Fusion for reduced-order model, measurement data and machine learning techniques of digital twin construction method for multi-physical device system

Abstract

The present application aims to provide a digital twin construction method that combines a reduced-order model of a multi-physical system and field data, artificial intelligence technology, real-time monitoring in an industrial field, improved operation, and a response to accidents. The digital twin construction method of the present application is characterized by including: a network definition step of defining a multi-physical device system as a network composed of a combination of elemental devices; an elemental model establishment step of establishing a 0-D model based on a relational expression for individual elemental devices; a system model establishment step of reflecting an additional relational expression by machine learning from a 3-D CAERO model or data for a core elemental device to form a closure of all relational expressions for the system; a system ROM construction step of constructing a system ROM for the system model established in the system model establishment step from a calculation result for a condition sampled in an operating variable parameter space; a system ROM correction step of minimizing an error between a model prediction value for an elemental device and system and measured data; and a real-time algorithm construction step of constructing an algorithm for grasping a system state expected under a virtual operating condition or an optimal operating condition based on a real-time monitoring result.

Description

Technical Field

[0001] This invention relates to a method for constructing digital twins, and more particularly, to a method for constructing digital twins that integrates field data from the Internet of Things (IoT) and artificial intelligence technologies such as machine learning, based on a 0-dimensional (hereinafter referred to as "0-D") or multi-dimensional (1-D, 2-D, or 3-D) reduced order model (hereinafter referred to as "ROM") for multi-physical device systems. Background Technology

[0002] A digital twin is a virtual twin model of a system consisting of multiple physical devices or multiple components that exist in the real world. When all the conditions that determine the operating state are provided, the digital twin also needs to determine the operating state and the variables of interest, just like in the real world.

[0003] Therefore, digital twins need to accurately grasp all variables of interest (diagnostic model) under current operating conditions, and similarly, predict all variables of interest (predictive model) under virtual operating conditions. Furthermore, the time required for this diagnosis and prediction must be sufficiently short, requiring real-time information delivery to on-site operators.

[0004] The method for constructing a reduced-order model by combining field measurement data and CAE analysis in "Patent Document 1" below is limited to the technology of combining measurement data and three-dimensional distribution as the result of CAE analysis. It has the limitation that it is difficult to extend its applicability to multi-physics systems composed of networks of multiple element devices.

[0005] (Patent Document 1) KR 10-2048243 B1 (2019.11.25) Summary of the Invention

[0006] Technical issues

[0007] The present invention was conceived to solve the problems described above. The purpose of the present invention is to provide a digital twin construction method that combines simulated ROM and field data based on multi-physical device systems with machine learning technology to monitor and improve operation in real time in industrial fields and to respond to accidents.

[0008] Solution to the problem

[0009] The digital twin construction method of the present invention for solving the problems described above is characterized by comprising: a network definition step, defining a multi-physical device system as a network composed of combinations of element devices; an element model establishment step, establishing a 0-D model based on the relations for individual element devices; a system model establishment step, reflecting the additional relations formed by machine learning from the 3-D CAE ROM or data of the core element devices to form a closure of all relations for the system; a system ROM construction step, constructing a system ROM for the system model established in the above system model establishment step from the calculation results of conditions sampled from the parameter space of the operating variables; a system ROM correction step, minimizing the error between the model prediction values ​​and measured data for the element devices and the system; and a real-time algorithm construction step, constructing an algorithm based on real-time monitoring results for mastering the expected system state or optimal operating conditions under virtual operating conditions.

[0010] The aforementioned system ROM correction steps may include one or more of the following steps: a Gappy-POD correction step, deriving the ROM from a matrix formed by all variables of interest obtained under the conditions of sampling in the operating variable parameter space, applying the Gappy-POD method to adjust the principal component coefficients of the ROM in order to minimize the sum of squares of the errors between the predicted and measured values; and an artificial intelligence correction step, when the causal or functional relationship between the error between the predicted value and the data and the operating variables and the predicted physical quantities is unclear, using artificial intelligence techniques such as machine learning through neural network circuits to correct the error based on accumulated data.

[0011] The Gappy-POD method described above can be applied to the correlation Gappy-POD method for adjusting the principal component coefficients of the ROM using measurements of fused heterogeneous species.

[0012] The Gappy-POD calibration steps and the artificial intelligence calibration steps described above can be performed independently or sequentially and simultaneously.

[0013] The above-described system ROM correction steps can assign appropriate weights to each error based on the uncertainty of the measured value or the importance of the main performance indicators while minimizing the error of the predicted value.

[0014] In the aforementioned system ROM calibration steps, when there are periodic or significant changes in equipment operation, automatic calibration can be performed to minimize the error between online measurement data and model predictions in order to maintain the accuracy of the digital twin.

[0015] In the above-mentioned real-time algorithm construction steps, conditions that maximize or minimize predefined performance variables or cost functions in the runtime variable parameter space can be proposed to the operators in real time.

[0016] The effects of the invention

[0017] The digital twin construction method of the present invention has the following excellent effects.

[0018] i) A real-time operating system ROM with fast response characteristics and a digital twin based thereon can be constructed based on 0-D simulation of each element device, 3-DC AE simulation of the core device, and machine learning results from the data.

[0019] ii) By reflecting various online and offline measurement data through Gappy-POD or associated Gappy-POD methods, a 3-D ROM for core equipment calibration or a ROM for the entire system calibration can be constructed, maintaining the reliability and accuracy of the digital twin based on it.

[0020] iii) During the data fusion process, different weighting values ​​can be applied to the variable based on the uncertainty of the measured data or its importance to the final performance variable, thereby minimizing uncertainty.

[0021] iv) Digital twins can be used in both diagnostic and predictive modes to provide optimal operating conditions in real time within permissible operating limits, thus reflecting them in operational automation. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the execution steps of the digital twin construction method of the present invention.

[0023] Figure 2 The diagram illustrates a network defined for a city gas production process system as an example of the present invention.

[0024] Figure 3 To indicate the composition Figure 2 A diagram showing the 0-D model relationships between the various elements and devices of the network.

[0025] Figure 4 A diagram representing a seawater vaporizer, which is a core device requiring 3D CAE analysis.

[0026] Figure 5 is a diagram illustrating the correction method of the system ROM correction steps according to the present invention. Figure 5a This indicates the Gappy-POD calibration steps. Figure 5b This indicates the correction method based on the artificial intelligence correction steps.

[0027] Figure 6 is a diagram illustrating the optimal running example of the real-time algorithm using the real-time algorithm construction steps according to the present invention. Figure 6a This represents the optimal operating case for the operating variable x (the flow ratio to the seawater vaporizer). Figure 6bThis represents the optimal operating case for the performance variable E (energy cost required to produce one unit of NG).

[0028] *Explanation of key reference numerals in the figures*

[0029] S100: Network Definition Steps

[0030] S200: Steps for Establishing an Element Model

[0031] S300: System Model Establishment Steps

[0032] S400: System ROM Construction Steps

[0033] S500: System ROM Calibration Steps

[0034] S510: Gappy-POD Calibration Procedure

[0035] S520: Artificial Intelligence Correction Steps

[0036] S600: Steps for building a real-time algorithm. Detailed Implementation

[0037] The digital twin construction method of the present invention will now be described in detail with reference to the accompanying drawings. However, detailed descriptions of well-known functions and structures that may unnecessarily obscure the essence of the present invention will be omitted.

[0038] Figure 1 This is a flowchart illustrating the execution steps of the digital twin construction method of the present invention.

[0039] Reference Figure 1 The digital twin construction method of the present invention includes a network definition step S100, an element model establishment step S200, a system model establishment step S300, a system ROM construction step S400, a system ROM correction step S500, and a real-time algorithm construction step S600.

[0040] Each step of the present invention can be implemented using the program of the present invention on a computer, and the measurement data can be obtained using various IoT sensors set up on site.

[0041] The terms used in this invention are defined as follows.

[0042] - Operating variables: Boundary conditions and operable variables that determine the overall state of an element, device, or system; all other variables that affect the current state.

[0043] - Variables of interest: All variables representing the state of an element, equipment, or system that are of interest to the equipment operators (typically, measurement and performance variables are included in the variables of interest).

[0044] - Variable measurement: Utilizing various sensors to measure variables of elements, equipment, or systems online.

[0045] - Performance variables: Variables that represent the main performance characteristics of an element, equipment, or system (e.g., efficiency, lifespan, energy consumption, safety, amount of harmful emissions, etc.).

[0046] The network definition step S100, which is the first step in the digital twin construction method of the present invention, is to define the system as a network composed of a combination of element devices and connection media, etc., in order to construct a digital twin.

[0047] Figure 2 This serves as an example of a network defined for a city gas production process system. Figure 2 The illustrated city gas production process network refers to a system that supplies liquefied natural gas (NG) as gaseous NG through seawater vaporizers utilizing heat transferred from seawater and combustion vaporizers utilizing the combustion energy of fuel gas. In this example, the heat transferred to the seawater is provided by... This indicates that the energy supplied to the burner is from This indicates the flow rate, temperature, pressure, and enthalpy at each location. T, P, h represent...

[0048] In this example, for simplicity, the inbound traffic is... The flow ratio (x) to the seawater vaporizer, and the seawater flow rate. and temperature (T) 10 ) are defined as operating variables. All variables accompanying the factor model and performance variables such as the energy cost (E) required to produce one unit of NG can be defined as variables of interest. Table 1 shows the operating variables, variables of interest, measurement variables, and performance variables arbitrarily determined for the purpose of illustrating this invention.

[0049] Table 1

[0050]

[0051] Then, the element model establishment step S200 is the step of establishing the relationship between the main variables of individual element devices.

[0052] The relational expression represents a 0-D model that determines the state of a device based on operating variables. When it is difficult to construct a model based on fundamental physical laws, machine learning models such as data regression and neural network circuits can be used, as well as hybrid models that combine physical models and data-based models.

[0053] Then, the system model establishment step S300 is a step that reflects the addition of relations from 3-D computer-aided engineering (CAE) ROM or data of core element equipment to form a closure of all relations for the system.

[0054] Figure 3 Indicates the composition Figure 2 The 0-D model relationship of the various elements and devices of the network consists of operating variables and variables of interest. Figure 3 In this context, only the heat transferred from the seawater heater to the seawater is provided. Only with the addition of information can a closure be formed for all relations in the system.

[0055] When execution Figure 4 During CAE analysis of the seawater vaporizer shown, under the sampled operating conditions, the data was analyzed using a multidimensional distribution or... Such a method focuses on the property orthogonal decomposition (POD) of the matrix formed by the variables, which yields the result of the transformation of the matrix by the property orthogonal decomposition of the variables. Figure 5a The principal component vectors and coefficients shown form a 3-D ROM.

[0056] The new 0-D relational or arbitrary operating conditions required for the closure can be easily derived from the data analysis results obtained through this 3-D ROM or machine learning. Focus on variable values, such as those mentioned above.

[0057] According to one example, through... Figure 2 The city gas production process network shown is similarly connected to the BOG compressor, reliquefaction unit, pump, combustion vaporizer, seawater heater, and seawater vaporizer, which are equipment elements constructed from 0-D models and connected to 3-D CAE analysis, thus constructing a system model for the entire process.

[0058] On the other hand, directly deriving the solution to the relational formulas for all element devices from the system model usually requires excessive computation time, making it difficult to handle in real time. To address this issue, the present invention also includes a step S400 of constructing the system ROM from the calculation results of conditions sampled from the operating variable parameter space.

[0059] Therefore, in this invention, the solution for the sampled operating conditions in the system operating variable space can be obtained in advance, forming a matrix composed of all variables of interest. The system ROM of the system model established in the system model establishment step S300 can be derived through methods such as POD analysis, and a fast prediction result that can be responded to in real time can be obtained.

[0060] Then, the system ROM correction step S500 is a step to minimize the error between the model prediction value and the measurement data for the element equipment and system, including the Gappy-POD correction step S510 and the artificial intelligence correction step S520.

[0061] Gappy-POD correction step S510 is a step to derive the ROM from the matrix formed by all variables of interest obtained under the conditions of sampling in the operating variable parameter space, applying the Gappy-POD method to adjust the principal component coefficients of the ROM in order to minimize the sum of squares of the errors between the predicted and measured values ​​(see [reference]). Figure 5a The Gappy-POD method can also be applied to the correlation Gappy-POD method, which adjusts the principal component coefficients of ROM by fusing heterogeneous measurements (see "Patent Document 1").

[0062] Artificial intelligence correction step S520 is a step that corrects errors based on accumulated data and using artificial intelligence techniques such as machine learning through neural network circuits when the causal or functional relationship between the error between the predicted value and the data and the operating variables and the predicted physical quantities is unclear (see reference). Figure 5b ).

[0063] In the system ROM calibration step S500, the Gappy-POD calibration step S510 and the artificial intelligence calibration step S520 can be executed independently or simultaneously in sequence.

[0064] In this invention, during the process of minimizing the error of the predicted value, the system ROM calibration step S500 can assign appropriate weighting values ​​to each error based on the uncertainty of the measured value or the importance to the main performance indicators. Furthermore, when periodically or when significant changes occur in equipment operation, automatic correction can be performed to minimize the error between the online measured data and the model predicted value, thereby maintaining the accuracy of the digital twin.

[0065] Finally, the real-time algorithm construction step S600 is a step of constructing an algorithm based on real-time monitoring results to understand the expected system state or optimal operating conditions under virtual operating conditions. It can find the conditions for maximizing or minimizing the predefined performance variables or cost functions in the operating variable parameter space based on the system ROM constructed in the previous steps and propose them to the operators in real time (see Figure 6).

[0066] The embodiments disclosed in this specification and accompanying drawings are used only for the purpose of readily illustrating the technical concept of the invention and are not intended to limit the scope of the invention as described in the claims of the patent. Therefore, those skilled in the art will understand that various modifications and equivalent embodiments can be implemented therein.

Claims

1. A method for constructing a digital twin, characterized in that, include: The network definition step defines a multi-physical device system as a network composed of combinations of element devices; The element model establishment step involves establishing a 0-D model based on the relationships between operating variables and variables of interest for the individual element devices constituting the network. The system model establishment step involves constructing a 0-D model associated with 3-DCAE analysis for core element devices that contain additional information needed to realize the closure of the 0-D model established in the above element model establishment step, thereby forming a closure for all relations of the system. The system ROM construction step involves constructing a matrix of all variables of interest from the calculation results of the conditions sampled in the runtime variable parameter space, thereby constructing the system ROM for the system model established in the above system model establishment step. The system ROM calibration step minimizes the error between the model predictions and measured data for the elements, equipment, and systems. as well as The real-time algorithm construction steps involve building an algorithm based on real-time monitoring results to understand the expected system state or optimal operating conditions under virtual operating conditions.

2. The digital twin construction method according to claim 1, characterized in that, The above system ROM calibration steps include one or more of the following steps: The Gappy-POD correction step derives the ROM from a matrix formed by all variables of interest obtained under conditions sampled in the running variable parameter space, and applies the Gappy-POD method to adjust the principal component coefficients of the ROM in order to minimize the sum of squares of the errors between the predicted and measured values. as well as The artificial intelligence correction step corrects errors based on accumulated data when the causal or functional relationship between the error between the predicted value and the data and the operating variables and the predicted physical quantities is unclear. This correction utilizes artificial intelligence techniques such as machine learning through neural network circuits.

3. The digital twin construction method according to claim 2, characterized in that, The Gappy-POD method described above can be applied to the correlation Gappy-POD method for adjusting the principal component coefficients of ROM using measurements from fused heterogeneous species.

4. The digital twin construction method according to claim 2, characterized in that, The Gappy-POD correction steps and the artificial intelligence correction steps described above are executed independently or sequentially and simultaneously.

5. The digital twin construction method according to claim 1, characterized in that, The above system ROM correction steps assign appropriate weights to each error based on the uncertainty of the measured value or the importance of the main performance indicators in the process of minimizing the error of the predicted value.

6. The digital twin construction method according to claim 1, characterized in that, In the aforementioned system ROM calibration steps, when periodically or when significant changes occur in equipment operation, automatic calibration is performed to minimize the error between online measurement data and model predictions in order to maintain the accuracy of the digital twin.

7. The digital twin construction method according to claim 1, characterized in that, In the above real-time algorithm construction steps, conditions for maximizing or minimizing predefined performance variables or cost functions in the runtime variable parameter space are proposed to the operators in real time.

Images