Information processing device, glass production system, information processing method, program, and calibration parameter estimation method

Abstract

An information processing device according to an embodiment comprises a state estimation unit that estimates the state of an actual machine by using measurement data stored in a measurement data storage unit that stores measurement data of multiple sensors disposed inside or outside a glass melting furnace in order to measure values relating to molten glass, and by using a glass melting furnace simulation model that models the glass melting furnace.

Description

Information processing device, glass manufacturing system, information processing method, program, and calibration parameter estimation method

[0001] The present disclosure relates to an information processing device, a glass manufacturing system, an information processing method, a program, and a calibration parameter estimation method.

[0002] A glass melting furnace is used as a glass article manufacturing device. The glass melting furnace includes, for example, a melting tank having a bottom and a sidewall, and an upper structure covering the melting tank.

[0003] Convection currents occurring in molten glass obtained by melting glass raw materials affect the quality of the final glass article, specifically, bubble density, foreign matter density, homogeneity, etc. Therefore, the technology described in Patent Document 1 proposes that, of the upstream circulating flow formed upstream of the bubbler, a flow near the surface of the molten glass is defined as an upstream surface flow, and, of the downstream circulating flow formed downstream of the bubbler, a flow near the surface of the molten glass is defined as a downstream surface flow, and the average flow velocities of the upstream surface flow and the downstream surface flow are adjusted (see Patent Document 1).

[0004] International Publication No. 2015 / 033931

[0005] Here, the glass melting process is required to improve the quality of the produced glass, improve production efficiency, reduce costs, reduce environmental impact, etc. However, because the glass melting process is a high-temperature process, it is difficult to grasp the details of the inside of a glass melting furnace, and it has sometimes been difficult to stably achieve the required items.

[0006] Numerical simulations can be used to understand the internal state of a glass melting furnace or to study its operation, but conventional techniques have limited their use for the following reasons: the model adjustment work (workflow for executing a simulation) required to obtain simulation results with sufficient accuracy takes time, and the number of personnel who can execute the simulation (e.g., dedicated analysts) is limited because specialized knowledge and skills are required to execute the simulation. For these reasons, it is sometimes difficult to execute a simulation in a timely manner, and the demand for solving problems at manufacturing sites in a short period of time cannot be fully met.

[0007] The present disclosure has been made in consideration of the above circumstances, and aims to provide an information processing device, a glass manufacturing system, an information processing method, a program, and a calibration parameter estimation method that can realize automatic simulations in glass manufacturing, thereby making simulation-based support more efficient than conventional techniques.

[0008] One aspect of the present disclosure is an information processing device including: a measurement data storage unit that stores measurement data from a plurality of sensors installed inside or outside a glass melting furnace to measure values ​​related to molten glass; and a state estimation unit that estimates the state of an actual machine using the measurement data stored in the measurement data storage unit; and a glass melting furnace simulation model that models the glass melting furnace.

[0009] One aspect of the present disclosure is a glass manufacturing system having a glass manufacturing apparatus including a glass melting furnace and an information processing device, wherein the information processing device includes a measurement data storage unit inside or outside the glass melting furnace that stores measurement data from a plurality of sensors installed inside or outside the glass melting furnace to measure values ​​related to molten glass, and the information processing device includes a glass melting furnace simulation model that models the glass melting furnace, and a state estimation unit that estimates the state of an actual device using the measurement data stored in the measurement data storage unit.

[0010] One aspect of the present disclosure is an information processing method in which an information processing device estimates the state of an actual furnace using measurement data stored in a measurement data storage unit that stores measurement data from multiple sensors installed inside or outside a glass melting furnace to measure values ​​related to molten glass, and a glass melting furnace simulation model that models the glass melting furnace.

[0011] One aspect of the present disclosure is a program for causing a computer to realize a state estimation function for estimating the state of an actual machine using measurement data stored in a measurement data storage unit that stores measurement data from multiple sensors installed inside or outside a glass melting furnace to measure values ​​related to molten glass, and a glass melting furnace simulation model that models the glass melting furnace.

[0012] One aspect of the present disclosure is a calibration parameter estimation method, which uses operating conditions and calibration parameters of a glass melting furnace as set variables, performs a plurality of simulations on a glass melting furnace simulation model based on an experimental design method within a range that comprehensively includes conditions expected in an actual glass manufacturing process, performs machine learning using the simulation results from the simulations to generate a predictive model, and evaluates the sensitivity of the operating conditions of the glass melting furnace and the calibration parameters to changes in at least one of the simulation results of the molten glass temperature, the resistivity of the molten glass, the current value, voltage value, or resistance value between heating electrodes, the molten glass flow rate, the furnace material temperature, or the atmospheric temperatures inside and outside the furnace at a plurality of observation points in the glass melting furnace.

[0013] According to the information processing device, information processing device, glass manufacturing system, information processing method, program, and calibration parameter estimation method disclosed herein, automatic simulation can be realized for glass manufacturing, thereby making simulation-based support more efficient than conventional techniques.

[0014] 1 is a diagram showing a schematic configuration example of a glass manufacturing system according to an embodiment; FIG. 2 is a diagram showing an outline of an online simulation according to an embodiment; FIG. 3 is a diagram showing an outline of digital prototyping according to an embodiment; FIG. 4 is a diagram showing an example of a cross section of a glass melting furnace according to an embodiment; FIG. 5 is a diagram showing another example of a cross section of a glass melting furnace according to an embodiment; FIG. 6 is a diagram showing an example of a procedure of processing performed in an information processing device according to an embodiment; FIG. 7 is a diagram showing an example of a procedure of a calibration parameter estimation process performed in an information processing device according to an embodiment; FIG. 8 is a diagram showing another example of a procedure of a calibration parameter estimation process performed in an information processing device according to an embodiment; FIG. 9 is a diagram showing a table representing examples of calibration parameters according to an embodiment; FIG. 10 is a diagram showing a schematic first process in a specific example of a calibration parameter estimation process according to an embodiment; FIG. 11 is a diagram showing a schematic result of the first process in a specific example of a calibration parameter estimation process according to an embodiment; FIG. 12 is a diagram showing a schematic second process in a specific example of a calibration parameter estimation process according to an embodiment; FIG. 13 is a diagram showing a table representing examples of estimation results of calibration parameters according to an embodiment; FIG. 14 is a diagram showing a comparative example of characteristics of a simulation result using estimation results of calibration parameters according to an embodiment and characteristics of actual machine data; FIG. 15 is a diagram showing a comparative example of characteristics based on a simulation result in a case where calibration parameter estimation process is not performed and characteristics of actual machine data; FIG. 16 is a diagram showing a schematic example of aging change in furnace wall heat radiation of a glass melting furnace.

[0015] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0016] First Embodiment A first embodiment will be described.

[0017] 1 is a diagram showing a schematic configuration example of a glass manufacturing system 1 according to an embodiment. The glass manufacturing system 1 includes an information processing device 111, a database 112, and a glass manufacturing apparatus 131. In this embodiment, the information processing device 111 and the database 112 are communicatively connected, and the database 112 and the glass manufacturing apparatus 131 are communicatively connected.

[0018] The glass manufacturing apparatus 131 includes a glass melting furnace 311. In this embodiment, the glass manufacturing apparatus 131 also includes a sensor unit 331 installed inside the glass melting furnace 311 and a sensor unit 351 installed outside the glass melting furnace 311. Here, the sensor unit 331 includes one or more sensors. In addition, the sensor unit 351 also includes one or more sensors.

[0019] In this embodiment, the sensor unit 331 installed inside the glass melting furnace 311 and the sensor unit 351 installed outside the glass melting furnace 311 are shown, but both of them are not necessarily provided, and for example, only one of them may be provided. When only one of the sensor units (sensor unit 331 or sensor unit 351) is provided, the sensor unit includes multiple sensors.

[0020] Database 112 stores measurement data R1, which is data measured by sensor unit 331 and sensor unit 351. Here, in the present embodiment, a case is shown in which database 112 is provided as a separate entity from information processing device 111, but as another example, the function of database 112 may be integrated into information processing device 111. In this case, database 112 is not provided, and, for example, information processing device 111 and glass manufacturing apparatus 131 are connected so as to be able to communicate with each other.

[0021] <Information Processing Device> In this embodiment, the information processing device 111 is configured using a computer. The information processing device 111 has a processor such as a CPU (Central Processing Unit), and performs various processes and controls by executing predetermined programs using the processor.

[0022] The information processing device 111 includes an input unit 211, an output unit 212, a communication unit 213, a storage unit 214, and a control unit 215. The input unit 211 includes an operation unit 231. The output unit 212 includes a display unit 251. The control unit 215 includes a state estimation unit 271, a calibration parameter estimation unit 272, a model calculation unit 273, a design of experiments execution unit 274, a prediction model generation unit 275, a change sensitivity evaluation unit 276, and an observation point data acquisition unit 277. Note that, in this embodiment, the calibration parameter estimation unit 272 includes the design of experiments execution unit 274, the prediction model generation unit 275, the change sensitivity evaluation unit 276, and the observation point data acquisition unit 277, but is not necessarily limited to this configuration.

[0023] The input unit 211 inputs information from an external device or a user (e.g., a person). For example, the operation unit 231 accepts an operation performed by a user and inputs information corresponding to the operation. The output unit 212 outputs information to an external device or a user. For example, the display unit 251 has a screen and displays and outputs information to be displayed on the screen. The information may be in any format, and may be, for example, image information or numerical information.

[0024] The communication unit 213 communicates with an external device. The communication may be wired or wireless. In this embodiment, for the sake of convenience, the communication unit 213 is shown as a functional unit separate from the input unit 211 and the output unit 212. However, as another example, the function of the communication unit 213 may be considered to be included in the input unit 211 and the output unit 212.

[0025] The storage unit 214 stores various types of information. In the present embodiment, the storage unit 214 stores, for example, a glass melting furnace simulation model R11, a calibration parameter estimation model R12, a prediction model R13, and a simulation result R14. Note that, in the present embodiment, a method is shown in which the prediction model R13 is used when estimating calibration parameters using the calibration parameter estimation model R12. However, if another method is used, the prediction model R13 does not need to be used. In the present embodiment, for example, the prediction model R13 may be considered to be a common model with the calibration parameter estimation model R12.

[0026] Here, the information stored in each of the database 112 and the storage unit 214 may be arbitrary. The various pieces of information in this embodiment may be stored in either the database 112 or the storage unit 214, or may be stored in both. Furthermore, although this embodiment illustrates information stored in the database 112 and the storage unit 214, other arbitrary information may also be stored. Note that in this embodiment, the various pieces of data may be referred to as information.

[0027] The database 112 has a function of a measurement data storage unit, and stores measurement data of a plurality of sensors installed inside or outside the glass melting furnace 311 to measure values ​​related to the molten glass. In this embodiment, these plurality of sensors are collectively a plurality of sensors included in the sensor unit 331 and the sensor unit 351.

[0028] The state estimation unit 271 estimates the state of the actual machine using the measurement data R1 stored in the measurement data storage unit (in this embodiment, the database 112) and a glass melting furnace simulation model R11 that models the glass melting furnace 311. Here, in this embodiment, the actual machine is the glass melting furnace 311 in the actual glass manufacturing apparatus 131. Note that estimation may also be called, for example, prediction or inference.

[0029] The calibration parameter estimation unit 272 estimates the calibration parameters using a calibration parameter estimation model R12 that estimates the calibration parameters of the glass melting furnace simulation model R11 and the measurement data R1 stored in the measurement data storage unit. In this case, the state estimation unit 271 estimates the state of the actual machine using the calibration parameters estimated by the calibration parameter estimation unit 272. Note that in this embodiment, for convenience of explanation, the parameters for calibrating the glass melting furnace simulation model R11 will be referred to as calibration parameters, but they may be called by other names or simply referred to as parameters.

[0030] The model calculation unit 273 performs an optimization calculation or a sensitivity analysis calculation using the glass melting furnace simulation model R11 or a summary model based on the glass melting furnace simulation model R11, based on the proposed operation change guideline based on the result of state estimation by the state estimation unit 271. In this embodiment, the summary model represents a model that summarizes a phenomenon, and includes, for example, a contraction model, a machine learning model, or both. Here, the proposed operation change guideline may be specified, for example, by a user, or may be automatically generated and specified by a device that performs predetermined processing. Note that both the optimization calculation and the sensitivity analysis may be performed; for example, the optimization calculation may be performed using the calculation result of the sensitivity analysis.

[0031] The output unit 212 outputs information relating to at least one of the result of the state estimation by the state estimation unit 271 and the result of the calculation by the model calculation unit 273 .

[0032] Here, an example configuration for estimating the calibration parameters is shown. Note that any other method may be used as the method for estimating the calibration parameters. The design of experiments execution unit 274 uses the operating conditions and calibration parameters of the glass melting furnace 311 as setting variables and executes multiple simulations for the glass melting furnace simulation model R11 based on the design of experiments within a range that comprehensively includes conditions expected in an actual glass manufacturing process. The memory unit 214 has the function of a simulation result memory unit and stores simulation results R14 by the design of experiments execution unit 274. The prediction model generation unit 275 performs machine learning using the simulation results R14 stored in the simulation result memory unit (the memory unit 214 in this embodiment) to generate a prediction model R13. The change sensitivity evaluation unit 276 evaluates the change sensitivity of the operating conditions and calibration parameters of the glass melting furnace 311 to the simulation results of at least one of the molten glass temperature and the molten glass flow rate at multiple observation points in the glass melting furnace 311.

[0033] The observation point data acquisition unit 277 acquires operating conditions at a predetermined time point for a plurality of observation points in the glass melting furnace 311, and also acquires observation point data, which is data on at least one of the molten glass temperature and the molten glass flow rate at the predetermined time point. The calibration parameter estimation unit 272 estimates calibration parameters based on the evaluation results by the change sensitivity evaluation unit 276 so that the deviation between the observation point data acquired by the observation point data acquisition unit 277 for each observation point and the results obtained by the simulation is small.

[0034] The calibration parameter estimation unit 272 may restrict the range of the calibration parameter values ​​based on changes in the amount of heat radiation from the furnace materials due to aging of the glass melting furnace 311.

[0035] [Example of Digital Twin Technology in a Glass Melting Process] An example of digital twin technology in a glass melting process is shown with reference to Figures 2A and 2B. A digital twin refers to recreating a real-world environment in a virtual space based on information from the real space. The online simulator may have, as a function for generating a trigger for executing an operation, a function for automatically generating a trigger (automatic function) by the online simulator, a function for generating a trigger in response to user input (manual function), or both of these functions. When the trigger is generated automatically, the online simulator executes the operation, for example, at a predetermined timing (e.g., a fixed time each day). When the trigger is generated manually, the online simulator executes the operation in response to information input from the user (trigger information). In this case, the user can, for example, instruct the online simulator to execute the operation at any timing and specify any simulation period. In this embodiment, the online simulator is described as a system that automatically generates and updates a simulation model in a virtual space in real time. As another example, the online simulator may be a system that generates and updates a simulation model that reproduces the state at any point in time through user operation. The digital prototyping tool is a tool that executes simulations by changing conditions for sensitivity analysis and preview studies.

[0036] As an example of digital twin technology for the glass melting process, a system that integrates an online simulator and a digital prototyping tool has been developed. This system can be used to verify the operation of a float furnace, for example. This technology makes it possible to quickly and precisely grasp the state of the glass melting process, which was previously difficult, and to quickly consider production conditions in advance.

[0037] In summary, the system uses data from the actual manufacturing process to run a simulation based on a simulation model using an online simulator. Parameters may be estimated during the simulation. The system also achieves optimal actions, optimal designs, information, and added value through digital prototyping. The online simulator and digital prototyping tool may operate synchronously or independently.

[0038] Specific examples of operational flows in this system are (Process L1) to (Process L4) below. (Process L1) The system acquires operation data and actual equipment status data. Here, operation data includes, for example, data on heating distribution from burners or electric boosters. Furthermore, actual equipment status data includes, for example, data on temperature, flow rate, resistivity, or quality. (Process L2) The system automatically generates a simulation model of the current kiln using an online simulator. As another example, the system may generate a simulation model that reproduces the kiln status at any point in time using the online simulator and user input information (trigger information). Here, for example, if both automatic and manual functions are possible, (Process L2) may be interpreted as including the automatic function (Process L2a) and the manual function (Process L2b). (Process L3) The system uses the simulation model as a reference to perform a case study simulation in which the operating conditions are arbitrarily changed. (Process L4) In the system, the optimum operating conditions derived from the simulation are reflected in the actual machine, and the process returns to the above (Process L1).

[0039] The temperature inside a glass melting furnace or the convection of molten glass changes daily due to various factors, such as the condition of raw materials and refractory bricks, and affects the quality of the glass produced. Each time these factors change, it is necessary to re-derive the optimal operating conditions, but this adjustment process takes time, resulting in a decrease in production volume during that time. Because the temperature inside a glass melting furnace is as high as 1,600°C and it is difficult to grasp the details of the internal conditions, simulation technology for the glass melting process has traditionally been used, but collecting the necessary data is time-consuming and it is difficult to run simulations in a timely manner.

[0040] To solve these problems, we developed a simulation tool that can confirm changes in temperature distribution inside a glass melting furnace and changes in molten glass convection, etc., from a simulation model generated using operation data from the glass melting furnace. This allows process engineers to easily grasp the detailed state inside the glass melting process and consider production conditions in advance, tasks that previously required a dedicated simulator to carry out.

[0041] FIG. 2A is a diagram illustrating an overview of an online simulation according to an embodiment. The online simulation automatically (or manually) constructs a simulation model C1 that reproduces the state of an actual machine. In an actual machine manufacturing process A1, actual machine data B2 is obtained for the actual machine B1. Here, operation information includes production volume, gas flow rate, power, pull, batch ratio, etc. Furthermore, actual machine state data includes temperature, quality, composition, etc. An integrated data platform A3 stores (memorizes) the actual machine data B2. In a simulation process A2 of the online simulation, a process flow is automatically executed when a trigger occurs. Specifically, the simulation process A2 issues a data request (e.g., a simulation execution request) to the integrated data platform A3, obtains data from the integrated data platform A3, creates an input file, and executes a simulation using the simulation model C1. In this case, parameter estimation D1 may be performed based on data from the integrated data platform A3, and the estimated parameters may be used in the simulation.

[0042] In the simulation process A2, the timing of acquiring data by a data request may be arbitrary. For example, the data may be acquired continuously by issuing a data request at regular time intervals. In addition, in the simulation process A2, the user may issue a data request by specifying a date and time of their choice. In this case, the processing after the user issues a data request (e.g., trigger information) is automatically executed.

[0043] FIG. 2B is a diagram illustrating an overview of digital prototyping according to an embodiment. This digital prototyping enables easy execution of simulations for sensitivity analysis and reconnaissance studies. Here, a simulation model C1 obtained through online simulation is used. A user E1 operates a computer F1 to set simulation conditions. A server G1 executes a simulation based on the simulation conditions. In this case, for example, a regression model based on actual machine data may be generated and used. The server G1 may be, for example, a cloud server or an on-premise server.

[0044] User E2 checks the results of the simulation using computer F2. Note that user E2 and computer F2 may be the same as user E1 and computer F1, or they may be different. Here, server G1 may perform a simulation to estimate the current kiln state, a simulation to change conditions, or a prediction of changes using a summary model (e.g., a reduced model or machine learning model). Note that a reduced model may be generated based on the results of multiple simulations, for example, using machine learning.

[0045] Such online simulation and digital prototyping techniques can solve problems at glass manufacturing sites in a short period of time. Although the present embodiment illustrates a case where an online simulation is performed, as another example, a configuration in which the simulation is performed by an offline device (e.g., a computer) may also be used.

[0046] [Glass Melting Furnace] Referring to Figures 3A and 3B, a glass melting furnace 100 is shown as an example of a glass melting furnace. Furnaces with other configurations may also be used as the glass melting furnace. Although not shown in Figures 3A and 3B, multiple sensors (sensors included in one or both of sensor unit 331 and sensor unit 351 in the example of Figure 1) may be provided inside or outside the glass melting furnace 100 as needed.

[0047] Fig. 3A is a diagram showing an example of a cross section of the glass melting furnace 100 according to the embodiment. Fig. 3B is a diagram showing another example of a cross section of the glass melting furnace 100 according to the embodiment. For convenience of explanation, each of Fig. 3A and Fig. 3B shows an XYZ Cartesian coordinate system, which is a three-dimensional Cartesian coordinate system. Fig. 3A shows a cross section of the glass melting furnace 100 in the XZ plane, and Fig. 3B shows a cross section of the glass melting furnace 100 in the YZ plane.

[0048] 3A and 3B, the XY plane is a horizontal plane. As shown in Fig. 3A and 3B, the glass melting furnace 100 includes a melting tank 10 (glass melting furnace) into which a glass material 70 (for convenience of explanation, the glass material 70 represents glass raw materials before melting and represents molten glass formed by melting the glass raw materials after melting, and may include a mixture of these. The same applies below) is supplied, a plurality of current-carrying electrodes 50, a plurality of thermocouples 40, an upper structure 20 covering the upper part of the melting tank 10, and a plurality of burners 60.

[0049] The glass melting furnace 100 melts the glass material 70 by burner combustion of the burners 60 and application of voltage to the powered electrodes 50. The molten glass material 70 is supplied from the left end of FIG. 3B , flows in the positive direction of the Y-axis, and is removed from the right end of FIG. 3B . A plurality of burners 60 are arranged on both sides of the upper structure 20 in the Y-axis direction (the flow direction of the glass material 70). Each burner 60 heats the surface of the glass material 70 near its location through burner combustion. Therefore, the amount of gas supplied to each of the plurality of burners 60 determines the distribution of the amount of heat on the surface of the glass material 70 (distribution in the flow direction of the glass material 70 and the width direction of the melting tank 10).

[0050] The current-carrying electrodes 50 are arranged so that pairs of electrodes to which a voltage is applied protrude from the bottom of the melting tank 10, with multiple pairs in the Y-axis direction (the flow direction of the glass material 70) and one or more pairs in the X-axis direction (the width direction of the melting tank 10). Each pair of current-carrying electrodes 50 heats the glass material 70 between the pair of current-carrying electrodes 50 when a voltage is applied. Therefore, the voltage applied to each of the multiple pairs of current-carrying electrodes 50 determines the distribution of the amount of heat within the glass material 70 (the distribution in the flow direction of the glass material 70 and the width direction of the melting tank 10). Furthermore, the amount of gas supplied to the burner 60 and the voltage applied to the pair of current-carrying electrodes 50 determine the distribution of the amount of heat across the depth of the melting tank 10. For simplicity of illustration, only one pair of current-carrying electrodes 50 is shown in the example of FIG. 3A , but multiple pairs of current-carrying electrodes 50 may be arranged within the X-Z cross section.

[0051] Each of the thermocouples 40 arranged at the bottom of the melting tank 10 measures the temperature of the glass material 70 at the bottom of the melting tank 10. Each of the thermocouples 40 arranged at the upper structure 20 measures one or both of the ambient temperature at the top of the interior of the glass melting furnace 100 and the temperature of the upper structure 20 as a substitute for the surface temperature of the glass material 70. At the bottom of the melting tank 10, multiple thermocouples 40 are arranged in the Y-axis direction (the flow direction of the glass material 70) and multiple thermocouples 40 are arranged in the X-axis direction (the width direction of the melting tank 10). Furthermore, multiple thermocouples 40 are arranged at the upper structure 20 in the Y-axis direction. Therefore, the multiple thermocouples 40 arranged at the bottom of the melting tank 10 can measure the temperature distribution of the glass material 70 at the bottom of the melting tank 10 (the distribution in the flow direction of the glass material 70 and the width direction of the melting tank 10). Furthermore, the temperature distribution on the surface of the glass material 70 (distribution in the flow direction of the glass material 70) can be measured by a plurality of thermocouples 40 arranged on the upper structure 20. Here, the temperature distribution on the surface of the glass material 70 does not necessarily mean the temperature distribution on the surface of the molten glass material 70, but may also include, for example, the temperature distribution on the surface of the glass frit before melting that is supplied from the left end in the example of FIG. 3B. The temperature distribution on the surface of the glass frit before melting can be measured by the thermocouple 40 at the left end in FIG. 3B, which is arranged near the supply of the glass material 70.

[0052] Each pair of current-carrying electrodes 50 can measure the temperature of the glass material 70 between the pair by detecting the current (or resistance value) between the pair. Therefore, using multiple pairs of current-carrying electrodes 50, the temperature distribution of the glass material 70 in the center of the melting tank 10 (distribution in the flow direction of the glass material 70 and the width direction of the melting tank 10) can be measured. Furthermore, by combining two or more pairs of thermocouples 40 arranged at the bottom of the melting tank 10, thermocouples 40 arranged in the upper structure 20, and current-carrying electrodes 50, the temperature distribution in the depth direction of the melting tank 10 can be obtained. Furthermore, for example, by photographing the surface of the glass material 70 using a video camera (not shown) and analyzing the photographed image, the flow velocity distribution on the surface of the glass material 70 can be obtained. Here, such analysis is performed, for example, by detecting the movement speed of bubbles or the like inside the glass material 70.

[0053] 4 to 6, an example of the procedure of processing performed by the information processing device 111 will be shown. In this example, a case where time-series data is acquired as measurement data R1 of the actual machine is shown. In this example, it is assumed that the measurement data R1, a glass melting furnace simulation model R11, a calibration parameter estimation model R12, and a prediction model R13 have been acquired in advance.

[0054] 4 is a diagram showing an example of a processing procedure performed in the information processing device 111 according to the embodiment. (Step S1) In the information processing device 111, the control unit 215 acquires actual machine data. Then, the process proceeds to step S2. Here, in this example, the actual machine data is time-series measurement data R1.

[0055] (Step S2) In the information processing device 111, the calibration parameter estimation unit 272 estimates the calibration parameters. Then, the process proceeds to step S3. Here, in this example, the calibration parameter estimation unit 272 estimates the calibration parameters using the measurement data R1 and the calibration parameter estimation model R12.

[0056] (Step S3) In the information processing device 111, the state estimation unit 271 estimates the state of the actual machine. Then, the process proceeds to step S4. Here, in this example, the state estimation unit 271 estimates the state of the actual machine using the measurement data R1, the glass melting furnace simulation model R11, and the estimated calibration parameters.

[0057] For example, the state estimation is performed by estimating the state at a predetermined time (for example, a date of year / month / day or month / day, or a date and time, etc.). In this case, the calibration parameters are also estimated by estimating values ​​at that time (for example, a date of year / month / day or month / day, or a date and time, etc.). Note that the time may be, for example, the time since the actual device was put into use (for example, the amount of time elapsed). As another example, instead of directly specifying such a predetermined time, the time may be indirectly specified by other conditions. For example, the condition may be that the temperature of a predetermined part (for example, the temperature of molten glass) reaches a predetermined value.

[0058] (Step S4) The information processing device 111 acquires a proposed operation change guideline based on the result of the actual machine state estimation. Then, the process proceeds to step S5. In this example, the user considers a proposed operation change guideline based on the result of the actual machine state estimation and operates the operation unit 231 to input the determined proposed operation change guideline into the information processing device 111. As another example, a configuration may be used in which the control unit 215 or the like in the information processing device 111 automatically performs processing to consider a proposed operation change guideline based on the result of the actual machine state estimation.

[0059] Here, various proposed guidelines may be used as the proposed operation change guidelines, and may be determined taking into consideration one or more of, for example, increasing or decreasing the amount of gas used, increasing or decreasing the amount of electricity used, adjusting the convection conditions by changing the bubbler flow rate, improving the quality of the glass, and the like.

[0060] (Step S5) In the information processing device 111, the model calculation unit 273 performs an optimization calculation or a sensitivity analysis calculation based on the proposed operation change guideline using the glass melting furnace simulation model R11 or a summary model based on the glass melting furnace simulation model R11. Then, the process proceeds to step S6. Here, this calculation is performed in a manner that realizes the proposed operation change guideline. Furthermore, when a summary model is used, for example, the model calculation unit 273 generates the summary model based on the simulation results R14 of the glass melting furnace simulation model R11.

[0061] In the processing of steps S4 to S7 in the first round, for example, change prediction may be performed using a condition change simulation, or change prediction may be performed using a summary model (for example, a contraction model or a machine learning model).

[0062] (Step S6) In the information processing device 111, the output unit 212 outputs information related to the results of the optimization calculation or sensitivity analysis. Then, the process proceeds to step S7. Here, in this example, the information is visualized by being displayed on the screen of the display unit 251, for example. The user can then evaluate the information by looking at it. As another example, if the information processing device 111 automatically performs the evaluation, the information does not need to be output.

[0063] (Step S7) In the information processing device 111, if the control unit 215 determines based on the evaluation result that the goal has been achieved, the process proceeds to step S8, and if it determines that the goal has not been achieved, the process proceeds to step S4. Here, in this example, the user operates the operation unit 231 to input whether or not the goal has been achieved to the information processing device 111, and the control unit 215 determines whether or not the goal has been achieved. As another example, a configuration may be used in which the information processing device 111 automatically determines whether or not the goal has been achieved.

[0064] (Step S8) An operation determination is made, and the process of this flow ends. Note that in this example, the operation determination is made by the user, but as another example, a process similar to the operation determination may be performed by the information processing device 111.

[0065] Examples of calibration parameter estimation processes are shown with reference to Figures 5 and 6. The example in Figure 5 is an example using a function (not machine learning), while the example in Figure 6 is an example using machine learning.

[0066] 5 is a diagram showing an example of the procedure of the calibration parameter estimation process performed in the information processing device 111 according to the embodiment. In general, a calibration parameter estimation model is constructed in the processes of steps S31 and S32, and the calibration parameter estimation process is performed using the calibration parameter estimation model in the processes of steps S33 and S34.

[0067] (Step S31) In the information processing device 111, the control unit 215 acquires actual machine data. Then, the process proceeds to step S32. Here, in this example, time-series measurement data R1 is used as the actual machine data.

[0068] (Step S32) In the information processing device 111, the control unit 215 constructs a calibration parameter estimation model using a function. Here, in the calibration parameter estimation model using a function, for example, calibration parameters are obtained by the function in accordance with simulation conditions. In other words, the function uses the conditions as variables and the calibration parameters as a solution.

[0069] (Step S33) In the information processing device 111, the control unit 215 inputs information that is a condition for the simulation. Then, the process proceeds to step S34. Here, the information may be, for example, instructed (input) to the information processing device 111 by a user operating the operation unit 231, or may be instructed (input) to the information processing device 111 from another device. The condition may be, for example, a date and time (or a year, month, date, or month, date, etc.), a temperature, or the like.

[0070] (Step S34) In the information processing device 111, the calibration parameter estimation unit 272 performs calibration parameter estimation processing based on the calibration parameter estimation model using the function. As a result, the information processing device 111 obtains the estimation results of the calibration parameters.

[0071] 6 is a diagram showing another example of the procedure of the calibration parameter estimation process performed in the information processing device 111 according to the embodiment. In general, a calibration parameter estimation model is constructed in the processes of steps S71 to S73, and the calibration parameter estimation process is performed using the calibration parameter estimation model in the processes of steps S74 to S76.

[0072] (Step S71) In the information processing device 111, the control unit 215 (for example, the state estimation unit 271) executes a simulation for constructing a database for creating a calibration parameter estimation model. Then, the process proceeds to step S72.

[0073] (Step S72) In the information processing device 111, the control unit 215 acquires the simulation results, and then the process proceeds to step S73.

[0074] (Step S73) In the information processing device 111, the control unit 215 constructs a calibration parameter estimation model using machine learning. Here, in the calibration parameter estimation model using machine learning, for example, by inputting predetermined information into the machine learning model, the values ​​of the calibration parameters are determined based on output values ​​from the model. In this example, the model inputs simulation conditions and actual device data (in this example, time-series measurement data R1), and outputs the values ​​of the calibration parameters themselves or information for obtaining the values ​​of the calibration parameters.

[0075] (Step S74) In the information processing device 111, the control unit 215 inputs information that is a condition for the simulation. Then, the process proceeds to step S75. Here, the information may be, for example, instructed (input) to the information processing device 111 by a user operating the operation unit 231, or may be instructed (input) to the information processing device 111 from another device. The condition may be, for example, a date and time (or a year, month, date, or month, date, etc.), a temperature, or the like.

[0076] (Step S75) In the information processing device 111, the control unit 215 collects (acquires) actual machine data. Then, the process proceeds to step S76. Here, in this example, time-series measurement data R1 is used as the actual machine data.

[0077] (Step S76) In the information processing device 111, the calibration parameter estimation unit 272 performs a calibration parameter estimation process based on a calibration parameter estimation model using machine learning. As a result, the information processing device 111 obtains an estimated result of the calibration parameters.

[0078] <Specific Example of Calibration Parameter Estimation Processing> A specific example of the calibration parameter estimation processing will be shown with reference to FIGS. 7, 8, and 9A to 9C.

[0079] FIG. 7 is a diagram showing a table 3011 representing examples of calibration parameters according to an embodiment. While this example shows examples of calibration parameters in table format, the table format is not necessarily required. Table 3011 shows an outline of the calibration parameters. Table 3011 stores calibration parameter items, minimum values ​​(Min) related to constraints, maximum values ​​(Max) related to constraints, and units in association with each other. Note that the information stored in table 3011 is an example showing an outline and is not actual information. In this example, multiple calibration parameters are set. These multiple calibration parameters may include, for example, items that are difficult to measure (actually measure) in real time.

[0080] In the example of FIG. 7, for example, for the calibration parameter of the item "heat dissipation AAA", Min is "... ("..." represents the corresponding numerical value, the same applies below)", Max is "...", and the unit is [W / m ２ For example, for the calibration parameters of the item "heat transfer AA", Min is "...", Max is "...", and the unit is [W / m ２ For example, for the calibration parameter of the item "air volume aaa", Min is "...", Max is "...", and the unit is [Nm ３ / h]. Similarly, in the example of FIG. 7, information on the calibration parameters of other items is stored.

[0081] In this example, the calibration parameter estimation unit 272 estimates (determines) as the estimated value of each calibration parameter a value included in the range between the minimum value (Min) and maximum value (Max) related to each constraint. Furthermore, for example, only the minimum value (Min) or only the maximum value (Max) may be set as such a constraint. The value related to such a constraint may be set based on, for example, past performance. Note that such a constraint does not necessarily have to be set, and if such a constraint is not set, the calibration parameter estimation unit 272 may estimate (determine) any value as the estimated value of each calibration parameter.

[0082] Fig. 8 is a diagram schematically showing a first process in a specific example of the calibration parameter estimation process according to the embodiment. Fig. 9A is a diagram schematically showing a result of the first process in the specific example of the calibration parameter estimation process according to the embodiment. Fig. 9B is a diagram schematically showing a second process in the specific example of the calibration parameter estimation process according to the embodiment. Here, the names "first process" and "second process" are used for convenience of explanation, and the processes may be called by any other names.

[0083] (First Process) In the first process, the information processing device 111 executes a simulation under a comprehensive set of operating conditions and generates a machine learning model based on the results of the simulation. In this example, a machine learning model for predicting temperature is generated as the machine learning model. As another example, a machine learning model for predicting a physical quantity other than temperature may be generated.

[0084] In the first process, the design of experiments execution unit 274 executes a simulation of the glass melting furnace simulation model R11 under comprehensive conditions using the design of experiments. The conditions include, for example, the operating conditions of the glass melting furnace 311 and conditions for multiple calibration parameters (in this example, the values ​​of the multiple calibration parameters). The operating conditions of the glass melting furnace 311 are, for example, conditions related to heat input. Examples of the heat input conditions may include the gas volume, the output of the electric boosting, or the control mode of the bubbler. Furthermore, a mode for realizing comprehensive conditions may be, for example, a mode in which a combination of multiple parameters is mechanically determined using random numbers (random numbers) within a predetermined range (between an upper limit and a lower limit) to comprehensively (or nearly comprehensively) cover the range. Note that a mode in which other conditions are also used while using such random numbers may also be used.

[0085] In Figure 8, the universal set U1 of conditions is shown by a frame, and each of the multiple conditions (simulation conditions) included in this universal set U1 is shown by a dot. In the example of Figure 8, to simplify the illustration, a single condition V1 is assigned a symbol as a representative. Here, since the total number of conditions included in the universal set U1 can be infinite, for example, the design of experiments execution unit 274 performs simulations under each of a finite number of conditions. One simulation is performed for each condition. This generates a database containing multiple simulation results.

[0086] The prediction model generation unit 275 generates a prediction model R13 based on the results of simulations performed for each of a plurality of conditions. In this example, a machine learning model is generated as the prediction model R13. Figure 8 schematically shows how machine learning is performed on a set of simulation results (a simulation result group W1) to generate a machine learning model that predicts simulation results (pseudo simulation results) between the respective simulation results (between points in the example of Figure 8).

[0087] The generated prediction model R13 makes it possible to predict simulation results (pseudo-simulation results) corresponding to any conditions. Fig. 8 shows a characteristic 3111 representing the relationship between a value (predicted value) predicted by the prediction model R13 (in this example, a machine learning model for temperature prediction) and an actual value (measured value) for a predetermined temperature related to the glass melting furnace 311. Note that, although the example in Fig. 8 shows a single predetermined temperature, the prediction model R13 may predict multiple temperatures. In this case, a graph such as that shown in Fig. 8 is obtained for each of the multiple temperatures.

[0088] In this way, in this example, a highly accurate prediction model R13 is generated. Then, using such a prediction model R13, it is possible to quantify the sensitivity of the manipulated variables and calibration parameters to the simulation results. Note that the sensitivity is, for example, like a weight value for weighting, and may also be called a weight.

[0089] (Second Process) In the second process, the conditions of the actual machine (conditions of the glass melting furnace 311) at any point in time are introduced into the prediction model R13 to estimate the values ​​of the actual machine.

[0090] 9A shows a glass melting furnace model (glass melting furnace model M1), three heat sources H1 to H3, three temperatures I1 to I3, and two calibration parameters J1 to J2. Although shown in two dimensions, the glass melting furnace model M1 is a three-dimensional model (e.g., a model of an XYZ Cartesian coordinate system). For example, the temperatures I1 to I3 are the temperatures of multiple locations (observation points) arranged three-dimensionally. The number of heat sources, temperatures, and calibration parameters in this example are merely examples and are not limited to these.

[0091] Based on the results of the first process, the change sensitivity evaluation unit 276 can quantify the change sensitivity value (sensitivity coefficient) of each variable from the simulation results. The observation point data acquisition unit 277 acquires data of each observation point (in this example, data on the temperature of the molten glass).

[0092] 9A, sensitivity w1 is the sensitivity of heat source H1 to temperature I1, sensitivity w2 is the sensitivity of heat source H1 to temperature I2, and sensitivity w3 is the sensitivity of heat source H1 to temperature I3. Note that the other heat sources (heat source H2, heat source H3) also have sensitivities for each of the multiple temperatures I1 to I3, but the method of determining the sensitivities is the same as for heat source H1, and therefore illustrations and detailed explanations are omitted in this example.

[0093] 9A, sensitivity w11 is the sensitivity of calibration parameter J1 to temperature I1, sensitivity w12 is the sensitivity of calibration parameter J1 to temperature I2, and sensitivity w13 is the sensitivity of calibration parameter J1 to temperature I3. Note that the other calibration parameter (calibration parameter J2) also has a sensitivity for each of the multiple temperatures I1 to I3, but the method of determining the sensitivity is the same as for calibration parameter J1, and therefore illustrations and detailed explanations are omitted in this example.

[0094] 9A, the heat quantities of the heat sources H1 to H3, the temperatures I1 to I3, and the calibration parameters J1 to J2 are known in the experimental design method, and only the sensitivity is unknown.

[0095] Figure 9B shows a diagram illustrating the values ​​of each sensitivity for the configuration shown in Figure 9A. In the example of Figure 9B, sensitivity w1 is 0.6, sensitivity w2 is 0.3, sensitivity w3 is 0.1, sensitivity w11 is 0.4, sensitivity w12 is 0.4, and sensitivity w13 is 0.2, with the other sensitivities being as shown.

[0096] In the actual device, the heat quantities and temperatures I1 to I3 of the heat sources H1 to H3 are known, and only the calibration parameters J1 to J2 are unknown. Therefore, the calibration parameters J1 to J2 are calculated using the obtained sensitivities. As a result, the calibration parameter estimation unit 272 estimates (estimately calculates) the calibration parameters J1 to J2.

[0097] A specific example of the calibration parameter estimation process according to the embodiment will be described. For convenience of explanation, the process will be divided into a first mathematical expression part K1 and a second mathematical expression part K2. The following steps are performed in this order: (Process Q1) to (Process Q4).

[0098] (Process Q1) In this example, it is assumed that a simulation is performed under several tens to several hundreds of conditions using the design of experiments. The conditions include the operating conditions of the glass melting furnace 311 and the conditions of the calibration parameters, that is, the values ​​of the calibration parameters are also used as variables.

[0099] In the first mathematical expression section K1, symbols with numerical values ​​attached to x (x1, x2, x3, ...) represent manipulated variables, and their coefficients (a1, a2, ..., b1, b2, ..., c1, c2, ..., etc.) represent the respective sensitivities. Note that manipulated variables may include, for example, variables related to gas, EB output, bubbler flow rate, etc.

[0100] Symbols with numerical values ​​added to p (p1, p2, p3, ...) represent calibration parameters, and their coefficients (α1, α2, ..., β1, β2, ..., γ1, γ2, ..., etc.) represent the respective sensitivities. The calibration parameters may include, for example, the amount of heat radiation from the furnace wall or the amount of tertiary air. At this stage, the values ​​used in the experimental design are substituted for the calibration parameters.

[0101] Furthermore, the symbols (T1, T2, T3, ...) each consisting of a T followed by a numerical value represent the temperature at the observation point to be compared with the actual results. The temperature may be, for example, the base temperature or the peak temperature.

[0102] (Process Q2) Here, the prediction model R13 is generated so that the same number of equations as the number of observation points (number of temperatures) to be matched can be established in the first formula section K1. The first formula section K1 has an equation in which the sum of the multiplication results of each manipulated variable and each coefficient and the multiplication results of each calibration parameter and each coefficient is equal to the temperature of the observation point, and such equations are provided for multiple combinations of coefficients. In the first formula section K1, each sensitivity (each coefficient) is unknown, and each sensitivity is determined by performing regression analysis using a data set obtained by simulation. The same number of regression analyses are performed as there are temperature locations to be matched. In other words, as a specific example, if there are five temperature locations to be matched, five regression analyses are performed.

[0103] In this example, a linear model is used for each equation in the first mathematical expression section K1, but a nonlinear model may also be used as an alternative. Furthermore, other analytical techniques may be used instead of regression analysis, such as Bayesian estimation instead of maximum likelihood estimation.

[0104] (Process Q3) Next, after each sensitivity is calculated in the first equation section K1, values ​​based on actual machine data are substituted for values ​​other than the calibration parameters (p1, p2, p3, ...). In this case, terms related to the manipulated variables (x1, x2, x3, ...) are constant terms, and the calibration parameters (p1, p2, p3, ...) are unknown. The left side of the second equation section K2 is assumed to be the product of a matrix containing all coefficients related to the multiple calibration parameters as elements and a vector of the multiple calibration parameters. The right side of the second equation section K2 is assumed to be the product of terms (ΔT1, ΔT2, ΔT3, ...). The matrix shown in the second equation section K2 then reduces to a problem of solving a system of simultaneous equations. Here, the terms (ΔT1, ΔT2, ΔT3, ...) on the right side of the second equation section K2 are constants. In this example, it is optional whether or not to use data on time (for example, date or time, or date and time, etc.) as actual machine data.

[0105] (Process Q4) In this example, unknown calibration parameters are estimated by the maximum likelihood estimation method or the Bayesian estimation method.

[0106] In this embodiment, the temperature of the molten glass is used as the criterion for adjusting the calibration parameters (i.e., adjusting them to appropriate values). However, as other examples, items such as the resistivity of the molten glass, the current value between the heating electrodes, the voltage value between the heating electrodes, the resistance value between the heating electrodes, and the flow rate of the molten glass may be used, or any combination of two or more items may be used.

[0107] <Example of Estimation Results of Calibration Parameters> Fig. 10 is a diagram showing a table 3211 showing an example of estimation results of calibration parameters according to an embodiment. Here, in this example, an example of calibration parameters is shown in table format, but a table format does not necessarily have to be used. Table 3211 shows an outline of the estimation results of the calibration parameters. Table 3211 stores symbols (p1, p2, ...) representing the calibration parameters, calibration parameter items, and estimated values ​​of the calibration parameters in association with each other. Note that the information stored in table 3211 is an example showing an outline and is not actual information.

[0108] In the example of FIG. 10 , for example, for a calibration parameter of an item called "heat dissipation AAA" represented by the symbol "p1", the estimated value is stored as "... (where "..." represents the corresponding numerical value, and the same applies below)." Also, for example, for a calibration parameter of an item called "dissipation rate AA" represented by the symbol "p11", the estimated value is stored as "...." Also, for example, for a calibration parameter of an item called "air volume aaa" represented by the symbol "p21", the estimated value is stored as "...." Similarly, in the example of FIG. 10 , information about calibration parameters of other items is stored.

[0109] In the example of FIG. 10, the estimated values ​​of the respective calibration parameters are estimated within the performance range, and are values ​​included in the range between the minimum value (Min) and the maximum value (Max) related to each constraint shown in FIG. 7.

[0110] FIG. 11 is a diagram illustrating a comparison between the characteristics of a simulation result using the estimated results of calibration parameters according to the embodiment and the characteristics of actual machine data. This comparison is not an actual result, but is a schematic representation for illustrative purposes. The example of FIG. 11 illustrates a comparison example for temperatures at predetermined locations. In the graph shown in FIG. 11 , the horizontal axis represents the temperature measurement point, and the vertical axis represents the temperature. Note that specific scales are omitted from the horizontal and vertical axes. In this graph, the temperature characteristics of the simulation results are represented by points, and the temperature characteristics of the actual machine (characteristic 3521) are represented by lines. Note that in the example of FIG. 11 , for ease of illustration, only one point (characteristic 3511) of the multiple simulation result points is labeled. In the example of FIG. 11 , the simulation results are able to generally simultaneously reproduce the temperature levels and distributions at multiple observation points of the actual machine temperature characteristic 3521, resulting in highly accurate characteristics.

[0111] As described above, in the glass manufacturing system 1 according to this embodiment, for example, automatic simulation of glass manufacturing can be realized, thereby making the support provided by simulation more efficient than conventional techniques (for example, when simulations are performed by process engineers and dedicated analysts).

[0112] For example, in conventional technology (e.g., when simulations are performed by process engineers and dedicated analysts), heat dissipation rates and the like are set based on actual furnace information by comparing actual results over a period of about one to two months, and then simulations are run for about one to two weeks to estimate the molten glass flow results in a glass melting furnace, which requires a considerable amount of time. In contrast, the information processing device 111 according to the present embodiment can set heat dissipation rates and the like based on actual furnace information by comparing actual results over a short period of time (e.g., several tens of minutes), and then run simulations for a short period of time (e.g., several hours) to estimate the molten glass flow results in a glass melting furnace. Note that these required times are illustrative examples and are not necessarily limited to these times. As a result, the glass manufacturing system 1 according to the present embodiment allows anyone to obtain evaluation results of the molten glass circulation state in a timely manner.

[0113] In the glass manufacturing system 1 according to the present embodiment, when an automatic simulation is executed in the information processing device 111, the values ​​of the calibration parameters of the simulation are estimated and set to appropriate values, thereby improving the accuracy of the simulation even when, for example, the glass melting furnace has changed over time. In the present embodiment, for example, when glass is manufactured using a glass melting furnace by the float method or the like, operation support for the glass manufacturing process can be provided. Generally, in the raw material melting process of glass manufacturing, it is important to grasp and manage the state of the glass melting furnace, and this support is possible in the present embodiment.

[0114] The glass manufacturing system 1 according to this embodiment performs, for example, a time-series data management process, an actual machine data acquisition process, a calibration parameter estimation process, and a state estimation simulation process. In this embodiment, the time-series data management process is performed by the glass manufacturing apparatus 131 and the database 112, and the other processes are performed by the information processing device 111.

[0115] In the time-series data management step, measurement data from multiple sensors installed inside or outside a glass melting furnace (e.g., a melting tank) to measure various quantities related to molten glass is collected, and the collected measurement data is stored in a database as time-series data. Note that other forms of data may be used instead of time-series data (time-series measurement data). The measurement data is, for example, measurement results from a glass manufacturing process using an actual machine (an actual glass manufacturing device). In the actual machine data acquisition step, data (actual machine data) necessary for subsequent processing is acquired from the time-series data stored in the time-series data management step. In the calibration parameter estimation step, calibration parameters of a simulation model (a simulation model that models a melting furnace) are estimated using a calibration parameter estimation model and the actual machine data acquired in the actual machine data acquisition step. In the state estimation simulation step, a state of the actual machine is estimated using the simulation model, the calibration parameters estimated in the calibration parameter estimation step, and the time-series data stored in the time-series data management step (e.g., the actual machine data acquired in the actual machine data acquisition step).

[0116] The glass manufacturing system 1 according to the present embodiment may perform a calibration parameter estimation model construction process. In the calibration parameter estimation model construction process, a calibration parameter estimation model is constructed to estimate calibration parameters of a simulation model (a simulation model that models a melting furnace).

[0117] The glass manufacturing system 1 according to the present embodiment may perform an optimization calculation process, in which an optimization calculation or a sensitivity analysis is performed using a simulation model or a summary model constructed based on the analysis results of the simulation model, based on a proposed operation change guideline considered based on the results (simulation results) output by the state estimation simulation process.

[0118] The glass manufacturing system 1 according to the present embodiment may perform a visualization process. In the visualization process, predetermined information is visualized (e.g., displayed). The predetermined information may be, for example, one or both of data generated as a processing result of the state estimation simulation process and data generated as a processing result of the optimization calculation process, or may be other data.

[0119] Here, various data may be used as data (e.g., time-series data) that can be used for estimation, such as temperature data measured by a permanent thermocouple, data on the operation period of the melting furnace, etc. For example, in the calibration parameter estimation step, in this embodiment, even if the input information is only date and time information, it is possible to estimate the calibration parameters.

[0120] In addition, various data may be used as the calibration parameters to be estimated, for example, data on the heat radiation amount of the furnace wall or data on the amount of erosion of the furnace material (for example, the base tiles, throat, etc.) may be used. In addition, any method may be used as the estimation method, for example, a method using a regression model using data from a kiln opening survey, etc. may be used.

[0121] Furthermore, various modes may be used as the simulation model (method), and for example, a model (method) for performing an analysis by coupling two or more elements may be used. Note that the model (method) may be, for example, two-dimensional or three-dimensional, and may be a steady-state analysis or a non-steady-state analysis. The elements may be, for example, a raw material melting simulation that models the melting behavior of glass raw materials, a thermal fluid simulation of molten glass, a combustion simulation of the upper combustion space, a simulation of the potential and current distribution in molten glass by electric heating equipment, a heat transfer simulation and steady-state analysis inside the furnace material, etc.

[0122] Alternatively, for example, a model that summarizes the original simulation model (summary model) may be used instead of the original simulation model. As an example of reduction, a reduced order model (ROM) may be used.

[0123] As a simulation model (method), for example, 1D-CAE, a circuit network model, or a system model may be used. As a simulation model (method), for example, a deep learning method such as PINNS (Physics Informed Neural Nets) may be used.

[0124] Furthermore, various execution environments may be used as the execution environment for the simulation, such as a client PC and workstation, a cluster machine, or a cloud service.

[0125] The glass manufacturing system 1 may perform a prediction model generation process in which a prediction model of simulation results is generated using, for example, design of experiments and machine learning. In the prediction model generation process, the operating conditions of the glass melting furnace and simulation calibration parameters are set as setting variables, and multiple simulations are performed based on the design of experiments within a range that comprehensively encompasses conditions expected in an actual glass manufacturing process. In the prediction model generation process, a machine learning model is constructed by performing machine learning using data from the simulation results, and the sensitivity (sensitivity to change) of the operating conditions of the glass melting furnace and the simulation calibration parameters to a predetermined value at multiple observation points in the glass melting furnace is evaluated. Note that the predetermined value may be, for example, the molten glass temperature, the molten glass flow rate, or both of these values. In the prediction model generation process, for example, data from the simulation results may be stored in a database. In this case, the machine learning model is constructed by performing machine learning using the data stored in the database.

[0126] In the glass manufacturing system 1 according to the present embodiment, for example, a calibration parameter determination process may be performed in which actual machine data is input into the prediction model to determine calibration parameters (unknown variables) for the simulation. Here, data related to an actual glass melting furnace is used. This actual glass melting furnace has, for example, the same shape as the glass melting furnace assumed in the simulation (or a shape similar enough to fit the simulation), and multiple sensors are installed to measure various quantities related to the molten glass.

[0127] In the calibration parameter determination step, operating conditions at any point in time and predetermined values ​​at that time (in this embodiment, at least one of the molten glass temperature and the molten glass flow velocity) are acquired at multiple observation points. In addition, in the calibration parameter determination step, the calibration parameters are optimized to the most probable values ​​using a computer based on the results of sensitivity evaluation so that the deviation between each predetermined value acquired at the above observation points and a value based on the result obtained by simulation (a value corresponding to the predetermined value) is minimized (if there are multiple values, they are minimized simultaneously). In this way, the calibration parameters are determined.

[0128] The parameter estimation method in this process may be Bayesian estimation instead of maximum likelihood estimation. The estimation algorithm may be not only a machine learning method, but also a calculation method that applies an approximation formula such as a polynomial or exponential function formulated from conventional knowledge about the process.

[0129] In the glass manufacturing system 1 according to the present embodiment, a range of constraints may be set for the values ​​that the estimated results of the calibration parameters can take. For example, a change in the heat radiation amount of the furnace material of the glass melting furnace over time, which is formulated based on experiments or past measurement results, may be calculated or reflected, and different constraints may be imposed on the range of the optimized values ​​depending on the time period. The constraints may be, for example, a minimum value, a maximum value, or both.

[0130] Second Embodiment A second embodiment will be described.

[0131] [Modification: Example in which Calibration Parameter Estimation is Omitted] In the first embodiment, the case where the process of step S2 shown in FIG. 4 (the process of estimating the calibration parameters) is performed has been described. However, as a modification, a configuration in which the process of step S2 (the process of estimating the calibration parameters) is not performed may be used. In this case, the information processing device 111 does not need to be provided with functions and data for estimating the calibration parameters. In this modification, the process of step S3 is performed after the process of step S1 in the processing flow shown in FIG. In this modification, the process of estimating the calibration parameters is not performed, and, for example, values ​​corresponding to the calibration parameters are maintained constant.

[0132] Fig. 12 is a diagram schematically illustrating an example of characteristics based on simulation results when the calibration parameter estimation process is not performed. Note that the example in Fig. 12 is not necessarily an actual example, but is merely a rough example. Fig. 12 shows, for a single temperature location, a characteristic 1011 based on simulation results when the calibration parameter estimation process is not performed, and a characteristic 1021 based on actual device data. In the graph shown in Fig. 12, the horizontal axis represents time (aging), and the vertical axis represents temperature. Note that detailed scales are omitted on the vertical and horizontal axes.

[0133] As shown in the example of Figure 12, when the calibration parameter estimation process is not performed, the deviation from the characteristic 1021 based on the actual machine data becomes larger than when the calibration parameter estimation process is performed. In other words, for example, even if the most recent actual conditions are input into a glass melting furnace simulation model fixed with past calibration parameters, the most recent temperature and other conditions of the actual machine cannot be accurately reproduced. The main reason for this is thought to be the significant influence of changes over time in the amount of heat dissipated in the furnace (the amount of heat dissipated in the melting furnace). For example, changes over time increase the surface temperature and cross-sectional temperature of the furnace, which is thought to reduce the accuracy of the simulation results not only for temperature but also for flow velocity and the like.

[0134] Note that, when the influence of such aging changes is not large (for example, when the time is close to the time when the actual results comparison condition is met as shown in Fig. 12 ), it is also effective to configure the system so that the process of step S2 (process of estimating the calibration parameters) shown in Fig. 4 is not performed. As an example, the time when the influence of aging changes is not large can be considered to be a period until several months have passed since the time when the actual results comparison condition is met, but is not necessarily limited to this.

[0135] An example of aging of a glass melting furnace is shown with reference to Fig. 13. Note that these are only examples showing the outline, and detailed explanations will be omitted.

[0136] FIG. 13 is a diagram schematically illustrating an example of aging of furnace wall heat radiation of a glass melting furnace. In the graph shown in FIG. 13, the horizontal axis represents the number of months of operation, and the vertical axis represents furnace wall heat radiation. The graph shows a characteristic 1221 representing the average of the characteristics of multiple furnaces (e.g., five or more furnaces) consisting of glass melting furnaces. In the example of FIG. 13, it can be seen that furnace wall heat radiation increases with aging. Here, by formulating such aging, furnace wall heat radiation at any time may be calculated and used as a calibration parameter. This calibration parameter setting method is not necessarily limited to furnace wall heat radiation and may be applied to other purposes.

[0137] As described above, in this embodiment, it is thought that the accuracy of the simulation will be lower than, for example, a configuration in which calibration parameter estimation processing is performed as in the first embodiment. However, even in this embodiment, by realizing automatic simulation for glass manufacturing, it is possible to improve the efficiency of simulation support compared to conventional techniques (for example, when simulations are performed by process engineers and dedicated analysts).

[0138] For example, in conventional techniques (e.g., when simulations are performed by process engineers and dedicated analysts), heat dissipation rates and the like are set based on actual furnace information and comparisons over a period of about one to two months, and then simulations are run for about one to two weeks to estimate the glass melting furnace's surface roughness results (glass flow results), which requires a considerable amount of time. In contrast, in the present embodiment, heat dissipation rates and the like can be set based on actual furnace information and comparisons over a short period of time (e.g., several tens of minutes), and then simulations are run for a short period of time (e.g., several hours) to estimate the glass melting furnace's surface roughness results (glass flow results). Note that these required times are illustrative examples and are not necessarily limited to these times. As a result, in the present embodiment, anyone can obtain evaluation results of the molten glass circulation state in a timely manner.

[0139] (Regarding the above embodiments) A program for implementing the functions of any of the components of any of the above-described devices may be recorded on a computer-readable recording medium, and the program may be loaded and executed by a computer system. The term "computer system" as used herein includes hardware such as an operating system or peripheral devices. Furthermore, "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs, and compact disc (CD)-read-only memories (ROMs), as well as storage devices such as hard disks built into computer systems. Furthermore, "computer-readable recording medium" also includes devices that retain a program for a certain period of time, such as volatile memory within a computer system that serves as a server or client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. Such volatile memory may be, for example, random access memory (RAM). The recording medium may also be, for example, a non-transitory recording medium.

[0140] The above program may also be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by transmission waves in the transmission medium. Here, the "transmission medium" that transmits the program refers to a medium that has the function of transmitting information, such as a network such as the Internet or a communication line such as a telephone line. The above program may also be intended to realize part of the above-mentioned functions. Furthermore, the above program may be a so-called differential file that can realize the above-mentioned functions in combination with a program already recorded in the computer system. A differential file may also be called a differential program.

[0141] Furthermore, the functions of any of the components in any of the above-described devices may be implemented by a processor. For example, each process in the embodiments may be implemented by a processor that operates based on information such as a program and a computer-readable recording medium that stores information such as the program. Here, the functions of each unit of the processor may be implemented by, for example, individual hardware, or may be implemented by integrated hardware. For example, the processor may include hardware, and the hardware may include at least one of a circuit for processing digital signals and a circuit for processing analog signals. For example, the processor may be configured using one or more circuit devices mounted on a circuit board, or one or both of one or more circuit elements. An integrated circuit (IC) or the like may be used as the circuit device, and a resistor or a capacitor may be used as the circuit element.

[0142] Here, the processor may be, for example, a CPU. However, the processor is not limited to a CPU, and various types of processors, such as a GPU (Graphics Processing Unit) or a DSP (Digital Signal Processor), may be used. The processor may also be, for example, a hardware circuit such as an ASIC (Application Specific Integrated Circuit). The processor may also be, for example, composed of multiple CPUs, or may also be composed of a hardware circuit such as multiple ASICs. The processor may also be, for example, composed of a combination of multiple CPUs and a hardware circuit such as multiple ASICs. The processor may also include, for example, one or more of an amplifier circuit or a filter circuit that processes analog signals.

[0143] The embodiments of this disclosure have been described in detail above with reference to the drawings, but the specific configuration is not limited to this embodiment, and includes designs within the scope that do not deviate from the gist of this disclosure.

[0144] [Note] (Configuration example 1) to (Configuration example 11) are shown.

[0145] (Configuration Example 1) An information processing device including: a state estimation unit that estimates a state of an actual device using measurement data stored in a measurement data storage unit that stores measurement data from a plurality of sensors installed inside or outside a glass melting furnace to measure values ​​related to molten glass, and a glass melting furnace simulation model that models the glass melting furnace.

[0146] (Configuration Example 2) The information processing device according to (Configuration Example 1), further comprising: a calibration parameter estimation model that estimates calibration parameters of the glass melting furnace simulation model; and a calibration parameter estimation unit that estimates the calibration parameters using the measurement data stored in the measurement data storage unit; and the state estimation unit performs the state estimation of the actual machine using the calibration parameters estimated by the calibration parameter estimation unit.

[0147] (Configuration Example 3) The information processing device described in (Configuration Example 1) or (Configuration Example 2), further comprising a model calculation unit that performs optimization calculation or sensitivity analysis calculation using the glass melting furnace simulation model or a summary model based on the glass melting furnace simulation model, based on a proposed operation change guideline based on the result of the state estimation by the state estimation unit.

[0148] (Configuration Example 4) The information processing device according to (Configuration Example 3), further comprising an output unit that outputs information relating to at least one of a result of the state estimation by the state estimation unit or a result of the calculation by the model calculation unit.

[0149] (Configuration Example 5) The information processing device described in (Configuration Example 2), wherein the calibration parameter estimation unit comprises: an experimental design execution unit that executes a plurality of simulations for the glass melting furnace simulation model based on experimental design, with the operating conditions of the glass melting furnace and the calibration parameters as set variables, within a range that comprehensively includes conditions expected in an actual glass manufacturing process; a simulation result storage unit that stores results of the simulations by the experimental design execution unit; a prediction model generation unit that performs machine learning using the simulation results stored in the simulation result storage unit to generate a prediction model; and a change sensitivity evaluation unit that evaluates change sensitivity of the operating conditions of the glass melting furnace and the calibration parameters to at least one of the simulation results of the molten glass temperature, the resistivity of the molten glass, the current value, voltage value or resistance value between heating electrodes, the molten glass flow rate, the furnace material temperature, or the atmospheric temperatures inside and outside the furnace, at a plurality of observation points of the glass melting furnace.

[0150] (Configuration Example 6) The information processing device described in (Configuration Example 5), wherein the calibration parameter estimation unit further includes an observation point data acquisition unit that acquires the operating conditions at a predetermined time point for a plurality of observation points in the glass melting furnace and acquires observation point data that is at least one of the molten glass temperature, the resistivity of the molten glass, the current value or voltage value or resistance value between heating electrodes, the molten glass flow rate, the furnace material temperature, or the atmospheric temperature inside and outside the furnace at the predetermined time point, and estimates the calibration parameters based on the evaluation result by the change sensitivity evaluation unit so that a deviation between the observation point data acquired by the observation point data acquisition unit for each of the observation points and the result obtained by the simulation is small.

[0151] (Configuration Example 7) The information processing device according to (Configuration Example 6), wherein the calibration parameter estimation unit restricts a range of values ​​of the calibration parameters based on a change in heat radiation amount of furnace materials due to aging of the glass melting furnace.

[0152] It is possible to provide a glass manufacturing system including the above-described information processing device. (Configuration Example 8) A glass manufacturing system having a glass manufacturing apparatus including a glass melting furnace and an information processing device, wherein a measurement data storage unit is provided inside or outside the information processing device to store measurement data from a plurality of sensors installed inside or outside the glass melting furnace to measure values ​​related to molten glass, and the information processing device includes a glass melting furnace simulation model that models the glass melting furnace, and a state estimation unit that estimates a state of an actual apparatus using the measurement data stored in the measurement data storage unit.

[0153] A method performed by the information processing device as described above can be provided. (Configuration Example 9) An information processing method in which an information processing device estimates a state of an actual device using measurement data stored in a measurement data storage unit that stores measurement data from a plurality of sensors installed inside or outside a glass melting furnace to measure values ​​related to molten glass, and a glass melting furnace simulation model that models the glass melting furnace.

[0154] It is possible to provide a program (computer program) executed by a computer constituting the above-described information processing device. (Configuration Example 10) A program for causing a computer to realize a state estimation function of estimating the state of an actual device using measurement data stored in a measurement data storage unit that stores measurement data from a plurality of sensors installed inside or outside a glass melting furnace to measure values ​​related to molten glass, and a glass melting furnace simulation model that models the glass melting furnace.

[0155] It is possible to provide a calibration parameter estimation method applicable to the above-described information processing devices, etc. (Configuration Example 11) A calibration parameter estimation method comprising: setting operation conditions and calibration parameters of a glass melting furnace as set variables, performing a plurality of simulations of a glass melting furnace simulation model based on an experimental design method within a range that comprehensively includes conditions expected in an actual glass manufacturing process, performing machine learning using simulation results from the simulations to generate a prediction model, and evaluating the sensitivity of the operation conditions of the glass melting furnace and the calibration parameters to changes in at least one of the simulation results of the molten glass temperature, the resistivity of the molten glass, the current value, voltage value, or resistance value between heating electrodes, the molten glass flow rate, the furnace material temperature, or the atmospheric temperatures inside and outside the furnace at a plurality of observation points in the glass melting furnace.

[0156] DESCRIPTION OF SYMBOLS 1...Glass manufacturing system 20...Superstructure 40...Thermocouple 50...Current-carrying electrode 60...Burner 70...Glass material (glass raw material or molten glass) 100...Glass melting furnace 111...Information processing device 112...Database 131...Glass manufacturing apparatus 211...Input unit 212...Output unit 213...Communication unit 214...Storage unit 215...Control unit 231...Operation unit 251...Display unit 271...State estimation unit 272...Calibration parameter estimation unit 273...Model calculation unit 274...Experimental design execution unit 275...Prediction model generation unit 276...Change sensitivity evaluation unit 277...Observation point data acquisition unit 311...Glass melting furnace 331, 351...Sensor unit 3011, 3211...Table 1011, 1021, 1221, 3111, 3511, 3521...Characteristics A1...Actual manufacturing process A2...Simulation process A3...Integrated data platform B1...Actual machine B2...Actual machine data C1...Simulation model D1...Parameter estimation E1, E2...User F1, F2...Computer G1...Server H1-H3...Heat source I1-I3...Temperature J1-J2...Calibration parameters M1...Glass melting furnace model R1...Measurement data R11...Glass melting furnace simulation model R12...Calibration parameter estimation model R13...Prediction model R14...Simulation results U1...Universal set of conditions V1...Condition W1...Simulation result group w1-w3, w11-w13...Sensitivity

Claims

1. An information processing apparatus comprising: a measurement data storage unit that stores measurement data of a plurality of sensors installed inside or outside a glass melting furnace for measuring values related to molten glass; and a state estimation unit that performs state estimation of an actual machine using a glass melting furnace simulation model that models the glass melting furnace.

2. The information processing apparatus according to claim 1, further comprising: a calibration parameter estimation model that estimates calibration parameters of the glass melting furnace simulation model; and a calibration parameter estimation unit that estimates the calibration parameters using the measurement data stored in the measurement data storage unit, wherein the state estimation unit performs the state estimation of the actual machine using the calibration parameters estimated by the calibration parameter estimation unit.

3. The information processing apparatus according to claim 1 or 2, further comprising: a model calculation unit that performs an optimization calculation or a sensitivity analysis calculation using the glass melting furnace simulation model or a summary model based on the glass melting furnace simulation model, based on an operation change guideline proposal based on a result of the state estimation by the state estimation unit.

4. The information processing apparatus according to claim 3, further comprising: an output unit that outputs information related to at least one of a result of the state estimation by the state estimation unit or a result of the calculation by the model calculation unit.

5. The calibration parameter estimation unit sets the operation conditions of the glass melting furnace and the calibration parameters as set variables, and executes a plurality of simulations on the glass melting furnace simulation model based on the experimental design method within a range comprehensively including the conditions assumed in the actual glass manufacturing process. An experimental design method execution unit, a simulation result storage unit that stores the simulation results by the experimental design method execution unit, and a prediction model generation unit that performs machine learning using the simulation results stored in the simulation result storage unit to generate a prediction model. And a change sensitivity evaluation unit that evaluates the change sensitivity of at least one of the melting glass temperature, the specific resistance of the melting glass, the current value, voltage value, or resistance value between the heating electrodes, the melting glass flow rate, the furnace material temperature, or the furnace internal and external atmosphere temperature at a plurality of observation points of the glass melting furnace with respect to the operation conditions and the calibration parameters of the glass melting furnace. The information processing apparatus according to claim 2, comprising:

6. The calibration parameter estimation unit further includes an observation point data acquisition unit that acquires the operation conditions at a predetermined time and observation point data that is at least one of the melting glass temperature, the specific resistance of the melting glass, the current value, voltage value, or resistance value between the heating electrodes, the melting glass flow rate, the furnace material temperature, or the furnace internal and external atmosphere temperature at the predetermined time for a plurality of the observation points in the glass melting furnace. Based on the evaluation result by the change sensitivity evaluation unit, the calibration parameter is estimated so that the deviation between the observation point data acquired by the observation point data acquisition unit for each observation point and the result obtained by the simulation is reduced. The information processing apparatus according to claim 5.

7. The calibration parameter estimation unit restricts the range of values of the calibration parameters based on the change in the heat dissipation amount of the furnace material over time of the glass melting furnace. The information processing apparatus according to claim 6.

8. A glass manufacturing system having a glass manufacturing apparatus including a glass melting furnace and an information processing apparatus, the information processing apparatus including or being externally provided with a measurement data storage unit that stores measurement data of a plurality of sensors installed inside or outside the glass melting furnace for measuring values related to molten glass, the information processing apparatus including a state estimation unit that performs state estimation of an actual machine using a glass melting furnace simulation model that models the glass melting furnace and the measurement data stored in the measurement data storage unit. A glass manufacturing system.

9. An information processing method in which an information processing apparatus performs state estimation of an actual machine using the measurement data stored in a measurement data storage unit that stores measurement data of a plurality of sensors installed inside or outside a glass melting furnace for measuring values related to molten glass and a glass melting furnace simulation model that models the glass melting furnace.

10. A program for causing a computer to realize a state estimation function for performing state estimation of an actual machine using the measurement data stored in a measurement data storage unit that stores measurement data of a plurality of sensors installed inside or outside a glass melting furnace for measuring values related to molten glass and a glass melting furnace simulation model that models the glass melting furnace.

11. A calibration parameter estimation method in which operating conditions and calibration parameters of a glass melting furnace are set as variables, a plurality of simulations are performed on a glass melting furnace simulation model based on an experimental design method within a range comprehensively including conditions assumed in an actual glass manufacturing process, machine learning is performed using simulation results from the simulations to generate a prediction model, and the sensitivity of changes in the operating conditions and calibration parameters of the glass melting furnace to at least one of the simulation results of the molten glass temperature, specific resistance of the molten glass, current value, voltage value, or resistance value between heating electrodes, molten glass flow rate, furnace material temperature, or furnace internal and external ambient temperature at a plurality of observation points of the glass melting furnace is evaluated.

Images