A method for constructing a digital twin for a multi-physical facility system by combining reduced-order models, measurement data, and machine learning techniques

Abstract

The object of the present invention is to provide a method for constructing a digital twin that is capable of real-time monitoring, operation improvement, and response to accidents in industrial sites by combining a reduced-order model of a multi-physical system, on-site data, and artificial intelligence techniques. The digital twin construction method of the present invention includes a network definition step of defining a multi-physical equipment system as a network composed of a combination of element equipment; an element model establishment step of establishing a 0-D model based on relational equations for individual element equipment; a system model establishment step of closing all relational equations for the system by reflecting additional relational equations by machine learning from 3-D CAE ROM or data for core element equipment; a system ROM construction step of constructing a system ROM for the system model established in the system model establishment step from calculation results for conditions sampled in the operating variable parameter space; a system ROM correction step of minimizing errors between model prediction values ​​and measurement data for element equipment and the system; and a real-time algorithm construction step of constructing an algorithm for grasping the system state predicted under virtual operating conditions and optimal operating conditions based on the real-time monitoring results.

Description

[Technical field]

[0001] The present invention relates to a method for constructing a digital twin, and more particularly to a method for constructing a digital twin by combining a 0-dimensional (hereinafter referred to as "0-D") or multidimensional (1-D, 2-D or 3-D) reduced order model (hereinafter referred to as "ROM") for a multi-physical equipment system with on-site data based on the Internet of Things (hereinafter referred to as "IoT") and artificial intelligence techniques such as machine learning. [Background technology]

[0002] A digital twin is a twin model created in a virtual space for a system that consists of multiple physical facilities or multiple component facilities that exist in the real world. If all the conditions that determine the operating state are given, the digital twin should be able to determine the operating state and variables of interest just like in the real world.

[0003] Therefore, the digital twin must be able to accurately determine all variables of interest for current operating conditions (diagnosis mode) and predict all variables of interest for hypothetical operating conditions (prediction mode). Furthermore, the time required for such diagnosis and prediction must be short enough to provide information to the operator on-site in real time.

[0004] The following "Patent Document 1" presents a method for constructing a reduced-order model by combining on-site measurement data and CAE analysis through correlation-POD. However, this method is limited to only combining measurement data and 3D distribution given as a CAE analysis result, and has limitations in that it is difficult to extend the method to a multi-physics system configured as a network of multiple component equipment. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Korean Patent No. 10-2048243 Summary of the Invention [Problem to be solved by the invention]

[0006] The present invention has been devised to solve the above problems, and an object of the present invention is to provide a method for constructing a digital twin that can enable real-time monitoring, operational improvement, and response to accidents in industrial sites by combining ROM based on simulation of a multi-physical equipment system, field data, and machine learning techniques. [Means for solving the problem]

[0007] The digital twin construction method according to the present invention for solving the above problems includes a network definition step of defining a multiple physical equipment system as a network composed of a combination of element equipment; an element model establishment step of establishing a 0-D model based on relational equations for individual element equipment; a system model establishment step of closing all relational equations for the system by reflecting additional relational equations by machine learning from 3-D CAE ROM or data for core element equipment; a system ROM construction step of constructing a system ROM for the system model established in the system model establishment step from calculation results for conditions sampled in the operating variable parameter space; a system ROM correction step of minimizing errors between model prediction values ​​and measurement data for element equipment and the system; and a real-time algorithm construction step of constructing an algorithm for grasping the system state predicted under virtual operating conditions and optimal operating conditions based on the real-time monitoring results.

[0008] The system ROM correction step may include one or more steps of: a Gappy-POD correction step, which derives a ROM from a matrix consisting of all variables of interest obtained under sampled conditions in the operating variable parameter space, and applies the Gappy-POD method to adjust the principal component coefficients of the ROM so as to minimize the sum of squares of the errors between the predicted values ​​and the measured values; and an artificial intelligence correction step, which corrects the errors by utilizing artificial intelligence techniques such as machine learning using a neural network circuit based on accumulated data when the causal relationship or functional relationship between the errors between the predicted values ​​and the data and the operating variables and the predicted physical quantities is unclear.

[0009] The Gappy-POD method may be applied to an associative Gappy-POD method that combines different measurements to adjust the principal component coefficients of the ROM.

[0010] The Gappy-POD correction step and the AI ​​correction step may be performed independently of each other or may be performed sequentially and simultaneously.

[0011] In the system ROM correction step, in the process of minimizing the error in the predicted value, an appropriate weight may be given to each error depending on the uncertainty of the measurement value and its importance to the key performance index.

[0012] The system ROM correction step can perform automatic correction to minimize errors between online measurement data and model predictions periodically or when significant changes occur in equipment operation to maintain the accuracy of the digital twin.

[0013] The real-time algorithm construction step can find conditions that maximize or minimize a predefined performance variable or cost function in the driving variable parameter space and present the conditions to the driver in real time. Effect of the Invention

[0014] The digital twin construction method according to the present invention has the following excellent effects.

[0015] i) It is possible to build a system ROM capable of real-time operation with fast response characteristics based on the results of 0-D simulation for each component facility, 3-D CAE simulation for core facilities, and machine learning from data, and a digital twin based on it.

[0016] ii) By reflecting various online and offline measurement data through the GAPDY-POD or linked GAPDY-POD method, a calibrated 3-D ROM for core equipment or a calibrated ROM for the entire system can be constructed, and the reliability and accuracy of the digital twin based on it can be maintained.

[0017] iii) During the data fusion process, uncertainty can be minimized by applying different weights to the corresponding variables depending on the uncertainty of the measurement data and its importance to the final performance variable.

[0018] iv) Digital twins can be used in two modes: diagnosis and prediction, and can provide optimal operating conditions in real time within the allowable operating range, which can be reflected in driving automation.

[0019]

number

[0020]

number

[0021] [Figure 1] 1 is a flowchart sequentially illustrating steps of a digital twin construction method according to the present invention. [Diagram 2] FIG. 1 is a diagram showing a network defined for a city gas production process system as an example of the present invention. [Diagram 3] FIG. 3 is a diagram showing a 0-D model relational expression for each element facility constituting the network of FIG. 2. [Figure 4] FIG. 4 shows a seawater vaporizer as a core equipment that requires 3-D CAE analysis. [Diagram 5] 1A and 1B are diagrams showing a correction method relating to the system ROM correction stage of the present invention, where (a) shows a correction method relating to the Gappy-POD correction stage and (b) shows a correction method relating to the artificial intelligence correction stage. [Figure 6] FIG. 1 shows optimal operating examples using a real-time algorithm in the real-time algorithm construction stage of the present invention, where (a) shows an optimal operating example for an operating variable (formula 1 above, flow rate ratio to the seawater vaporizer), and (b) shows an optimal operating example for a performance variable (formula 2 above, energy cost per unit of NG production). DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0022] Hereinafter, a digital twin construction method according to the present invention will be described in detail with reference to the accompanying drawings, in which detailed description of known functions and configurations that may unnecessarily obscure the gist of the present invention will be omitted. FIG. 1 is a flowchart sequentially illustrating the steps of the digital twin construction method according to the present invention.

[0023] Referring to FIG. 1, the digital twin construction method according to 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).

[0024] Each step of the present invention can be performed by a program in which the present invention is embodied on a computer, and measurement data can be obtained by various IoT sensors installed in the field.

[0025] Some terms used in the present invention are defined as follows: Operating Variables: The boundary conditions and operational variables that determine the overall state of a component equipment or system, as well as all variables that affect its current state. Variables of interest: All variables that indicate the state of an element of equipment or a system and that are of interest to the equipment operator (measurement variables and performance variables are generally included in variables of interest). Measurement variable: A variable that is measured online by various sensors for a component equipment or system. Performance variable: A variable that indicates the main performance of a component facility or system (e.g., efficiency, lifespan, energy consumption, safety, amount of harmful exhaust gas, etc.).

[0026] The first step of the digital twin construction method according to the present invention, the network definition step (S100), is a step of defining a system as a network composed of a combination of elemental equipment, connecting media, etc. in order to construct a digital twin.

[0027] FIG. 2 shows a network defined for a city gas production process system as an example. The city gas production process network shown in FIG. 2 shows a system that supplies gaseous natural gas (NG) from a seawater vaporizer that utilizes heat transfer from seawater and a combustion vaporizer that utilizes the combustion energy of fuel gas. In this example, the amount of heat transferred to seawater is JPEG0007611617000003.jpg1316,The energy supplied from the burner is, JPEG0007611617000004.jpg1316, Flow rate, temperature, pressure, and enthalpy at each point are They are represented as JPEG0007611617000005.jpg88JPEG0007611617000006.jpg87JPEG0007611617000007.jpg87JPEG0007611617000008.jpg87 respectively.

[0028] In this example, for simplicity, we use the inlet flow rate ( JPEG0007611617000009.jpg119, JPEG0007611617000010.jpg1110), flow rate ratio to seawater vaporizer ( JPEG0007611617000011.jpg2618), seawater flow rate ( JPEG0007611617000012.jpg1010) and temperature ( JPEG0007611617000013.jpg810) were defined as operation variables. All variables associated with the element model and the energy cost per unit of NG production ( A performance variable such as (JPEG0007611617000014.jpg2724) can be defined as a variable of interest. Table 1 shows the operating variables, variables of interest, measurement variables, and performance variables arbitrarily determined for purposes of illustrating the present invention.

[0029] [Table 1]

[0030] Next, the element model establishment step (S200) is a step of establishing relationships between main variables for individual element equipment.

[0031] Here, the relational equation represents a 0-D model that can determine the state of the equipment according to the operating variables. When it is difficult to build a model based on fundamental physical laws, models based on machine learning such as data regression and neural network circuits can be used, and hybrid models that combine physical models and data-based models are also possible.

[0032] Next, the system model establishment step (S300) is a step of closing all the relationships for the system by reflecting additional relationships by machine learning from 3-D CAE (Computer Aided Engineering) ROM or data for the core component equipment.

[0033] Figure 3 shows the 0-D model relational equations for each elemental equipment that constitutes the network in Figure 2, and is composed of operational variables and variables of interest. In Figure 3, the amount of heat transferred to the seawater from the seawater heater ( Only when additional information about the sigma-based sigma filter (JPEG0007611617000016.jpg1316) is given can all the relationships for the system be closed.

[0034] When the CAE analysis of the seawater vaporizer shown in Figure 4 is performed, multi-dimensional distribution and A 3-D ROM consisting of the principal component vectors and coefficients shown in FIG. 5(a) is obtained by proper orthogonal decomposition (hereinafter referred to as "POD") of a matrix consisting of variables of interest such as JPEG0007611617000017.jpg1316.

[0035] From the data analysis results using such 3-D ROM and machine learning, new 0-D relations required for closure and One can easily determine the variable value of interest, such as JPEG0007611617000018.jpg1316.

[0036] As an example, a system model for the entire process can be constructed by connecting the equipment elements constructed using a 0-D model, such as a BOG compressor, a reliquefaction machine, a pump, a combustion-type vaporizer, and a seawater heater, and a seawater vaporizer constructed using a 0-D model linked with 3-D CAE analysis, to the city gas production process network shown in Figure 2.

[0037] On the other hand, the process of directly solving the equations for all element equipment from the system model generally requires excessive calculation time, and is difficult to respond to in real time. To solve this problem, the present invention further includes a step (S400) of constructing a system ROM from the calculation results for the conditions sampled in the parameter space of the operating variables.

[0038] As a result, the present invention can obtain fast prediction results that can respond in real time by finding solutions for the operating conditions sampled in the system operating variable space in advance, constructing a matrix composed of all variables of interest, and deriving a system ROM for the system model established in the system model establishment step (S300) through a method such as POD analysis.

[0039] Next, the system ROM correction step (S500) is a step of minimizing the error between the model prediction value and the measurement data for the component equipment and the system, and includes a Gappy-POD correction step (S510) and an artificial intelligence correction step (S520).

[0040] The Gappy-POD correction step (S510) is a step of applying the Gappy-POD method, which derives a ROM from a matrix consisting of all variables of interest obtained under sampled conditions in the operating variable parameter space, and adjusts the principal component coefficients of the ROM so as to minimize the sum of squares of errors between predicted values ​​and measured values ​​(see FIG. 5(a)). The Gappy-POD method can also be applied to the associative Gappy-POD method, which adjusts the principal component coefficients of the ROM by fusing different measured values ​​(see Patent Document 1).

[0041] The artificial intelligence correction step (S520) is a step in which, when there is no clear causal or functional relationship between the error between the predicted value and data on the one hand and the operating variables and predicted physical quantities on the other, the error is corrected based on accumulated data using artificial intelligence techniques such as machine learning using neural network circuits (see Figure 5(b)).

[0042] In the system ROM correction step (S500), the Gappy-POD correction step (S510) and the artificial intelligence correction step (S520) may be performed independently or sequentially at the same time.

[0043] In the present invention, during the system ROM correction step (S500), in the process of minimizing the error of the predicted value, an appropriate weighting can be given to each error according to the uncertainty of the measured value and the importance of the error to the key performance indicator. Also, in order to maintain the accuracy of the digital twin, automatic correction can be performed to minimize the error between the online measured data and the model predicted value periodically or when a significant change occurs in the equipment operation.

[0044] Finally, the real-time algorithm construction step (S600) is a step of constructing an algorithm for grasping the predicted system state and optimal driving conditions under virtual driving conditions based on the real-time monitoring results. Based on the system ROM constructed in the previous step, conditions that maximize or minimize predefined performance variables or cost functions in the driving variable parameter space can be found and presented to the driver in real time (see Figures 6(a) and 6(b)).

[0045] The embodiments disclosed in this specification and the accompanying drawings are merely used to easily explain the technical idea of ​​the present invention, and are not used to limit the scope of the present invention described in the claims. Therefore, a person having ordinary knowledge in the art will understand that various modifications and equivalent other embodiments are possible. [Explanation of symbols]

[0046] S100 Network definition stage S200 Element model establishment stage S300 System model establishment stage S400 System ROM Construction Stage S500 System ROM correction stage S510 Gappy-POD correction stage S520 AI correction stage S600 real-time algorithm construction stage

Claims

1. A network definition step in which a computer executes a program to define a multiple physical equipment system as a network composed of a combination of element equipment; An element model establishment stage in which a computer executes a program to establish a 0-D model based on a relational equation for each element equipment; A system model establishment step in which the computer executes a program to reflect additional relationships based on machine learning from 3-D CAE ROM or data for core element equipment in the 0-D model established in the element model establishment step, thereby closing all relationships for the system; a system ROM construction step in which the computer executes the program to construct a system ROM for the system model established in the system model establishment step from calculation results for conditions sampled in the operating variable parameter space; A system ROM correction step in which the computer executes a program to minimize errors between model predictions and measured data for component equipment and systems; and A real-time algorithm construction stage in which the computer executes a program to construct an algorithm for grasping the predicted system state and optimal operating conditions under virtual operating conditions based on the real-time monitoring results; How to build a digital twin, including:

2. In claim 1, The system ROM correction step includes: a Gappy-POD correction step in which a ROM is derived from a matrix of all variables of interest determined under conditions sampled in the operating variable parameter space, and the Gappy-POD method is applied to adjust the principal component coefficients of the ROM so as to minimize the sum of squares of the errors between the predicted values ​​and the measured values; and an artificial intelligence correction step, in which errors are corrected by utilizing artificial intelligence techniques such as machine learning using neural network circuits based on accumulated data when there is no clear causal or functional relationship between the error between the predicted value and the data on the one hand and the operating variables and the predicted physical quantities on the other hand.

3. In claim 2, The Gappy-POD method is characterized in that the associated Gappy-POD method, which combines different measured values ​​to adjust the principal component coefficients of the ROM, can be applied to the Gappy-POD method.

4. In claim 2, The digital twin construction method, wherein the Gappy-POD correction step and the artificial intelligence correction step are performed independently of each other or sequentially and simultaneously.

5. In claim 1, The system ROM correction step is a digital twin construction method characterized in that an appropriate weighting is given to each error depending on the uncertainty of the measurement value and its importance on the key performance indicator in the process of minimizing the error in the predicted value.

6. In claim 1, The system ROM correction step is a digital twin construction method characterized in that automatic correction is performed to minimize errors between online measurement data and model predicted values ​​when significant changes occur periodically or in equipment operation in order to maintain the accuracy of the digital twin.

7. In claim 1, The real-time algorithm construction step is characterized in that it finds conditions that maximize or minimize predefined performance variables or cost functions in the driving variable parameter space and presents them to the driver in real time.

Images