KR-102961218-B1 - Apparatus for updating a digital twin-based model and method therefor

KR102961218B1KR 102961218 B1KR102961218 B1KR 102961218B1KR-102961218-B1

Abstract

A method for updating a model comprises: a measurement unit collecting sensor data representing the measured physical quantities of a component of a gas turbine unit measured through a sensor during the operation of the gas turbine unit; an analysis unit analyzing the sensor data using a reduction-order model generated using training data derived through numerical analysis to derive state data indicating whether the state of the component is normal or abnormal; a verification unit performing inference on the sensor data using a verification model generated through training data to derive a state vector representing the state of the component as a probability; if the state data and the state vector are different, the verification unit updating the training data using the state vector corresponding to the sensor data; and a physical model generation unit updating the reduction-order model using the updated training data.

Inventors

이승민
박누가
황범철

Assignees

두산에너빌리티 주식회사

Dates

Publication Date: 20260507
Application Date: 20231006

Claims (20)

A step in which the measuring unit collects sensor data representing the measured physical quantity of a component of the gas turbine device measured through a sensor during the operation of the gas turbine device; A step in which the analysis unit analyzes the sensor data using a reduction-order model generated using training data derived through numerical analysis to derive state data indicating whether the state of the component is normal or abnormal; A step in which a verification unit performs inference on the sensor data through a verification model generated from training data to derive a state vector representing the state of the component as a probability; and A step of comparing the state data and the state vector, and if the state data and the state vector are different, the verification unit updates the training data using the state vector corresponding to the sensor data; A step in which a physical model generation unit updates the order reduction model using the updated training data; Characterized by including Method for updating the model.
In paragraph 1, Before the step of collecting the above sensor data, A step in which the physical model generation unit generates a reduction-order model that derives state data representing the state of a component from the measured physical quantity of the component of a gas turbine device using training data derived through numerical analysis; Characterized by further including Method for updating the model.
In paragraph 2, The step of generating the above-mentioned order reduction model is A step in which the physical model generation unit generates a virtual part that simulates the part through 3D modeling; A step in which the physical model generation unit derives sensor data representing the measured physical quantities of the virtual components according to the operating conditions of the gas turbine device through numerical analysis; A step in which the physical model generation unit derives state data representing the state of the component according to the operating conditions through numerical analysis; The step of the physical model generation unit mapping sensor data and state data according to the driving conditions to configure training data; and The above physical model generation unit generates a reduction-order model that derives state data from the sensor data based on the training data; Characterized by including Method for updating the model.
In paragraph 1, Before the step of collecting the above sensor data, A step in which a learning model generation unit generates a verification model that derives a state vector representing the state of the component from the sensor data using learning data acquired during the operation of the gas turbine device; Characterized by further including Method for updating the model.
In paragraph 4, The step of generating the above verification model is The above learning model generation unit prepares learning data including sensor data representing a measured physical quantity of the component measured through a sensor during the operation of the above gas turbine device, and a label representing the state of the component corresponding to the sensor data; The step of the above-mentioned learning model generation unit inputting sensor data into a verification model; A step in which the verification model performs inference on the sensor data to derive a state vector; A step in which the above learning model generation unit calculates a loss representing the difference between the state vector and the label; A step in which the learning model generation unit performs optimization to update the weights of the verification model so that the loss is minimized; Characterized by including Method for updating the model.
In paragraph 5, The above label is It is a vector indicating whether the state of the above component is normal or abnormal, and The above state vector is Characterized as a probability corresponding to each of the above-mentioned normal state and above-mentioned abnormal state Method for updating the model.
In paragraph 5, The step of preparing the above-mentioned training data is The above learning model generation unit continuously collects the operating conditions, sensor data, and output of the gas turbine device during the actual operation of the gas turbine device; and A step in which the learning model generation unit, in response to the operating conditions and sensor data, assigns a normal state label when the output of the gas turbine device is a normal value relative to the specifications of the gas turbine device, and assigns an abnormal state label when the output of the gas turbine device differs from the normal value relative to the specifications of the gas turbine device by more than a predetermined value; Characterized by including Method for updating the model.
In paragraph 1, After the step of deriving state data indicating the state of the above-mentioned component, Before the step of deriving a state vector representing the state of the above component, A step in which a display unit visualizes the physical quantity of the analysis of the above-mentioned part and the state of the above-mentioned part and displays them on a screen; Characterized by further including Method for updating the model.
In paragraph 1, The step of analyzing the sensor data to derive state data indicating the state of the component The above-mentioned order reduction model derives the analysis physical quantity of a part from sensor data representing the measurement physical quantity of the part; and A step in which the above-described order reduction model derives state data indicating whether the state of the part is a normal state or an abnormal state from the analysis physical quantity of the part; Characterized by including Method for updating the model.
In paragraph 1, The step of deriving the above state vector The step of the verification unit inputting sensor data into the verification model; and A step in which the verification model performs multiple operations on the sensor data in which learned weights between multiple layers are applied to derive a state vector representing the probability that the component is in a normal state and the probability that it is in an abnormal state; Characterized by including Method for updating the model.
A measuring unit that collects sensor data representing the measured physical quantity of a component of the gas turbine device measured through a sensor during the operation of the gas turbine device; An analysis unit that derives state data indicating whether the state of the component is normal or abnormal by analyzing the sensor data through a reduction-order model generated using training data derived through numerical analysis; Inference is performed on the sensor data through a validation model generated from the training data to derive a state vector representing the state of the component as a probability, and A verification unit that compares the state data and the state vector, and if the state data and the state vector are different, updates the training data using the state vector corresponding to the sensor data; and A physical model generation unit that updates the order reduction model using the above updated training data; Characterized by including A device for updating the model.
In Paragraph 11, The above physical model generation unit Characterized by generating a reduction-order model that derives state data representing the state of a component from the measured physical quantity of a component of a gas turbine device using training data derived through numerical analysis. A device for updating the model.
In Paragraph 12, The above physical model generation unit A virtual part that simulates the above part is created through 3D modeling, and Sensor data representing the measured physical quantities of the virtual component according to the operating conditions of the gas turbine device are derived through numerical analysis, and State data representing the state of the above parts according to the above operating conditions is derived through numerical analysis, and Training data is configured by mapping sensor data and status data according to the above driving conditions, and Characterized by generating a reduction-order model that derives state data from sensor data based on the training data. A device for updating the model.
In Paragraph 11, A learning model generation unit that generates a verification model for deriving a state vector representing the state of the component from the sensor data using learning data acquired during the operation of the above gas turbine device; Characterized by further including A device for updating the model.
In Paragraph 14, The above learning model generation unit When operating the above gas turbine device, learning data is provided that includes sensor data representing a measured physical quantity of the above component measured through a sensor, and a label representing the state of the above component corresponding to the sensor data. Input sensor data into the verification model, and When the above verification model performs inference on the above sensor data to derive a state vector, Calculate a loss representing the difference between the above state vector and the above label, and Characterized by performing optimization to update the weights of the verification model so that the above loss is minimized. A device for updating the model.
In paragraph 15, The above label is It is a vector indicating whether the state of the above component is normal or abnormal, and The above state vector is Characterized as a probability corresponding to each of the above-mentioned normal state and above-mentioned abnormal state A device for updating the model.
In paragraph 15, The above learning model generation unit During the actual operation of the above gas turbine device, the operating conditions, sensor data, and output of the above gas turbine device are continuously collected, and Characterized by assigning a normal state label when the output of the gas turbine device is at a normal value relative to the specifications of the gas turbine device in response to the above operating conditions and the above sensor data, and assigning an abnormal state label when the output of the gas turbine device differs from the normal value relative to the specifications of the gas turbine device by more than a predetermined value. A device for updating the model.
In Paragraph 11, A display unit that visualizes and displays on a screen the physical quantity of the analysis of the above-mentioned part and the state of the above-mentioned part; Characterized by further including A device for updating the model.
In Paragraph 11, The above order reduction model is Deriving the physical quantity of an analysis of a part from sensor data representing the physical quantity of measurement of the part, and Characterized by deriving state data indicating whether the state of a part is normal or abnormal from the physical quantity analyzed of the part. A device for updating the model.
In Paragraph 11, The above verification model is When the above sensor data is input, the method is characterized by performing multiple operations in which learned weights between multiple layers are applied to the above sensor data to derive a state vector representing the probability that the component is in a normal state and the probability that it is in an abnormal state. A device for updating the model.

Description

Apparatus for updating a digital twin-based model and method therefor The present invention relates to a model update technology, and more specifically, to an apparatus for updating a digital twin-based model and a method for doing the same. Recently, the term "digital twin" is being used in everyday life as well as in various industrial sectors such as ports, transportation, buildings, energy, and shipbuilding. As digital twins are utilized in the construction and operation of smart cities, they are also being exposed to citizens in their daily lives, appearing on billboards in subways and on highways to reach a wider audience. On the surface, a digital twin involves creating a virtual twin of a real-world object and establishing the movements and behaviors of the real object as a role model for the virtual twin, thereby enabling the simulation and modeling of the real world within a virtual environment. Although the concept of a digital twin was first introduced in 2002 and has been used partially in the manufacturing sector, it has recently been gaining prominence across various industries. A digital twin is a technology that enables the virtualization of objects, systems, and environments existing in the real world into the virtual space of a software system, allows the simulation of dynamic motion characteristics and resulting changes of real objects and systems within the software system, applies the optimal state based on the simulation results to the real system, and transmits changes in the real system back to the virtual system, thereby implementing a continuous cyclic adaptation and optimization system. Figure 1 is a diagram illustrating the configuration of a system for updating a digital twin-based model. FIG. 2 is a partially cutaway perspective view of a gas turbine device according to one embodiment of the present invention. FIG. 3 is a cross-sectional view showing the schematic structure of a gas turbine device according to one embodiment of the present invention. FIG. 4 is a diagram illustrating the configuration of a device for updating a digital twin-based model according to an embodiment of the present invention. FIG. 5 is a flowchart illustrating a method for generating an order reduction model using training data according to an embodiment of the present invention. FIG. 6 is a flowchart illustrating a method for generating a verification model using training data according to an embodiment of the present invention. FIG. 7 is a diagram illustrating a method for updating a digital twin-based model according to an embodiment of the present invention. FIG. 8 is a drawing showing a computing device according to an embodiment of the present invention. The present invention is capable of various modifications and may have various embodiments, and specific embodiments are illustrated and described in detail in the detailed description. However, this is not intended to limit the present invention to specific embodiments, and it should be understood that it includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention. The terms used in this invention are used merely to describe specific embodiments and are not intended to limit the invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this invention, terms such as "comprising" or "having" are intended to indicate the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof. First, a system for updating a digital twin-based model according to an embodiment of the present invention will be described. FIG. 1 is a diagram illustrating the configuration of a system for updating a digital twin-based model. FIG. 2 is a partially cutaway perspective view of a gas turbine device according to an embodiment of the present invention, and FIG. 3 is a cross-sectional view showing the schematic structure of a gas turbine device according to an embodiment of the present invention. FIG. 4 is a diagram illustrating the configuration of a device for updating a digital twin-based model according to an embodiment of the present invention. Referring to FIG. 1, a system for updating a digital twin-based model according to an embodiment of the present invention includes a gas turbine device (1000) and a virtualization device (100). Here, the virtualization device (100) is for managing parts of the gas turbine device (1000) using a digital twin. Referring to FIGS. 2 and 3, the gas turbine device (1000) includes a compressor (1100), a combustor (1200), and a turbine (1300). The compressor (1100) is equipped with a plurality of blades (1110) installed radially. The compressor (1100) rotates the blades (1110), and air moves as it i