the outside. In nuclear power plants, massive amounts of heat are generated through nuclear reactions inside the reactor; therefore, efficiently transferring and managing this heat is essential, and insulation helps ensure that the generated heat is utilized efficiently within the reactor. Insulation is generally composed of materials that possess properties for absorbing, transferring, or blocking heat.
[0024] Additionally, the data collection unit (110) can receive previously performed finite element analysis data from the finite element analysis database (160). The previously performed finite element analysis data consists of past pipe temperature distribution and thermal stress data. The previously performed finite element analysis data may also include actual measurement data that served as the basis for the analysis of pipe temperature distribution and thermal stress data.
[0025] Finite element analysis (FEA) refers to a numerical analysis technique used to predict the physical behavior of complex structures or systems.
[0026] FIG. 2 is a diagram showing a flowchart regarding a method for constructing a finite element analysis database according to an embodiment of the present invention.
[0027] Referring to FIG. 2, in step 210, the pipe transient state prediction device (100) defines the geometric shape for the reactor coolant system.
[0028] In step 220, the pipe transient state prediction device (100) performs a sensitivity analysis.
[0029] A pipe transient state prediction device (100) evaluates the effect of geometric shape conditions (input parameters) on analysis results such as stress. Sensitivity analysis is an analysis technique that identifies the sensitivity of a system or structure by analyzing how much a change in conditions (input parameters) affects the analysis results.
[0030] In one embodiment of the present invention, a pipe transient state prediction device (100) defines the mesh size and time step in finite element analysis and performs conjugate heat transfer analysis by analyzing the phenomenon of heat transfer between a solid and a fluid through a turbulence model that numerically models and analyzes turbulence phenomena, which are irregular and complex movements occurring in fluid flow.
[0031] In step 230, the pipe transient state prediction device (100) performs thermal flow analysis.
[0032] Thermal Flow Analysis is a process of analyzing heat transfer related to fluid flow, and is used to determine how heat is transferred as a fluid flows, how that heat is affected by structures or other fluids, and how the temperature distribution changes.
[0033] In one embodiment of the present invention, a pipe transient state prediction device (100) performs heat flow analysis through the temperature distribution within the branch pipe.
[0034] In step 240, the pipe transient state prediction device (100) maps the data.
[0035] In one embodiment of the present invention, temperature data in the form of CFX Output is mapped to ANSYS input data.
[0036] In step 250, the pipe transient state prediction device (100) performs thermal stress analysis.
[0037] Thermal stress analysis is a method for analyzing the effects of thermal expansion or contraction caused by temperature changes on an object, and it is the process of calculating the stress generated when an object deforms due to temperature changes.
[0038] In one embodiment of the present invention, a pipe transient state prediction device (100) generates thermal stress analysis data by performing thermal stress analysis through CFD-FE Coupled Analysis.
[0039] CFD-FE Coupled Analysis is a method of analysis that combines Computational Fluid Dynamics (CFD) and Finite Element Analysis (FEA). This analysis provides a more realistic simulation by simultaneously considering the interaction between fluid flow (CFD) and structural analysis (FEA).
[0040] Thermal stress analysis data includes pipe temperature distribution and thermal stress data.
[0041] In step 260, the pipe transient state prediction device (100) stores thermal stress analysis data in the finite element analysis database (160). Additionally, the pipe transient state prediction device (100) may store the data from each step and the actual measurement data that formed the basis of the thermal stress analysis in the finite element analysis database (160) in conjunction with the thermal stress analysis data.
[0042] In another embodiment of the present invention, the pipe transient state prediction device (100) may receive thermal stress analysis data stored in a predetermined storage and store it in a finite element analysis database (160).
[0043] In addition, thermal stress analysis data may be generated by a separate analysis device (not shown) rather than the pipe transient state prediction device (100).
[0044] The preprocessing unit (120) can remove outliers and noise and convert the data into a learnable form by preprocessing the actual measurement data and past finite element analysis data from the data collection unit (110).
[0045] The actual measurement data and past finite element analysis data received by the data collection unit (110) can be stored in the storage unit (130). Additionally, the actual measurement data and past finite element analysis data preprocessed by the data collection unit (110) can also be stored in the storage unit (130). The storage unit (130) is a concept that includes at least one database.
[0046] The machine learning model learning unit (140) learns a machine learning model generated according to the purpose using input data.
[0047] The machine learning model according to the present invention is a machine learning model for predicting the pipe temperature distribution and thermal stress of a nuclear power generation system. The machine learning model can be generated according to the task based on actual measurement data and past finite element analysis data.
[0048] The machine learning model according to the present invention may be generated in the pipe transient state prediction device (100) of the present invention, or may be received from an external source and stored in a storage unit (130).
[0049] In the present invention, the machine learning model learning unit (140) learns actual measurement data and past finite element analysis data using a machine learning model for predicting the pipe temperature distribution and thermal stress of a nuclear power generation system, and predicts the pipe temperature distribution and thermal stress. Through learning, the machine learning model learning unit (140) generates pipe temperature distribution and thermal stress prediction data.
[0050] Machine learning algorithms include Convolutional Neural Networks (CNN), Deep Neural Networks (DNN), Radial Basis Function Neural Networks (RBFN), Recurrent Neural Networks (RNN), Long Short-Term Memory Networks (LSTM), Generative Adversarial Networks (GAN), Autoencoders Deep Networks, Deep Belief Networks, Multilayer Perceptrons (MLP), Self Organizing Maps (SOM), Restricted Boltzmann Machines (RBM), Feedforward Neural Networks (FNN), and Deep Q-Networks (DQN).
[0051] To briefly summarize, Convolutional Neural Networks (CNNs) function as a specific form of artificial intelligence primarily used to understand visual data such as images and videos. Deep Neural Networks (DNNs) are artificial neural networks with multiple hidden layers. Radial Basis Function Networks (RBFNs) are a distinct subset of artificial neural networks that operate by using radial basis functions as activation functions within the hidden layers. Recurrent Neural Networks (RNNs) constitute a specific type of neural network architecture specifically designed to process sequential data by retaining information over time through loops within the network. Long Short-Term Memory Networks (LSTMs) are a sophisticated type of recurrent neural network architecture designed to address the vanished gradient problem of traditional RNNs. Generative Adversarial Networks (GANs) represent a powerful class of machine learning models composed of two interconnected networks called generators and discriminators. Autoencoder deep learning algorithms are a type of deep learning algorithm used for unsupervised learning tasks that operates by compressing input data into latent or compressed representations and then reconstructing the original data as accurately as possible. Deep Belief Networks are complex neural networks composed of multiple layers of probabilistic models. Multilayer Perceptrons (MLPs) are a basic type of artificial neural network composed of multiple layers of interconnected neurons. Self-Organizing Maps (SOMs) are artificial neural networks used for unsupervised learning and pattern recognition. Restricted Boltzmann Machines (RBMs) are a specific type of neural network composed of two layers: a visible layer and a hidden layer. Feedforward Neural Networks (FNNs) are a basic type of artificial neural network in which information flows unidirectionally from the input layer through the hidden layer to the output layer without cycles. Deep Q-Networks (DQNs) are a type of deep reinforcement learning algorithm that combines deep neural networks and Q-learning, enabling machines to learn optimal behavior in complex environments.
[0052] In one embodiment of the present invention, a machine learning algorithm used for learning may be a recurrent neural network (RNN). A recurrent neural network (RNN) is an algorithm that has the advantage of processing sequence data of various lengths and capturing temporal dependencies.
[0053] FIG. 3 is a diagram showing a schematic diagram of the operation of a recurrent neural network used in the present invention.
[0054] Figure 3 illustrates an input layer (Input Layer, 310) for receiving data, a hidden layer (Hidden Layer, 320) located between the input layer (310) and the output layer (330) for extracting and learning features of the input data, and an output layer (Output Layer, 330) for producing a final result.
[0055] The input data includes actual data such as flow rate, fluid temperature, convective heat transfer coefficient, pipe diameter, and branch pipe direction, as well as past finite element analysis data. The output data includes pipe temperature distribution and maximum thermal stress values. Although the input data is shown as above in Figure 3, all of the actual data mentioned above may be included.
[0056] Although FIG. 3 illustrates a recurrent neural network, this is merely an example, and there are no limitations on the machine learning algorithms used for learning according to the present invention.
[0057] The control unit (150) embeds the machine learning model that has completed training into the platform, stores the machine learning model that has completed training in the storage unit (130), and manages it. In addition, the control unit (150) also stores data learned using the machine learning model in the storage unit (130) and manages it.
[0058] The pipe transient state prediction device (100) may also store data on a cloud basis.
[0059] Additionally, although not shown in the drawings, a machine learning model performance measurement unit (not shown) measures the performance of the completed machine learning model. If the performance of the completed machine learning model is below a threshold, the machine learning model may be replaced or regenerated.
[0060] If the performance of a machine learning model decreases during monitoring by a machine learning model monitoring unit (not shown), the control unit (150) may retrain the machine learning models through a machine learning model learning unit (140), measure the performance of the learned machine learning model through a machine learning model performance measurement unit, and update the modularized machine learning model through a machine learning model update unit (not shown).
[0061] Additionally, the control unit (150) may periodically retrain machine learning models through the machine learning model learning unit (140), measure the performance of the trained machine learning models through the machine learning model performance measurement unit, and update the machine learning models through the machine learning model update unit.
[0062] FIG. 4 is a flowchart illustrating a method for predicting a reactor piping transient state based on machine learning according to an embodiment of the present invention.
[0063] Referring to FIG. 4, in step 410, a device for performing a machine learning-based reactor piping transient state prediction method (hereinafter referred to as the "reactor piping transient state prediction device") receives actual data related to the operation of a nuclear power generation system and previously performed finite element analysis data.
[0064] The measured data includes at least two of the following: pipe temperature and pressure data measured during inspection while the nuclear power plant is in operation, flow rate and fluid temperature which are operating conditions of the nuclear power plant, pipe shape information such as pipe diameter, pipe layout shape, direction of branch pipes, and convective heat transfer coefficient of the insulation material.
[0065] In nuclear power generation, insulation is a critical component primarily associated with the reactor. It plays a role in maintaining heat generated in high-temperature environments, enhancing reactor efficiency, and preventing heat from escaping to the outside. In nuclear power plants, massive amounts of heat are generated through nuclear reactions inside the reactor; therefore, efficiently transferring and managing this heat is essential, and insulation helps ensure that the generated heat is utilized efficiently within the reactor. Insulation is generally composed of materials that possess properties for absorbing, transferring, or blocking heat.
[0066] Previously performed finite element analysis data can be received from a finite element analysis database. The previously performed finite element analysis data consists of historical pipe temperature distribution and thermal stress data. The previously performed finite element analysis data may also include actual measurement data that served as the basis for the analysis of pipe temperature distribution and thermal stress data.
[0067] As the method for constructing a finite element analysis database has been described in Figure 2 above, it will be omitted to avoid redundancy.
[0068] Finite element analysis (FEA) refers to a numerical analysis technique used to predict the physical behavior of complex structures or systems.
[0069] In step 420, the pipe transient state prediction device preprocesses the received actual data and historical finite element analysis data. The pipe transient state prediction device removes outliers and noise from the received actual data and historical finite element analysis data and converts them into a learnable form.
[0070] Step 420 is shown in dashed lines, indicating that it is an additional step to the general form of the method.
[0071] In step 430, the pipe transient state prediction device stores received actual data and historical finite element analysis data. The pipe transient state prediction device may also store preprocessed actual data and historical finite element analysis data. The pipe transient state prediction device includes at least one database.
[0072] In step 440, the pipe transient state prediction device learns a machine learning model generated according to the purpose using the input data.
[0073] The machine learning model according to the present invention is a machine learning model for predicting the pipe temperature distribution and thermal stress of a nuclear power generation system. The machine learning model can be generated according to the task based on actual measurement data and past finite element analysis data.
[0074] The machine learning model according to the present invention may be generated in the pipe transient state prediction device of the present invention, or may be received from an external source and stored in the pipe transient state prediction device.
[0075] Specifically, the pipe transient state prediction device learns from actual data and past finite element analysis data using a machine learning model for predicting pipe temperature distribution and thermal stress in a nuclear power generation system, and predicts pipe temperature distribution and thermal stress. Through learning, the pipe transient state prediction device generates pipe temperature distribution and thermal stress prediction data.
[0076] As machine learning algorithms have been described above, they will be omitted to avoid redundancy.
[0077] The pipe transient state prediction device may embed a trained machine learning model into the platform. The pipe transient state prediction device may store the trained machine learning model in a designated repository of the platform and manage it. In addition, the pipe transient state prediction device may separately store training data learned using the machine learning model in a designated repository and manage it.
[0078] The pipe transient state prediction device can also store data on a cloud basis.
[0079] In addition, although not shown in the drawings, the pipe transient state prediction device may also measure the performance of the machine learning model after training. If the performance of the machine learning model after training falls below a threshold, the machine learning model may be replaced or updated.
[0080] If the performance of a machine learning model deteriorates during monitoring by the pipe transient state prediction device, the pipe transient state prediction device may retrain the machine learning models, measure the performance of the trained machine learning models, and update the machine learning models.
[0081] In addition, the pipe transient state prediction device may periodically retrain machine learning models through machine learning model training, measure the performance of the trained machine learning models, and update the machine learning models through the machine learning model update unit.
[0082] Meanwhile, the embodiments of the present disclosure described above can be written as a program that can be executed on a computer, and the written program can be stored on a medium.
[0083] A medium may be a medium that continuously stores programs executable by a computer, or temporarily stores them for execution or download. Additionally, a medium may be a variety of recording or storage means in the form of a single or multiple pieces of hardware combined, and is not limited to a medium directly connected to a computer system but may exist distributed over a network. Examples of media may include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and media configured to store program instructions, including ROM, RAM, and flash memory. Furthermore, other examples of media may include recording or storage media managed by app stores that distribute applications, or sites and servers that supply or distribute various other software. Explanation of the symbols
[0084] Piping transient state prediction device: 100 Data Collection Unit: 110 Preprocessing section: 120 Storage: 130 Machine learning model training section: 140 Control unit: 150 Finite Element Analysis Database: 160
Claims
Claim 1 A machine learning-based reactor piping transient state prediction device characterized by comprising: a data collection unit for receiving actual measurement data and past finite element analysis data related to the operation of a nuclear power generation system; a storage unit for storing the received actual measurement data and past finite element analysis data; and a machine learning model learning unit for learning a machine learning model for predicting pipe temperature distribution and thermal stress based on the received actual measurement data and past finite element analysis data to generate pipe temperature distribution and thermal stress prediction data. Claim 2 A machine learning-based reactor piping transient state prediction device according to claim 1, characterized in that the actual measurement data includes at least two of the following: pipe temperature and pressure data measured during inspection while the nuclear power plant is operating, flow rate and fluid temperature which are operating conditions of the nuclear power plant, pipe shape information such as pipe diameter, pipe arrangement shape, direction of branch pipes, and convective heat transfer coefficient of insulation material. Claim 3 A machine learning-based method for predicting transient states of reactor piping, characterized by comprising: a step of receiving actual data and historical finite element analysis data related to the operation of a nuclear power generation system; a step of storing the received actual data and historical finite element analysis data; and a step of training a machine learning model for predicting pipe temperature distribution and thermal stress based on the received actual data and historical finite element analysis data to generate pipe temperature distribution and thermal stress prediction data. Claim 4 A method for predicting a transient state of a reactor piping based on machine learning according to claim 3, characterized in that the actual measurement data includes at least two of the following: pipe temperature and pressure data measured during inspection while the nuclear power plant is in operation, pipe shape information such as flow rate and fluid temperature which are operating conditions of the nuclear power plant, pipe diameter, pipe arrangement shape, direction of branch pipes, and convective heat transfer coefficient of the insulation material. Claim 5 A computer-readable recording medium comprising a program for executing a method for predicting a transient state of a reactor piping based on machine learning, wherein the method for predicting a transient state of a reactor piping based on machine learning comprises: a step of receiving actual data and past finite element analysis data related to the operation of a nuclear power generation system; a step of storing the received actual data and past finite element analysis data; and a step of training a machine learning model for predicting pipe temperature distribution and thermal stress based on the received actual data and past finite element analysis data to generate pipe temperature distribution and thermal stress prediction data.
