Parameter adjustment method for simulation calculation and information processing device capable of executing parameter adjustment

Abstract

To provide a parameter adjustment method for simulation calculation regarding heat or fluid in a given environment.SOLUTION: The parameter adjustment method includes the following steps: acquiring observation data obtained by observing the internal state of an environment; and acquiring model information regarding a simulation model. The model information includes equations of state that include parameters that define the behavior of heat or fluid moving through the environment, and multiple grid-point data describing the environment. The multiple grid points include: observation grid points showing observation locations, multiple obstacle grid points indicating obstacles in the environment, and multiple environmental grid points excluding observation grid points and multiple obstacle grid points. The method also includes the following steps: calculating the first distance, which is the distance of the path connecting the observation grid points and the environment grid points for each of the multiple environment grid points; when the path intersects an edge connecting two obstacle grid points, determining a detour path that does not cross any edge; calculating the second distance, which is the distance of the detour path; and calculating the adjusted parameters to output the same.SELECTED DRAWING: Figure 7

Description

[Technical Field]

[0001] The present disclosure relates to a simulation adjustment method and a simulation adjustment device that use an ensemble Kalman filter, which is a sequential data assimilation method. [Background technology]

[0002] Numerical simulation is a powerful tool for predicting phenomena that are difficult to grasp through direct measurement. However, to obtain accurate prediction results, it is necessary to properly adjust the simulation. Simulation adjustments are usually performed manually by simulation engineers by comparing the results with observed data. This method takes a lot of time and relies on the intuition and skills of the simulation engineers. Data assimilation is a simulation adjustment method that solves the above problems. Data assimilation is a technology that can improve prediction accuracy by statistically combining observation data and simulations to modify the parameters of the simulation model and the simulation estimation results. Data assimilation makes it possible to adjust simulations without relying on the intuition or skills of engineers. Patent Document 1 discloses technology that applies a data assimilation method to a system that performs weather forecasts for a specified area based on a meteorological model. According to Patent Document 1, by dividing the specified area into multiple regions and using data assimilation in each region, it is possible to improve forecast accuracy without relying on intuition or tips, and to perform forecasts efficiently. Patent Document 1 mentions the ensemble Kalman filter as one of the data assimilation techniques. The ensemble Kalman filter performs a process called "localization processing." Localization processing is a method for estimating the state of a system using observation data and a numerical model. By focusing on the area where observation data is available, localization processing can improve computational efficiency and reduce errors in the numerical model. In other words, it can improve estimation accuracy. In the ensemble Kalman filter, it is important to perform localization processing appropriately. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2018-194497 Summary of the Invention [Problem to be solved by the invention]

[0004] The ensemble Kalman filter is easy to implement and can be applied to a variety of simulations. However, if the technology disclosed in Patent Document 1 is used more generally in fields other than meteorology, there is a problem that estimation accuracy may decrease. For example, in the localization process of a conventional ensemble Kalman filter, estimation accuracy may decrease if an obstacle is present in the simulation model. This is because the positional relationship between the obstacle and the observation data is not taken into consideration, and the localization process is not performed appropriately.

[0005] One object of the present disclosure is to provide a technology that can suppress a decrease in estimation accuracy when applying an ensemble Kalman filter to a simulation model that includes an obstacle. [Means for solving the problem]

[0006] A simulation adjustment method according to one aspect of the present disclosure is a method for adjusting parameters for a simulation calculation related to heat or fluid in a predetermined environment, executed by an arithmetic circuit of an information processing device. The arithmetic circuit acquires observation data that observes an internal state of the environment and acquires model information related to a simulation model. The model information includes data on a state equation including parameters that define the behavior of heat or fluid moving through the environment, and data on a plurality of grid points that describe the environment. The plurality of grid points include observation grid points that indicate positions in the environment where the observation data was observed, a plurality of obstacle grid points that describe obstacles present in the environment, and a plurality of environment grid points other than the observation grid points and the plurality of obstacle grid points. The arithmetic circuit includes: for each of the plurality of environment lattice points, calculating a first distance which is the distance of a path connecting the observation lattice point and the environment lattice point; when the path intersects with an edge connecting two obstacle lattice points selected from the plurality of obstacle lattice points, determining a detour path which does not intersect with any edge; calculating a second distance which is the distance of the detour path; calculating adjusted parameters by data assimilating the parameters with the observation data; and outputting the adjusted parameters.

[0007] An information processing device according to one aspect of the present disclosure adjusts parameters for a simulation calculation related to heat or fluid in a predetermined environment. The information processing device includes an input device for acquiring data, an arithmetic circuit, and an output device for outputting the adjusted parameters. The input device acquires observation data that observes the internal state of the environment and model information related to a simulation model. The model information includes data on a state equation including parameters that define the behavior of heat or fluid moving through the environment, and data on a plurality of grid points that describe the environment. The plurality of grid points include observation grid points that indicate positions within the environment where the observation data was observed, a plurality of obstacle grid points that describe obstacles present in the environment, and a plurality of environment grid points other than the observation grid points and the plurality of obstacle grid points. The arithmetic circuit executes the following operations: for each of the plurality of environment lattice points, calculating a first distance which is the distance of a path connecting the observation lattice point and the environment lattice point; when the path intersects with an edge connecting two obstacle lattice points selected from the plurality of obstacle lattice points, determining a detour path which does not intersect with any edge; calculating a second distance which is the distance of the detour path; and calculating adjusted parameters by data assimilating the parameters with the observation data. [Effects of the Invention]

[0008] According to the parameter adjustment method for simulation calculations and the information processing device that performs parameter adjustment according to the present disclosure, it is possible to perform simulation model adjustment through data assimilation for a simulation model that includes obstacles while suppressing a decrease in prediction accuracy. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a schematic diagram of a drying oven 10, which is one example of an obstacle included in a simulation model. [Figure 2] 1 is a conceptual diagram of data assimilation. [Figure 3]This is a conceptual diagram of an ensemble Kalman filter for sequential data assimilation. [Figure 4] FIG. 10 is a conceptual diagram of a localization process. [Figure 5A] 10A and 10B are diagrams for explaining problems with localization processing depending on the presence or absence of an obstacle inside a simulation model. [Figure 5B] 10A and 10B are diagrams for explaining problems with localization processing depending on the presence or absence of an obstacle inside a simulation model. [Figure 6] 1 is a block diagram showing an example configuration of a simulation adjustment device 100 according to an exemplary embodiment. [Figure 7] 1 is a flowchart illustrating a procedure of a simulation adjustment method in an exemplary embodiment. [Figure 8A] FIG. 10 is a diagram illustrating a process for calculating a first distance. [Figure 8B] FIG. 10 is a diagram showing an example of a matrix in which the first distance and the coordinates and lattice numbers of the environment lattice points when the first distance is calculated are arranged in correspondence with each other. [Figure 9A] FIG. 1 is a diagram illustrating an example of a procedure for determining matrix elements that require correction. [Figure 9B] A diagram showing an example of a matrix in which elements that require correction are assigned a "1" and elements that do not require correction are assigned a "0" [Figure 10] 10 is a flowchart illustrating the second distance calculation process (step S7) in detail. [Figure 11A] 8B is a diagram showing an observation grid point Po, an obstacle 11s, and an environment grid point PL3 ​​on the model shown in FIG. 8A. [Figure 11B] FIG. 10 is a diagram illustrating an example of a first distance correction process. [Figure 12] 1 is a schematic diagram of an information processing system 200 including a server device 210 and a client device 220. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, embodiments of the present disclosure will be described in detail with appropriate reference to the drawings. However, more detailed description than necessary may be omitted. For example, detailed description of well-known matters or redundant description of substantially identical configurations may be omitted. This is to avoid unnecessary redundancy in the following description and to facilitate understanding by those skilled in the art. Note that the inventors provide the accompanying drawings and the following description to enable those skilled in the art to fully understand the present disclosure, and do not intend for them to limit the subject matter described in the claims.

[0011] An exemplary embodiment of the present invention will now be described with reference to the accompanying drawings. This embodiment is directed to a thermal fluid simulation for analyzing phenomena related to the interaction between heat and fluid, and can be used to deepen knowledge of heat transfer and fluid behavior in various engineering fields. The simulation model used numerically represents the flow velocity, pressure, and temperature of a fluid by discretely dividing the analysis domain, more specifically, by dividing the space into a finite number of grids. While various grid shapes are possible, any shape can be adopted in this embodiment.

[0012] However, the processing according to the present embodiment does not necessarily have to target both heat and fluid, and may also be applied to, for example, only fluid. The present disclosure relates to an adjustment method for simulation calculations related to heat or fluid in a predetermined environment, which is executed by an arithmetic circuit of an information processing device.

[0013] In the examples described below, for simplicity, a two-dimensional square or rectangular lattice model is used. However, the lattice shape may be a rhombic or triangular shape. Similarly, when applying to a three-dimensional model to more closely resemble reality, any spatial lattice model composed of a tetrahedron, pentahedron, hexahedron, or seven or more polyhedrons can be used.

[0014] (Embodiment) 1. Overview of the environment and simulation of the drying furnace 10 assumed in this embodiment FIG. 1 is a schematic diagram of a drying furnace 10, which is a specific example of a simulation model that includes an obstacle. In a drying furnace such as the drying furnace 10, a workpiece placement shelf 11 for placing workpieces is installed inside the furnace to be predicted. The workpiece placement shelf 11 partially divides the space inside the drying furnace 10 into two. The two divided spaces are connected to each other via an opening 12. Heat and fluids such as air can move from one space to the other through the opening 12. In this embodiment, when creating a model of such a drying furnace 10, the workpiece placement shelf 11 is modeled as an obstacle.

[0015] As used herein, the term "obstacle" refers to an object or structure that impedes or modifies the flow of fluid or the transfer of heat. For example, any physical object that has a shape and is in the path of a fluid (liquid or gas) or thermal energy can be considered an obstacle. Such an "object" may block, deflect, or cause turbulence in the flow of fluid, thereby changing the velocity, pressure, or flow pattern of the fluid. Alternatively, such an "object" may affect the distribution or rate of heat transfer by altering the pathways of heat conduction, convection, or radiation.

[0016] Figure 2 is a conceptual diagram of data assimilation. Data assimilation is a method for predicting phenomena using observed values ​​obtained from sensors and predicted values ​​obtained from numerical simulations. Data assimilation assumes that the observed values ​​and predicted values ​​each contain errors, and predicts phenomena statistically. In this embodiment, the simulation model is adjusted using an ensemble Kalman filter, which is a sequential data assimilation method.

[0017] Figure 3 is a conceptual diagram of the ensemble Kalman filter for sequential data assimilation. In the ensemble Kalman filter, ensembles with multiple different initial conditions are generated and each is allowed to evolve over time. At the time when measurements are obtained, data assimilation is performed using each ensemble and the observation data. The average value of the ensemble corrected by data assimilation is treated as the predicted value. The ensemble Kalman filter is a method that approximates the error covariance matrix with a finite number of ensemble members, allowing the Kalman filter to be applied to nonlinear problems. It can also be applied to problems where the dimensions of the model parameters are extremely large, making it applicable to a variety of problems.

[0018] In this embodiment, ensemble members are constructed by generating multiple state vectors consisting of state variables (e.g., temperature, flow velocity, pressure, etc.). Temperature can be observed, for example, using a thermocouple placed in the drying oven 10 (FIG. 1). Flow velocity can be measured using a flow meter, and pressure can be measured using a pressure gauge. Note that the ensemble Kalman filter can also estimate flow velocity from temperature, so it is sufficient if temperature is primarily obtained as an observed value.

[0019] The mean and variance of the ensemble members are calculated, and the probability density function of the state vector before data assimilation is determined based on the calculated values. The probability density function of the state vector before data assimilation uses the probability density function of the observed values ​​as the likelihood, and the probability density function of the state vector after data assimilation is calculated based on Bayes' theorem. Each state variable evolves over time according to a state equation or mathematical model until the time when an observed value is obtained. The state equation expresses how the state variable changes over time, and is the "simulation model" in this embodiment.

[0020] When applying an ensemble Kalman filter to a thermal fluid simulation, as in this embodiment, the equations of state or mathematical models are fluid dynamics equations such as the Navier-Stokes equations and the energy equation. The Navier-Stokes equations describe the temporal changes in the velocity and pressure fields of a fluid, and the energy equation describes the temporal changes in the temperature field. Using these equations, it is possible to numerically analyze the behavior of the fluid being analyzed. In the ensemble Kalman filter, these equations of state are used to statistically combine observed values, making it possible to improve the accuracy of the simulation.

[0021] In the simulation model example above, variables called simulation model parameters are set to determine the model's behavior and output. In a fluid dynamics simulation model that includes the Navier-Stokes equations and the energy equation, the following parameters are considered:

[0022] A. Physical properties: Density (ρ): The mass density of the fluid. Viscosity coefficient (μ): This represents the viscosity of a fluid and is an index of how viscous the fluid is. Thermal conductivity (k): A parameter that indicates how well a material conducts thermal energy. Specific heat (cp): The heat capacity of a material, which indicates how much heat energy it can store per unit mass.

[0023] B. Initial and boundary conditions: Initial conditions: The state of the fluid, such as velocity and temperature, at the start of the simulation. Boundary conditions: Conditions that constrain the fluid behavior at the boundaries of the simulation domain, such as velocity at the wall (no-slip or slip condition), temperature (constant temperature or adiabatic condition), etc.

[0024] C. Source term or external force term: External forces acting on a fluid, such as gravity and electromagnetic force. Influences from within the system, such as heat sources and mass sources.

[0025] Using an ensemble Kalman filter, these parameters can be iteratively updated in a way that minimizes the discrepancy between the observed data and the simulation model output, thereby improving the model's predictive accuracy.

[0026] Figure 4 shows the concept of localization processing. Localization processing is a method for reducing the influence of sampling errors in the ensemble Kalman filter as much as possible.

[0027] 4, for convenience of description, the symbol with a bar above E and the symbol with a hat above V are expressed as "E" and "V", respectively. Both E and V represent matrices.

[0028] In Figure 4, the concepts of the two equations are shown in the upper and lower sections.

[0029] In the upper equation, matrix E't|t-1 is obtained by multiplying matrix Et|t-1 by a localization function. "Matrix Et|t-1" represents the error variance between the predicted state at time t-1 and the true state. In other words, Et|t-1 represents the prediction error covariance matrix of the state at time t-1. "Localization function" is a function used to combine observed values ​​with simulation model information in the region where observations are available. This function typically has the function of modifying or weighting the elements of the covariance matrix based on the distance from the observed value. Examples of localization functions are described below. Matrix E't|t-1 obtained by the upper equation is a localized prediction error covariance matrix. In other words, the upper equation localizes the elements of the prediction error covariance matrix at time t-1 by applying the localization function to the prediction error covariance matrix Et|t-1, thereby emphasizing information in the region where observation data is available.

[0030] The lower equation is an equation for calculating the matrix Vt|t-1, which is the product of the matrix E't|t-1 and its transpose. The obtained matrix Vt|t-1 is the prediction error covariance matrix for the state at the next time, time t. In other words, the lower equation is significant for calculating the prediction error covariance matrix for the state at time t.

[0031] The ensemble Kalman filter is a method for approximating the error covariance matrix using a finite number of ensemble members, but the number of ensemble members is extremely small compared to the dimensionality of the state vector. This results in large sampling errors being introduced into the error covariance matrix. On the other hand, correlations become smaller at points farther away from a given point in the simulation model. For example, in meteorological atmospheric predictions on Earth, it is conceivable that there will be almost no correlation between the atmosphere on the other side of the Earth and a given point in the near future. Based on this idea, there is a method for attenuating the error covariance and reducing the impact of sampling errors as the physical distance from the observation point increases; this type of processing is called localization processing.

[0032] In the ensemble Kalman filter, it is possible to reduce sampling error by increasing the number of ensemble members significantly, but due to computational resource limitations, it is almost always performed with a number of ensemble members that increases the impact of sampling error. Therefore, how to appropriately perform localization processing is important from the perspective of improving estimation accuracy and reducing computational load.

[0033] 5A and 5B are diagrams illustrating the problems with localization processing depending on the presence or absence of obstacles within a simulation model. The example in FIG. 5A shows a case where there are no obstacles in the space. Since the observation point and point A are close to each other, it is conceivable that they are correlated. On the other hand, in the example in FIG. 5B, the observation point and point A are located in the same positions as in the example on the left, but an obstacle 11 is sandwiched between them. If the localization processing of a conventional ensemble Kalman filter were applied to such a case, localization processing would be performed simply based on the linear distance to each grid point. In other words, the conventional method performs localization processing without considering the influence of the obstacle 11 at all. However, the presence of the obstacle 11 may reduce the correlation between the observation point located above the obstacle 11 and point A located below the obstacle 11. In other words, it is conceivable that the mutual correlation is reduced due to obstruction by the obstacle 11. When localization processing is performed based on the conventional method, it is determined that there is a large correlation between the observation point and point A, resulting in processing that deviates from the actual correlation. This is because the distance between the observation point and point A is calculated as a straight-line distance, resulting in an underestimation.

[0034] In this embodiment, a graph is created taking obstacles into account, and the total cost is calculated using a shortest path search algorithm. This total cost is then used as a correction value. This corrects distances that would have been underestimated by conventional methods due to the influence of obstacles, enabling localization processing that suppresses a decrease in estimation accuracy. In short, the distance with the smallest total value of the edges that make up the detour route can be used as the correction value.

[0035] 2. Configuration 6 is a block diagram showing an example of the configuration of the simulation adjustment device 100 according to this embodiment. The simulation adjustment device 100 according to this embodiment takes into account the presence of obstacles in the drying oven 10 (FIG. 1) when performing a thermal fluid simulation on the transfer of heat in the drying oven 10 (FIG. 1) by applying an ensemble Kalman filter.

[0036] The simulation adjustment device 100 is an information processing device, and more specifically, can be realized using a commonly available PC. The simulation adjustment device 100 includes a storage device 110, an arithmetic circuit 120, and an input / output device 130.

[0037] The storage device 110 stores a program 111, observation data 112, and simulation model data 113. The storage device 110 may be realized by, for example, a semiconductor storage device such as a flash memory or a solid state drive (SSD), a magnetic storage device such as a hard disk drive (HDD), or other recording media, either alone or in combination. The storage device 110 may include a volatile memory such as an SRAM or a DRAM.

[0038] The arithmetic circuit 120 functions as a simulation model generation unit 121, an observation data acquisition unit 122, a data processing unit 123, a simulation unit 124, and a data assimilation unit 125. For example, the arithmetic circuit 120 includes a CPU, performs information processing, and realizes each function of the simulation adjustment device 100 described below. Such information processing is realized, for example, by the CPU operating in accordance with instructions of a program 111 stored in the storage device 110. The CPU is an example of the arithmetic circuit 120 of the present disclosure, but is not limited to a CPU. For example, the arithmetic circuit 120 may be configured with a circuit such as an MPU or FPGA.

[0039] The input / output device 130 is a concept that encompasses input devices and output devices. The input device may include, for example, a keyboard, a mouse, a touch panel, and also a communication interface device for receiving data from other devices. The communication interface device may include a wired or wireless communication circuit, an Ethernet (registered trademark) terminal, a USB (registered trademark) terminal, and so on. The output device may include, in addition to a display device, a communication interface device for outputting data to other devices. Examples of communication interface devices are the same as those of input devices.

[0040] 3.Operation 7 is a flowchart illustrating the procedure of the simulation adjustment method according to this embodiment. Steps S1 to S9 represent processing by the arithmetic circuit 120 itself in the simulation adjustment device 100, or processing performed in response to an instruction from the arithmetic circuit 120. Hereinafter, the processing by the arithmetic circuit 120 will be described as processing based on the functional blocks shown in FIG.

[0041] First, the simulation model generation unit 121 acquires a simulation model via the input / output device 130 (step S1). The simulation model acquired in step S1 is realized as fluid dynamics equations such as the Navier-Stokes equations and the energy equation. In this embodiment, the simulation model is a mathematical model used in thermal fluid simulations based on the Navier-Stokes equations, and is constructed by dividing space into finite grids and taking into account fluid properties, boundary conditions, and the like. In this embodiment, this simulation model is used to predict the behavior of a fluid on a computer.

[0042] Following step S1, the observation data acquisition unit 122 acquires observation data observed at an observation point within the prediction target (step S2). There may be multiple observation points within the prediction target. Sensors such as thermometers and anemometers may be installed at the observation points. The observation data acquired via the sensors is data representing the state of a fixed point within the prediction target, and is time-series data at that point. The observation data acquisition unit stores the acquired observation data as "observation data 112" in the storage device 110.

[0043] Following step S2, the data processing unit 123 acquires simulation model information (hereinafter abbreviated as "model information") via the input / output device 130 (step S3). More specifically, the data processing unit 123 acquires the model information by reading out the model information that has been prepared in advance and stored in the storage device 110. The model information is specifically the coordinates and grid point numbers of the grid points that constitute the simulation model acquired in step S1. Information about obstacles that constitute the simulation model is also acquired as model information. Specifically, the acquired obstacle information is the grid points on the surface or inside of the area or space where the obstacle exists, i.e., the coordinates and grid point numbers of the grid points that constitute the obstacles. When the drying oven in FIG. 1 is represented by a simulation model, the obstacle information is information about the grid point coordinates and grid point numbers that constitute the shelves in the drying oven on the simulation model.

[0044] In this specification, the internal environment of the drying oven 10 is described using multiple grid points. Of the multiple grid points, grid points that indicate positions in the environment where observation data was observed are sometimes referred to as "observation grid points," and one or more grid points that describe obstacles present in the environment are sometimes referred to as "obstacle grid points." Additionally, one or more grid points other than the observation grid points and obstacle grid points are sometimes referred to as "environment grid points."

[0045] Following step S3, data processing unit 123 acquires the position of the observation data (step S4). Specifically, step S4 is a process for determining an observation grid point, and data processing unit 123 acquires which grid point in the simulation model acquired in step S1 corresponds to the position of the observation data acquired in step S2. At this time, there may be cases in which the simulation model does not have a grid point on coordinates that exactly match the coordinates of the observation data, and in such cases, for convenience, the grid point that is closest to the coordinates of the observation data is treated as the coordinates of that observation data.

[0046] Following step S4, the data processing unit 123 calculates a first distance (step S5). In this embodiment, the first distance is a straight-line distance connecting two points from a specific observation lattice point to each of the environment lattice points, and is calculated as many times as there are environment lattice points. The data processing unit 123 generates a matrix including the obtained first distances. In the matrix, each first distance is associated with the coordinates of the environment lattice point at which the first distance was calculated, and the lattice number. Note that if there are multiple observation points, a first distance is calculated for each observation lattice point as many times as there are environment lattice points, and as a result, multiple matrices may be calculated.

[0047] 8A is a diagram for explaining the procedure of step S5 for calculating the first distance. In the example of FIG. 8A, the straight-line distance to each environment lattice point is calculated with the observation lattice point Po as the reference. In FIG. 8A, as an example, three environment lattice points P L1 ~P L3 is shown.

[0048] In order to calculate the first distance, the data processing unit 123 acquires the observation data position (i.e., the position of the observation grid point) from the coordinate information on the simulation model of the observation point acquired in step S4, and acquires the coordinates of each environment grid point from the model information acquired in step S3.

[0049] Furthermore, FIG. 8A shows obstacles 11s present on the simulation model. L1 and P L2 is located above the obstacle 11s, i.e., on the same side as the observation grid point Po. L3 is located below the obstacle 11s, and the path when the first distance is calculated crosses the obstacle 11s. However, in calculating the first distance, the data processing unit 123 calculates the distances between the observation grid points and the environment grid points by straight lines without taking into account the positional relationship with the obstacle 11s. The three environment grid points P shown in FIG. L1 ~P L3 Regarding this, the data processing unit 123 calculates the distance between the observation grid point Po and the environment grid point P L1The linear distance r1 between the observation grid point Po and the environment grid point P L2 The linear distance r2 between the observation grid point Po and the environment grid point P L3 Calculate the straight-line distance r3.

[0050] 8B shows an example of a matrix in which the first distance and the coordinates of the environment grid points and grid numbers when the first distance is calculated are arranged in correspondence with each other. For simplicity, FIGS. 8A and 8B show a two-dimensional model as an example, but the first distance can be calculated using a similar procedure when applied to a three-dimensional model, not just a two-dimensional model.

[0051] Referring again to FIG. In step S6, the data processing unit 123 determines elements that require correction in consideration of the positional relationship between the observation lattice point and the obstacle 11s. Elements that require correction refer to row elements of the matrix determined in step S5 that correspond to pairs of observation lattice points and environment lattice points where a straight line connecting the observation lattice point and the environment lattice point crosses the obstacle 11s. If an element requires correction, a value of "1" is assigned to the end (rightmost) of the element, and if correction is not required, a value of "0" is assigned to the end of the element.

[0052] 9A shows an example of the procedure of step S6 in which the data processing unit 123 determines the elements of the matrix that need to be corrected. As described with reference to FIG. 8A, the observation grid point Po and the environment grid point P L1 and the line connecting the observation grid point Po and the environment grid point P L2 The line connecting the observation grid point Po and the environment grid point P on the model does not cross the obstacle 11s. L3 The line connecting the observation grid point Po and the environment grid point P crosses the obstacle 11s. L3 The straight line path connecting the observation grid point Po and the environment grid point P intersects with any edge (described later) connecting the obstacle grid points that make up the obstacle 11s. L3 This means that an obstacle 11s is sandwiched between the object and the target.

[0053] The data processing unit 123 determines that the elements of the matrix related to the environment grid points corresponding to the former do not need to be corrected, and determines that the elements of the matrix related to the environment grid points corresponding to the latter need to be corrected. In the example of FIG. 9A, the elements that need to be corrected are the grid point group P LG is the row element of the matrix with respect to

[0054] 9B shows an example of a matrix in which elements that require correction are assigned a "1" and elements that do not require correction are assigned a "0" by the data processing unit 123. In this embodiment, a flag value of "1" or "0" is assigned to the rightmost side of each row element to determine whether correction is required.

[0055] Referring again to FIG. Following step S6, the data processing unit 123 recalculates the distance for the element requiring correction (step S7). The target of the distance recalculation process is the element requiring correction calculated in step S6, i.e., the first distance of the row element in the matrix shown in FIG. 9B to which a flag value "1" is assigned, which determines whether correction is required. The data processing unit 123 performs correction processing to calculate the second distance. While the first distance is the straight-line distance between the two points, the observation lattice point and the environment lattice point, the second distance is the distance of the route that bypasses obstacles between the two points, the observation lattice point and the environment lattice point. The processing of step S7 will be described in detail later with reference to FIGS. 10, 11A, and 11B.

[0056] Following step S7, the data processing unit 123 performs a data assimilation calculation to modify the parameters of the simulation model (step S8). The simulation model parameters modified by data assimilation include, but are not limited to, the temperature and flow velocity of the boundary in the simulation model.

[0057] In this embodiment, an ensemble Kalman filter is used as a data assimilation calculation method. The specific flow of modifying the simulation model parameters using the ensemble Kalman filter is as follows.

[0058] First, the simulation unit 124 generates multiple ensemble members with different initial values. Next, the simulation unit 124 advances each ensemble member until the time when observation data is obtained, and obtains an analytical value. The data assimilation unit 125 performs an assimilation process using the obtained analytical value and the observation data obtained in step S2. As a result, the data assimilation unit 125 updates the values ​​of each ensemble member and calculates the average of these ensemble members, thereby correcting and calculating the parameters of the simulation model. The ensemble Kalman filter is a method of performing corrections by sequentially repeating this process until an arbitrary time.

[0059] Specifically, localization processing in ensemble Kalman filter calculations is a process that reduces the sampling error of the error covariance matrix, and is performed during the ensemble Kalman filter calculation. Localization processing is a process that forcibly attenuates the error covariance at points far from the observation grid points in order to reduce the sampling error. The localization distance and localization function L(r) determine how much the error covariance is attenuated depending on the distance. Any function can be used for the localization function L(r), but a quintic function that approximates the Gaussian function expressed in the following equation (1) is often used.

[0060]

number

[0061] The r in the localization function L(r) is the normalized value obtained by dividing the distance from the observation grid point to each tourist grid point by the localization distance. The quintic function approximating the Gaussian distribution in equation (1) does not strictly become 0 even at infinity, but becomes completely 0 when a certain distance is reached, specifically when the localization distance r is greater than 2. This makes it possible to completely suppress sampling errors at positions a certain distance away from the observation grid point. In this way, localization processing is performed by multiplying the localization function L(r) by the error covariance matrix of the ensemble Kalman filter.

[0062] As mentioned above, the value of the localization function L(r) in Equation (1) is determined by the normalized distance r, which is the distance from the observation grid point to each environment grid point divided by the localization distance. When there are no obstacles in the simulation model to be predicted, there is no problem with calculating the distance r as the straight-line distance from the observation grid point to each environment grid point. However, when predicting an object such as a drying oven shown in Figure 1, where obstacles exist in the simulation model, the distance r may be underestimated, and the covariance that should be attenuated during localization may not be attenuated. Therefore, if the ensemble Kalman filter is applied using conventional localization processing as is, the estimation error will be large.

[0063] In this embodiment, in the localization process performed when performing data assimilation using the ensemble Kalman filter in step S8, the distance r is calculated using the second distance calculated in advance in step S7. This makes it possible to perform the ensemble Kalman filter while maintaining prediction accuracy even for a simulation model in which an obstacle is present.

[0064] Following step S8, the arithmetic circuit 120 instructs the input / output device 130 to output the modified, adjusted simulation model (step S9). With respect to the initial simulation model as a reference, the adjusted values ​​in the adjusted simulation model are, for example, the boundary conditions (temperature, flow rate) of the simulation model. In other words, it can be said that the boundary conditions with errors have been adjusted to boundary conditions with reduced errors. Taking the drying oven 10 of FIG. 1 as an example of the simulation analysis target, the settings of the boundary values ​​of two locations, the hot air outlet and inlet, in the simulation model are adjusted in the process of FIG. 7. When an obstacle exists inside the simulation model, as in the drying oven 10 of FIG. 1, by performing simulation adjustment using this method, it is possible to construct a simulation model with higher accuracy than when using conventional methods.

[0065] 10 is a flowchart illustrating the second distance calculation process (step S7) in detail. The data processing unit 123 acquires model information when correcting the first distance calculated in step S5 (step S71).

[0066] Following step S71, the data processing unit 123 acquires the first distance acquired in step S5 (step S72). Note that steps S71 and S72 are not essential because the model information and the first distance are acquired in steps S3 and S5 of Fig. 7. This information is required when viewed as a subroutine, so it is described here just to be safe.

[0067] Following step S72, the data processing unit 123 acquires the elements that require correction calculated in step S6 (step S73). Correction processing is performed in order on each of the acquired elements that require correction.

[0068] Following step S73, the data processing unit 123 creates an undirected graph in which each lattice point is a node and straight lines connecting each lattice point to an adjacent lattice point are edges (step S74). The edges constituting the undirected graph created at this time are made to not include any that cross obstacles present inside the simulation model. In other words, an undirected graph is created in which movement that crosses obstacles is impossible.

[0069] Following step S74, the data processing unit 123 executes a shortest path search in the undirected graph generated in step S74 and calculates the total cost at that time (step S75). The cost is determined by the number of edges included in the path. The data processing unit 123 executes the shortest path search with the observation lattice point as the starting node and the environment lattice point as the ending node. By solving this shortest path search problem, it is possible to determine the shortest path and its distance from the observation point to the environment lattice point on the other side of the obstacle while avoiding the obstacle. The distance of the shortest path in this case corresponds to the total cost.

[0070] Following step S75, the data processing unit 123 replaces the first distance with the total cost calculated in S75 (step S76). As a result, the first distance, which has been calculated as the distance of a straight route, is corrected to the distance of a route that bypasses the obstacle (second distance). Specifically, the first distance is corrected to a value indicating a longer distance.

[0071] Following step S76, the data processing unit 123 determines whether the number of executions is equal to the number of elements that require correction (step S77). If the result of the determination indicates that the number of executions is equal to the number of elements that require correction, the data processing unit 123 executes step S78, and if not, executes step S74 again to correct the next element that requires correction.

[0072] Following step S77, the data processing unit 123 outputs the second distance (step S78).

[0073] FIG. 11A shows the observation grid point Po, the obstacle 11s, and the environment grid point P on the model shown in FIG. 8A. L3 The observation grid point Po and the environment grid point P L3 The line segment connecting and crosses the obstacle 11s, and its distance (length) is r3.

[0074] FIG. 11B is a diagram illustrating an example of the first distance correction process. As described above, in this embodiment, each grid point except for the obstacle grid point is treated as a node (for example, node v p ,v q ), and the edges connecting the lattice points are defined as edges (for example, edge e pq ={v p ,v q}) in an undirected graph. As a result, the first distance is corrected to the second distance. In the example shown in FIG. 11B, the starting node is the observation lattice point Po, and the ending node is the environment lattice point P L3To simplify the calculation, movement between nodes is limited to two orthogonal axes, the x and y directions. From the observation grid point Po, move two edges in the -y direction until you hit the obstacle 11s, then move five edges in the -x direction, one edge in the -y direction, and four edges in the +x direction along the obstacle 11s. Then move one edge along the -y direction to reach the environment grid point P L3 As a result, the first distance r3, which was the square root of 17, or approximately 4, is corrected to the second distance r3' = 12. Note that a detour route is a route from the observation grid point to the target environment grid point, and does not intersect with any edge connecting any two obstacle grid points selected from multiple obstacle grid points. The shortest route among such routes is found by the above-mentioned process.

[0075] In the example of FIG. 11A, for simplicity, movement between nodes is limited to two axes, the x direction and the y direction, which are orthogonal to each other, but they do not have to be orthogonal.

[0076] Known optimal path search algorithms include the Bellman-Ford method, the Dijkstra method, and the Warshall-Floyd method. Any of these methods may be employed in this embodiment. When employing these methods, a path search may be performed taking into account a more realistic environment, or a method for reducing computational costs may be combined. Regarding the thermal fluid simulation of this embodiment, the former method may further consider the material and thermal conductivity of the obstacle, and the cost (weight) of the path along the obstacle surface may be considered to search for the shortest path. Regarding the latter method, if there are paths in any direction, the distance from the measurement grid point to the environment grid point for which the distance is to be determined may be calculated only for paths in directions away from the environment grid point, and paths in directions away from the environment grid point may be excluded from the distance calculation. In other words, a so-called divide-and-conquer method may be used, in which each path is classified as either a viable path or an inviable path, and the cost is calculated only for the viable path to search for the shortest path. While FIG. 7 illustrates an example for a two-dimensional simulation model, the same method may also be applied to a three-dimensional simulation model.

[0077] By executing the sequential data assimilation ensemble Kalman filter using the second distance calculated by this series of processes, it is possible to estimate simulation model parameters while suppressing a decrease in accuracy even in simulation models that include obstacles.

[0078] As described above, the simulation adjustment device 100 performs localization processing using an ensemble Kalman filter by taking into account the influence of obstacles on the first distance during localization processing and correcting the first distance that has been undercalculated. This makes it possible to adjust simulation parameters while suppressing a decrease in prediction accuracy, regardless of the presence or absence of obstacles.

[0079] Furthermore, the correction value for the first distance is the total cost calculated as a shortest path search problem for a graph in which the grid points of the simulation model are nodes and the sides between the grid points are edges. This makes it possible to perform localization processing that suppresses a decrease in prediction accuracy regardless of the shape and number of obstacles included in the simulation model.

[0080] FIG. 12 schematically illustrates the configuration of an information processing system 200 including a server device 210 and a client device 220. The server device 210 and the client device 220 can transmit and receive data to and from each other via a communication line 230 such as the Internet. The information processing system 200 has an aspect that allows for distributed execution of the processing that the simulation adjustment device 100 according to the present embodiment described above performed in a standalone manner. The server device 210 and the client device 220 each have hardware similar to that of the simulation adjustment device 100 shown in FIG. 6. That is, the server device 210 and the client device 220 each have a storage device, an arithmetic circuit, and an input / output device. Specific examples of the storage device, the arithmetic circuit, and the input / output device have been described with reference to FIG. 6, so repeated description will be omitted.

[0081] First, the client device 220 receives the observation data 112 and the simulation model data 113 from the user, and transmits the information to the server device 210 via the communication line 230 .

[0082] The server device 210 performs the processing described with reference to FIGS. 7 to 11B using the observation data 112 and simulation model data 113 received from the client device 220 to calculate adjusted parameters. The server device 210 then transmits the adjusted parameters to the client device 220 or other devices via the communication line 230. By performing a simulation calculation related to heat or fluid using the adjusted parameters, it becomes possible to perform a simulation with reduced deterioration in prediction accuracy. By performing such a simulation, it is possible to more accurately reproduce the thermal behavior inside the actual drying furnace 10.

[0083] (Example) The following describes exemplary aspects of the present disclosure.

[0084] <Aspect 1> A method for adjusting parameters for a simulation calculation relating to heat or fluid in a predetermined environment, the method being executed by an arithmetic circuit of an information processing device, the arithmetic circuit comprising: Obtaining observation data that observes an internal state of the environment; obtaining model information relating to a simulation model; the model information includes data of a parameterized equation of state that defines the behavior of heat or fluid moving through the environment, and data of a plurality of grid points that describe the environment; The plurality of lattice points are observation grid points indicating locations within the environment where the observation data was observed; a plurality of obstacle grid points describing obstacles present in the environment; and a plurality of environment grid points other than the observation grid points and the plurality of obstacle grid points; obtaining model information, including: Calculating a first distance, which is the distance of a path connecting the observation grid point and the environment grid point, for each of the plurality of environment grid points; determining a detour route that does not intersect with any edge when the route intersects with an edge connecting two obstacle grid points selected from the plurality of obstacle grid points; calculating a second distance that is the distance of the detour route; calculating adjusted parameters by data assimilation of the parameters with the observation data; Outputting the adjusted parameters. A method for adjusting parameters, including:

[0085] <Aspect 2> The adjustment method according to aspect 1, wherein calculating the first distance is calculating, for each of the plurality of environment lattice points from the observation lattice point, the distance of a straight line path connecting the observation lattice point and the environment lattice point as the first distance.

[0086] <Aspect 3> determining the detour path includes determining a path including one or more edges, the one or more edges are sides connecting two adjacent lattice points, the two adjacent grid points are selected from one or more obstacle grid points defining a surface of the obstacle among the plurality of obstacle grid points, and the plurality of environment grid points; The method for adjusting according to embodiment 1 or 2.

[0087] <Aspect 4> Calculating the second distance includes: Each lattice point in the simulation model is a node, A side connecting the two adjacent lattice points is defined as an edge, The observation grid point is set as a starting node, and calculating a total cost of the shortest path in an undirected graph when the path connecting the observation grid point and the environment grid point constituting the first path intersects with an edge connecting the two obstacle grid points, with the environment grid point being the end node. The method for adjusting according to embodiment 3.

[0088] <Aspect 5> 4. The adjustment method according to aspect 3, wherein the second distance is calculated as a distance with the smallest total value of the one or more edges that form the detour route.

[0089] <Aspect 6> The adjustment method of aspect 4 or 5, wherein calculating the adjusted parameters includes performing localization processing of a sequential data assimilation ensemble Kalman filter using the second distance.

[0090] <Aspect 7> 7. The adjustment method according to any one of aspects 1 to 6, wherein the acquiring of the observation data comprises acquiring the observation data via an input device of the information processing device.

[0091] <Aspect 8> The adjustment method according to aspect 7, wherein acquiring the observation data is performed via a communication interface device that is an input device of the information processing device.

[0092] <Aspect 9> 7. The adjustment method according to any one of aspects 1 to 6, wherein the acquiring of the model information comprises acquiring the observation data via an input device of the information processing device.

[0093] <Aspect 10> 8. The adjustment method according to claim 7, wherein acquiring the model information comprises acquiring the model information via a communication interface device that is an input device of the information processing device.

[0094] <Aspect 11> A computer program for causing the arithmetic circuit to execute the adjusting method according to any one of aspects 1 to 6.

[0095] <Aspect 12> An information processing device that adjusts parameters for simulation calculations related to heat or fluid in a predetermined environment, An input device for acquiring data, an arithmetic circuit, and an output device for outputting adjusted parameters. Equipped with the input device acquires observation data obtained by observing an internal state of the environment and model information related to a simulation model; the model information includes data of an equation of state including parameters that defines the behavior of heat or fluid moving through the environment, and data of a plurality of grid points that describe the environment; The plurality of lattice points are observation grid points indicating locations within the environment where the observation data was observed; a plurality of obstacle grid points describing obstacles present in the environment; and a plurality of environment grid points other than the observation grid points and the plurality of obstacle grid points; The arithmetic circuit comprises: Calculating a first distance, which is the distance of a path connecting the observation grid point and the environment grid point, for each of the plurality of environment grid points; determining a detour route that does not intersect with any edge when the route intersects with an edge connecting two obstacle grid points selected from the plurality of obstacle grid points; calculating a second distance that is the distance of the detour route; calculating adjusted parameters by data assimilation of the parameters with the observation data; and An information processing device that executes the above. [Industrial Applicability]

[0096] The present invention provides a simulation adjustment method using data assimilation that takes into account the influence of obstacles included in a simulation model, in relation to a simulation adjustment device. [Explanation of symbols]

[0097] 10 Drying oven 11 Workpiece placement shelf 100 Simulation adjustment device (information processing device) 110 Storage device 111 Program 112 Observation Data 113 Simulation model data 120 Arithmetic circuit 121 Simulation model generation unit 122 Observation data acquisition unit 123 Data Processing Unit 124 Simulation Department 125 Data Assimilation Department 130 Input / Output Devices

Claims

1. A method for adjusting parameters for a simulation calculation relating to heat or fluid in a predetermined environment, the method being executed by an arithmetic circuit of an information processing device, the arithmetic circuit comprising: Obtaining observation data that observes an internal state of the environment; obtaining model information relating to a simulation model; the model information includes data of an equation of state including parameters that defines the behavior of heat or fluid moving through the environment, and data of a plurality of grid points that describe the environment; The plurality of lattice points are observation grid points indicating locations within the environment where the observation data was observed; a plurality of obstacle grid points describing obstacles present in the environment; and a plurality of environment grid points other than the observation grid points and the plurality of obstacle grid points; obtaining model information, including: Calculating a first distance, which is the distance of a path connecting the observation grid point and the environment grid point, for each of the plurality of environment grid points; determining a detour route that does not intersect with any edge when the route intersects with an edge connecting two obstacle grid points selected from the plurality of obstacle grid points; calculating a second distance that is the distance of the detour route; calculating adjusted parameters by data assimilation of the parameters with the observation data; Outputting the adjusted parameters. A method for adjusting parameters, including:

2. 2. The adjustment method according to claim 1, wherein calculating the first distance comprises calculating, for each of the plurality of environment lattice points from the observation lattice point, a distance of a straight line path connecting the observation lattice point and the environment lattice point as the first distance.

3. Determining the detour path includes determining a path including one or more edges, the one or more edges are sides connecting two adjacent lattice points, the two adjacent grid points are selected from one or more obstacle grid points defining a surface of the obstacle among the plurality of obstacle grid points, and the plurality of environment grid points; The adjusting method according to claim 1 .

4. Calculating the second distance includes: Each lattice point in the simulation model is a node, A side connecting the two adjacent lattice points is defined as an edge, The observation grid point is set as a starting node, and calculating a total cost of the shortest path in an undirected graph when the path connecting the observation lattice point and the environment lattice point constituting the first path intersects with an edge connecting the two obstacle lattice points, with the environment lattice point being the end node. The adjusting method according to claim 3 .

5. The adjustment method according to claim 3 , wherein the second distance is calculated as the distance with the smallest total value of the one or more edges that make up the detour route.

6. 5. The adjustment method of claim 4, wherein calculating the adjusted parameters includes performing a localization process of the ensemble Kalman filter based on a sequential data assimilation ensemble Kalman filter using the second distance.

7. The adjustment method according to claim 1 , wherein the acquisition of the observation data comprises acquiring the observation data via an input device of the information processing device.

8. The adjustment method according to claim 7 , wherein the acquisition of the observation data comprises acquiring the observation data via a communication interface device that is an input device of the information processing device.

9. The adjustment method according to claim 1 , wherein the acquisition of the model information comprises acquiring the observation data via an input device of the information processing device.

10. The adjustment method according to claim 7 , wherein the acquiring of the model information comprises acquiring the model information via a communication interface device that is an input device of the information processing device.

11. A computer program for causing the arithmetic circuit to execute the adjustment method according to any one of claims 1 to 6.

12. An information processing device that adjusts parameters for simulation calculations related to heat or fluid in a predetermined environment, An input device for acquiring data, an arithmetic circuit, and an output device for outputting adjusted parameters. Equipped with the input device acquires observation data obtained by observing an internal state of the environment and model information related to a simulation model; the model information includes data of an equation of state including parameters that defines the behavior of heat or fluid moving through the environment, and data of a plurality of grid points that describe the environment; The plurality of lattice points are observation grid points indicating locations within the environment where the observation data was observed; a plurality of obstacle grid points describing obstacles present in the environment; and a plurality of environment grid points other than the observation grid points and the plurality of obstacle grid points; The arithmetic circuit comprises: Calculating a first distance, which is the distance of a path connecting the observation grid point and the environment grid point, for each of the plurality of environment grid points; determining a detour route that does not intersect with any edge when the route intersects with an edge connecting two obstacle grid points selected from the plurality of obstacle grid points; calculating a second distance that is the distance of the detour route; calculating adjusted parameters by data assimilation of the parameters with the observation data; and An information processing device that executes the above.

Images