JP7746673B2 - Parameter specifying device and parameter specifying method - Google Patents

Description

The present invention relates to an apparatus and method for identifying parameters for maintaining appropriate sand properties in a mixing system that mixes foundry sand to produce mixed sand used in molding molds.

Technologies for analyzing particle characteristics are known. For example, Patent Document 1 describes a particle analyzer that runs multiple image processing algorithms (morphological operations, Hough transform, binarization, etc.) on raw images of particles and displays a list of test result images for each different image processing algorithm. Furthermore, Patent Document 2 describes technology that optimizes a blending plan for an optimization period and, based on the solution obtained, creates a blending plan from the results of simulating the inventory trends and quality/properties of each raw material.

WO2017/195785 publication (published on November 16, 2017) JP 2009-169823 A (published July 30, 2009)

In molding lines that produce casting molds, it is important to maintain the sand properties (e.g., compactability) of the mixed sand appropriately. Because the sand properties of the mixed sand change due to fluctuations in various parameters related to the mixing system (e.g., sand temperature of the mixed sand, ambient temperature, humidity, etc.), it can be difficult to maintain the sand properties at appropriate values. Even with the above-mentioned conventional technology, it was difficult to maintain appropriate sand properties in the mixing system.

One aspect of the present invention aims to realize a technology for identifying parameters to maintain the target sand properties of mixed sand in a mixing system that mixes foundry sand to produce mixed sand.

To solve the above problems, a parameter identification device according to one aspect of the present invention includes one or more processors that execute a parameter identification step. Furthermore, a parameter identification method according to one aspect of the present invention includes a parameter identification step.

In the parameter identification device and the parameter identification method, the identification step is a step in which the processor identifies the remaining parameters from a portion of a plurality of parameters related to a mixing system that mixes foundry sand, water, and additives to produce mixed sand, including parameters that represent the sand properties of the mixed sand.

According to one aspect of the present invention, in a mixing system that mixes foundry sand to produce mixed sand, parameters can be identified to maintain the desired sand properties of the mixed sand.

1 is a diagram showing the configuration of a sand property control system according to a first embodiment of the present invention. FIG. FIG. 2 is a block diagram showing the configuration of a parameter specifying device included in the sand property control system of FIG. 1. 3 is a flowchart showing the flow of a parameter specifying method performed by the parameter specifying device of FIG. 2 . FIG. 1 is a diagram illustrating a genetic algorithm. 10 is a flowchart showing the flow of a relational expression specifying step performed by the parameter specifying device. 10 is a flowchart showing the flow of processing performed by a parameter specifying device. FIG. 10 is a diagram illustrating the contents of a crossover. FIG. 10 illustrates the contents of a subtree mutation. FIG. 1 illustrates the contents of a hoist mutation. FIG. 1 is a diagram illustrating the contents of point mutations. FIG. 10 is a diagram showing the relationship between the actual measured value and the predicted value of compactability. FIG. 10 is a diagram showing the configuration of a sand property control system according to a second embodiment of the present invention. 13 is a flowchart showing the flow of a parameter specifying method implemented by a parameter specifying device included in the sand property control system of FIG. 12. FIG. 1 illustrates an example of a trained model and a genetic algorithm. FIG. 13 is a block diagram showing the configuration of a machine learning device included in the sand property control system of FIG. 12. 16 is a flowchart showing the flow of a machine learning method performed by the machine learning device of FIG. 15 .

[First embodiment]
(Configuration of sand property control system)
The configuration of a sand property control system S1 according to a first embodiment of the present invention will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the configuration of the sand property control system S1.

The sand property control system S1 is a system for controlling the properties of mixed sand. As shown in Figure 1, the sand property control system S1 includes a mixing system 1, a data logger 2, and a parameter identification device 3.

The mixing system 1 is a system for producing mixed sand by mixing foundry sand, water, and additives used in molding a casting mold. In this embodiment, the mixing system 1 is composed of a foundry sand property measuring device 10, a foundry sand charging device 11, an additive charging device 12, a water injection device 13, a mixer 14, a mixed sand property measuring device 15, and a group of sensors 16.

The molding sand property measuring device 10 is a device for measuring the sand properties of molding sand (hereinafter also referred to as "molding sand properties"). Examples of molding sand properties include the temperature and moisture content of the molding sand. Note that the moisture content of molding sand refers to, for example, the weight or volume of water contained in a unit weight or unit volume of molding sand. The molding sand property measuring device 10 provides the measured molding sand properties to the data logger 2.

The foundry sand charging device 11 is a device for charging a preset amount of foundry sand (hereinafter also referred to as the "foundry sand charging amount") into the mixer 14. Here, the foundry sand charging amount refers to, for example, the weight or volume of foundry sand charged into the mixer 14 per unit time (in the case of continuous processing) or per processing (in the case of batch processing). The foundry sand charging device 11 provides the set foundry sand charging amount to the data logger 2.

The additive dosing device 12 is a device for dosing a preset amount of additive (hereinafter also referred to as the "additive dosing amount") into the kneader 14. Here, the additive dosing amount refers to, for example, the weight or volume of additive added to the kneader 14 per unit time (in the case of continuous processing) or per processing (in the case of batch processing). The additive dosing device 12 provides the measured additive dosing amount to the data logger 2.

The water injection device 13 is a device for injecting a preset amount of water (hereinafter also referred to as the "water input amount") into the mixer 14. Here, the water input amount refers to, for example, the weight or volume of water input into the mixer 14 per unit time (in the case of continuous processing) or per processing (in the case of batch processing). The water injection device 13 provides the set water input amount to the data logger 2.

The kneader 14 is a device that kneads foundry sand, water, and additives according to preset kneading conditions. Examples of kneading conditions include the rotation speed of the kneading blade (in the case of continuous processing and batch processing) and the rotation time (in the case of batch processing). The rotation speed of the kneading blade refers to, for example, the number of times the kneading blade rotates per unit time. The rotation time of the kneading blade refers to, for example, the time it takes for the kneading blade to rotate in one processing run. The kneader 14 provides the set kneading conditions to the data logger 2.

The mixed sand property measuring device 15 is a device for measuring the sand properties of mixed sand (hereinafter also referred to as "mixed sand properties"). Examples of mixed sand properties include the temperature, moisture content, compactibility (CB value), and air permeability of the mixed sand. The moisture content of mixed sand refers to, for example, the weight or volume of water contained in a unit weight or unit volume of mixed sand. The compactability of mixed sand is a value that represents the moisture state of the surface layer of mixed sand particles and is a value used as a green sand evaluation index. The air permeability of mixed sand refers to, for example, the ease with which gases (hydrogen, oxygen, nitrogen, carbon monoxide, carbon dioxide, hydrocarbons, etc.) can pass through the mixed sand escaping from the mold. The mixed sand property measuring device 15 provides the measured mixed sand properties to the data logger 2.

The sensor group 16 is a collection of sensors for measuring external environmental variables of the kneading system 1. Examples of external environmental variables include the temperature and humidity of the room in which the kneading system 1 is installed. Each sensor that makes up the sensor group 16 provides the measured external environmental variables to the data logger 2.

The data logger 2 collects parameters related to the mixing system 1 (hereinafter also referred to as "mixing parameters") and provides the collected mixing parameters to the parameter identification device 3. As described above, the mixing parameters include molding sand properties (molding sand temperature and moisture content), molding sand input amount, additive input amount, moisture input amount, mixing conditions, mixed sand properties (mixed sand temperature, moisture content, compactibility, and air permeability), and external environmental variables (external environmental temperature and humidity).

The amount of foundry sand added, the amount of additives added, the amount of water added, and the mixing conditions are set values and are controllable mixing parameters. Hereinafter, controllable mixing parameters will be referred to as mixing parameters x1, x2, ..., xn. On the other hand, the foundry sand properties, mixed sand properties, and external environmental variables are measured values and are uncontrollable mixing parameters. Hereinafter, uncontrollable parameters other than the mixed sand properties will be referred to as y1, y2, ..., ym, and the mixed sand properties will be referred to as z. Regarding the mixed sand properties, any one of the temperature, water content, compactibility, and air permeability of the mixed sand is treated as the mixed sand property z. In particular, in this embodiment, compactability is treated as the mixed sand property z. Regarding the foundry sand properties, either the temperature or water content of the foundry sand may be treated as an uncontrollable mixing parameter, or both the temperature and water content of the foundry sand may be treated as uncontrollable mixing parameters. Furthermore, with regard to the external environmental variables, either the temperature or humidity of the external environment may be treated as an uncontrollable kneading parameter, or both the temperature and humidity of the external environment may be treated as uncontrollable kneading parameters.

The parameter identification device 3 records the set of kneading parameters (x1, x2, ..., xn, y1, y2, ..., ym, z) acquired from the data logger 2 each time kneading is performed. When kneading is repeated k times, the parameter identification device 3 stores the set of kneading parameters PS1 acquired in the first kneading, the set of kneading parameters PS2 acquired in the second kneading, ..., the set of kneading parameters PSk acquired in the kth kneading. The parameter identification device 3 then performs a parameter identification method M1, which includes a relational equation identification step M11 and a parameter identification step M14. The relational equation identification step M11 is a step in which the nonlinear relational equation z = f(x1, x2, ..., xn, y1, y2, ..., ym) that represents the relationship between the kneading parameters x1, x2, ..., xn, y1, y2, ..., ym, z is identified by referring to the accumulated parameter sets PS1 to PSk. Parameter identification step M14 is a step in which, using the relational equation z = f(x1, x2, ..., xn, y1, y2, ..., ym) identified in relational equation identification step M11, the values to which the controllable mixing parameters x1, x2, ..., xn should be set in order to make the mixed sand property z match the target value. The configuration of the parameter identification device 3 and the flow of parameter identification method M1 will be described later with reference to different drawings.

In this embodiment, the nonlinear relational expression expressing the relationship between the mixing parameters x1, x2, ..., xn, y1, y2, ..., ym, and z uses the relational expression z = f(x1, x2, ..., xn, y1, y2, ..., ym), with the mixed sand property z as the objective variable. However, the present invention is not limited to this. In other words, the nonlinear relational expression expressing the relationship between the mixing parameters x1, x2, ..., xn, y1, y2, ..., ym, and z may use a relational expression with a mixing parameter other than the mixed sand property z as the objective variable.

(Configuration of parameter specifying device)
The configuration of the parameter specifying device 3 will be described with reference to Fig. 2. Fig. 2 is a block diagram showing the configuration of the parameter specifying device 3.

The parameter identification device 3 is implemented using a general-purpose computer and includes a processor 31, a primary memory 32, a secondary memory 33, an input/output interface 34, a communication interface 35, and a bus 36. The processor 31, primary memory 32, secondary memory 33, input/output interface 34, and communication interface 35 are interconnected via the bus 36.

The secondary memory 33 stores a parameter identification program P2 and a kneading parameter set PS. The processor 31 loads the parameter identification program P2 and the kneading parameter set PS stored in the secondary memory 33 onto the primary memory 32. The processor 31 then executes each step included in the parameter identification method M1 in accordance with the instructions included in the parameter identification program P2 loaded onto the primary memory 32. The kneading parameter set PS loaded onto the primary memory 32 is used when the processor 31 executes the relational equation identification step M11 (described below) of the parameter identification method M1. Note that the parameter identification program P2 being stored in the secondary memory 33 means that the source code or an executable file obtained by compiling the source code is stored in the secondary memory 33.

Devices that can be used as processor 31 include, for example, a CPU (Central Processing Unit), GPU (Graphics Processing Unit), DSP (Digital Signal Processor), MPU (Micro Processing Unit), FPU (Floating Point Number Processing Unit), PPU (Physics Processing Unit), microcontroller, or a combination of these. Processor 31 is sometimes called a "computing device."

Devices that can be used as the primary memory 32 include, for example, semiconductor RAM (Random Access Memory). The primary memory 32 is sometimes referred to as a "main storage device." Devices that can be used as the secondary memory 33 include, for example, flash memory, HDD (Hard Disk Drive), SSD (Solid State Drive), ODD (Optical Disk Drive), FDD (Floppy (registered trademark) Disk Drive), or a combination thereof. The secondary memory 33 is sometimes referred to as an "auxiliary storage device." The secondary memory 33 may be built into the parameter identification device 3, or may be built into another computer (e.g., a computer constituting a cloud server) connected to the parameter identification device 3 via the input/output interface 34 or the communication interface 35. In this embodiment, the parameter identification device 3 uses two memories (the primary memory 32 and the secondary memory 33) to store data. However, this is not limiting. In other words, the parameter identification device 3 may use a single memory to store data. In this case, for example, one storage area of the memory can be used as primary memory 32, and another storage area of the memory can be used as secondary memory 33.

Input and/or output devices are connected to the input/output interface 34. Examples of the input/output interface 34 include interfaces such as USB (Universal Serial Bus), ATA (Advanced Technology Attachment), SCSI (Small Computer System Interface), and PCI (Peripheral Component Interconnect). An example of an input device connected to the input/output interface 34 is a data logger 2. Data acquired in parameter identification method M1 is input to the parameter identification device 3 via the data logger 2 and stored in the primary memory 32. Examples of input devices connected to the input/output interface 34 include a keyboard, mouse, touchpad, microphone, or a combination thereof. Examples of output devices connected to the input/output interface 34 include a display, projector, printer, speakers, headphones, or a combination thereof. Information provided to the user in parameter identification method M1 is output from the parameter identification device 3 via these output devices. Note that the parameter identification device 3 may have a built-in keyboard that functions as an input device and a built-in display that functions as an output device, like a laptop computer. Alternatively, the parameter identification device 3 may have a built-in touch panel that functions as both an input device and an output device, like a tablet computer.

Other computers are connected to the communication interface 35 via a network, either wired or wirelessly. Examples of the communication interface 35 include interfaces such as Ethernet (registered trademark) and Wi-Fi (registered trademark). Available networks include a PAN (Personal Area Network), LAN (Local Area Network), CAN (Campus Area Network), MAN (Metropolitan Area Network), WAN (Wide Area Network), GAN (Global Area Network), or an internetwork including these networks. The internetwork may be an intranet, an extranet, or the Internet. Data provided by the parameter identification device 3 to other computers is sent and received via these networks.

Note that, while this embodiment employs a configuration in which parameter identification method M1 is executed using a single processor (processor 31), the present invention is not limited to this. That is, a configuration in which parameter identification method M1 is executed using multiple processors may also be employed. In this case, the multiple processors that cooperate to execute parameter identification method M1 may be provided on a single computer and configured to be able to communicate with each other via a bus, or may be distributed across multiple computers and configured to be able to communicate with each other via a network. As an example, a processor built into a computer that constitutes a cloud server and a processor built into a computer owned by a user of that cloud server may cooperate to execute parameter identification method M1.

Furthermore, in this embodiment, the kneading parameter set PS is stored in memory (secondary memory 33) built into the same computer as the processor (processor 31) that executes parameter identification method M1, but the present invention is not limited to this. That is, the kneading parameter set PS may be stored in memory built into a computer different from the processor that executes parameter identification method M1. In this case, the computer incorporating the memory that stores the kneading parameter set PS is configured to be able to communicate with the computer incorporating the processor that executes parameter identification method M1 via a network. As an example, the kneading parameter set PS may be stored in memory built into a computer that constitutes a cloud server, and parameter identification method M1 may be executed by a processor built into a computer owned by a user of the cloud server.

Furthermore, while this embodiment employs a configuration in which the kneading parameter set PS is stored in a single memory (secondary memory 33), the present invention is not limited to this. That is, a configuration in which the kneading parameter set PS is distributed and stored in multiple memories may also be employed. In this case, the multiple memories storing the kneading parameter set PS may be provided in a single computer (which may or may not be a computer incorporating a processor that executes parameter identification method M1), or may be distributed and stored in multiple computers (which may or may not include a computer incorporating a processor that executes parameter identification method M1). As an example, a configuration in which the kneading parameter set PS is distributed and stored in memories incorporated in each of the multiple computers that make up the cloud server may be considered.

(Flow of parameter identification method)
The flow of the parameter specifying method M1 will be described with reference to Fig. 3. Fig. 3 is a flowchart showing the flow of the parameter specifying method M1.

Parameter identification method M1 includes a relational equation identification step M11, a relational equation output step M12, a determination step M13, a parameter identification step M14, and a control step M15.

The relational equation identification step M11 is a step in which the processor 31 references the parameter set PS1, PS2, ..., PSk stored in the secondary memory 33 to identify a nonlinear relational equation that represents the relationship between the kneading parameters x1, x2, ..., xn, y1, y2, ..., ym, and z. In this embodiment, the processor 31 uses a genetic algorithm to identify the relational equation z = f(x1, x2, ..., xn, y1, y2, ..., ym). Specific examples of the relational equation identification step M11 will be described later with reference to different drawings.

The relational equation output step M12 is a step in which the processor 31 outputs the relational equation z = f(x1, x2, ..., xn, y1, y2, ..., ym) identified in the relational equation identification step M11. In this embodiment, the processor 31 outputs (displays) the relational equation z = f(x1, x2, ..., xn, y1, y2, ..., ym) on a display. At this time, the processor 31 may also display a graph representing the relational equation on the display.

The user visually checks the relational equation z = f(x1, x2, ..., xn, y1, y2, ..., ym) output on the display and determines whether the relational equation z = f(x1, x2, ..., xn, y1, y2, ..., ym) identified by the parameter identification device 3 is an appropriate relational equation. After completing this determination, the user then performs a user operation to input the determination result into the parameter identification device 3.

Determination step M13 is a step in which processor 31 determines whether the relational equation z = f(x1, x2, ..., xn, y1, y2, ..., ym) identified in relational equation identification step M11 is an appropriate relational equation, in response to the user operation described above. If determination step M13 determines that the relational equation is "appropriate," processor 31 executes parameter identification step M14, described below. On the other hand, if determination step M13 determines that the relational equation is "not appropriate," processor 31 re-executes the processing from relational equation identification step M11 described above onward. Note that when re-executing the processing from relational equation identification step M11 onward, processor 31 may change the set of kneading parameters used in relational equation identification step M11 or change the parameters used in the genetic algorithm.

The parameter identification step M14 is a step in which, using the relational equation z = f(x1, x2, ..., xn, y1, y2, ..., ym) identified in the relational equation identification step M11, the values to which the controllable mixing parameters x1, x2, ..., xn should be set are identified in order to make the mixed sand property z match the target value. That is, in this embodiment, the processor 31 identifies the remaining parameters by substituting some of the parameters, including the parameters representing the sand properties of the mixed sand, into the relational equation and performing calculations.

In this embodiment, processor 31 obtains an equation with controllable mixing parameters x1, x2, ..., xn as unknowns by substituting the target value of mixed sand property z and the measured values of uncontrollable mixing parameters y1, y2, ..., ym into the relational expression z = f(x1, x2, ..., xn, y1, y2, ..., ym). Processor 31 then solves this equation to obtain set values for controllable mixing parameters x1, x2, ..., xn. Note that if the number of controllable mixing parameters x1, x2, ..., xn is two or more, it is not possible to uniquely determine a solution to the above equation. In this case, processor 31 sets at least one of the solutions to the above equation as the set values for the controllable mixing parameters x1, x2, ..., xn.

Control step M15 is a step in which processor 31 controls kneading system 1 so that the values of the controllable kneading parameters become the set values identified in parameter identification step M14. That is, in this embodiment, processor 31 controls kneading system 1 using the parameters identified in parameter identification step M14.

In this embodiment, the processor 31 controls the foundry sand charging device 11 so that the amount of foundry sand charged becomes the set value specified in the parameter specification step M14. The processor 31 also controls the additive charging device 12 so that the amount of additives charged becomes the set value specified in the parameter specification step M14. The processor 31 also controls the water injection device 13 so that the amount of water charged becomes the set value specified in the parameter specification step M14. The processor 31 also controls the kneader 14 so that each of the kneading conditions becomes the set value specified in the parameter specification step M14.

In this embodiment, a genetic algorithm is used as the algorithm for identifying the nonlinear relational equation representing the relationship between the kneading parameters x1, x2, ..., xn, y1, y2, ..., ym, and z. However, the present invention is not limited to this. In other words, a nonlinear regression algorithm other than a genetic algorithm, such as logistic regression, may be used as the algorithm for identifying the nonlinear relational equation representing the relationship between the kneading parameters x1, x2, ..., xn, y1, y2, ..., ym, and z.

(Specific example of relational expression specifying step)
A specific example of the relational equation specifying step M11 included in the parameter specifying method M1 will be described with reference to Figs. 4 and 5. Fig. 4 is a diagram illustrating a genetic algorithm GA. Fig. 5 is a flowchart showing the flow of the relational equation specifying step M11 executed by the processor 31. In the example of Fig. 4, the genetic algorithm GA includes first to fourth generations G1 to G4.

In the relational equation identification step M11 in this specific example, a genetic algorithm (GA) is used to identify the relational equation z = f(x1, x2, ..., xn, y1, y2, ..., ym). Here, the genetic algorithm (GA) refers to an algorithm that prepares multiple individuals i, each of which represents a solution candidate using genes, preferentially selects individuals i with high fitness levels Di, and searches for a solution by repeating operations such as crossover and mutation. In this embodiment, the individuals i represent a nonlinear relational equation in a tree structure, with the operators and arguments included in the relational equation represented by the nodes of the tree. The fitness level Di is given by a fitness function.

Processor 31 executes the relational equation identification step M11 using a predetermined module (hereinafter referred to as "Module A"). Module A is a module that executes a genetic algorithm. In Module A, Processor 31 begins by creating a population of simple random equations that represent the relationships between known independent variables and their dependent variable targets in order to predict new data. Processor 31 then evolves the population to generate the next generation of population by selecting the fittest individuals from the population to be genetically manipulated. Through the above operations, a relational equation that best represents the above relationship is identified.

In this specific example, a module that executes genetic programming is used as module A. Genetic programming is an extension of genetic algorithms, and uses a tree structure to represent genotypes. Note that the flow of relational equation identification step M11 shown in Figure 5 is an example, and the method of identifying a relational equation using a genetic algorithm GA is not limited to the method shown in Figure 5. Various other methods can be used as a method of identifying a relational equation using a genetic algorithm GA.

In step M121, the processor 31 acquires the kneading parameter set PS. In this operation example, the processor 31 acquires the kneading parameter set PS by reading the kneading parameter set PS stored in the secondary memory 33.

In step M122, the processor 31 acquires parameters (hereinafter referred to as "individual parameters") used in the genetic algorithm GA. The individual parameters include, for example, the number of individuals to be generated N, the tournament size Nt, the crossover probability Pc, the mutation probability Pms, the number of evolutionary generations Ng, the operator Oj used in the syntax tree, the maximum depth d of the syntax tree, and the event occurrence probabilities Pk1 to Pk5. The value of each individual parameter is input to the parameter identification device 3, for example, by the user.

The number of generated individuals N represents the number of individuals i to be included in the set. The tournament size Nt is the number of individuals i to be randomly selected from the set of the current generation. The mutation probability Pms is the probability that a gene will mutate. Operators Oi used in syntax trees include, for example, Max, Min, sqrt (root), log (natural logarithm), +, -, ×, ÷, sin (radians), cos (radians), tan (radians), abs, neg, and inv. Max is an operator that selects the maximum value. Min is an operator that selects the minimum value. neg is an operator that negates the sign. inv is an operator that sets arguments close to zero to 0.

Event occurrence probabilities Pk1 to Pk5 are the probabilities that operations m1 to m5 will be selected as operations to evolve the next-generation set. Processor 31 evolves the next-generation set using one of operations m1 to m5. The sum of event occurrence probabilities Pk1 to Pk5 is 1. As an example, the values of event occurrence probabilities Pk1, Pk2, Pk3, Pk4, and Pk5 are "0.1," "0.2," "0.3," "0.4," and "0.1," respectively. Operations m1 to m5 will be described later with reference to different drawings.

In step M123, processor 31 randomly generates N individuals i based on specified individual parameters (operator Oi to be used in the syntax tree, maximum depth d of the syntax tree, etc.), and generates a set of N individuals i that will become the initial current generation.

In step M124, the processor 31 calculates the fitness Di of each individual i included in the set of the current generation. The fitness Di is given by the fitness function.

In step M125, processor 31 randomly selects individuals i equal to the tournament size Nt from the current generation set, selects the individual i with the highest fitness Di among them, and adds it to the next generation set. The individual i selected in step M125, i.e., the individual i added to the next generation set, is also referred to as the "winning tree."

Processor 31 repeats the processing of step M125 until the number of individuals in the next generation reaches N, the same as the current generation, i.e., while the number of individuals in the next generation has not reached N (step M126; NO). When the number of individuals in the next generation reaches N (step M126; YES), processor 21 executes the processing of step M127.

In step M127, processor 31 executes a process to evolve the next-generation set. Details of step M127 will be described later with reference to different drawings.

In step M130, processor 31 overwrites the next-generation set on the current-generation set. In step M131, processor 31 determines whether the number of evolutionary generations Ng has been reached. If the number of evolutionary generations Ng has not been reached (step M131; NO), processor 31 returns to the processing of step M124. On the other hand, if the number of evolutionary generations Ng has been reached (step M131; YES), processor 31 proceeds to the processing of step M132.

In step M132, processor 31 identifies the individual i with the highest fitness Di from among the individuals i included in the set of the current generation. Through the above processing, processor 31 identifies a nonlinear relational expression that represents the relationship between multiple blending parameters.

Figure 6 is a flowchart illustrating the flow of step M127 executed by processor 31. In step M201, processor 31 selects one of operations m1 to m5 based on the event occurrence probabilities Pk1 to Pk5 set by the user. If operation m1 is selected (step M201; "operation m1"), processor 31 proceeds to processing of step M202. If operation m2 is selected (step M201; "operation m2"), processor 31 proceeds to processing of step M211. If operation m3 is selected (step M201; "operation m3"), processor 31 proceeds to processing of step M221. If operation m4 is selected (step M201; "operation m4"), processor 31 proceeds to processing of step M231. If operation m5 is selected (step M201; "operation m5"), processor 31 terminates processing.

Operation m1 is crossover. Crossover is a method of mixing genetic material between individuals. In the case of crossover, in step M202, processor 31 randomly selects subtrees included in each winning tree for the next generation set.

In step M203, processor 31 generates a next-generation set for donors. The processing content of step M203 is the same as the processing content of steps M123 to M125 in Figure 6. That is, processor 31 first randomly generates N individuals i based on individual parameters specified by the user, and generates a set of N individuals i (hereinafter referred to as the "donor set"). Next, processor 31 calculates the fitness Di of each individual i included in the donor set. Next, processor 31 randomly selects individuals i equal to the tournament size Nt from the donor set, selects the individual i with the highest fitness Di among them, and adds it to the next-generation donor set. The individual i selected by this process is also referred to as the "donor tree." Processor 31 repeats the donor tree selection process until the number of next-generation donor trees reaches N.

In step M204, processor 31 randomly selects a subtree included in the donor tree (hereinafter referred to as the "donor subtree").

In step M205, processor 31 replaces the subtrees in the winning tree. In this embodiment, processor 31 removes the subtree selected in step M202 from the winning tree and transplants the donor subtree selected in step M203 into the location where the subtree was previously located. In other words, processor 31 replaces the subtree included in the winning tree with the donor subtree. The winning tree with the replaced subtree becomes the descendant (individual) of the next generation.

Operation m2 is an operation for mutating a subtree. By mutating a subtree, extinct functions and operators can be reintroduced into the population, maintaining diversity. In this case, in step M211, processor 31 randomly selects a subtree to be included in the winning tree.

In step M212, processor 31 randomly generates a subtree. In step M213, processor 31 replaces the subtrees in the winning tree. In this embodiment, processor 31 removes the subtree selected in step M202 from the winning tree and transplants the subtree generated in step M212 into the location of the removed subtree. In other words, processor 31 replaces the subtree included in the winning tree with the subtree generated in step M212. The winning tree with the replaced subtree becomes the descendant (individual) of the next generation.

Operation m3 is hoist mutation. Hoist mutation is a mutation operation that combats tree growth. In step M221, the processor randomly selects a subtree to be included in the winning tree. In step M222, processor 31 randomly selects a subtree to be included in the subtree selected in step M221.

In step M223, processor 31 rolls up the subtree selected in step M222 to the position of the original subtree (the subtree selected in step M221). The winning tree into which this subtree is rolled up becomes the descendant (individual) of the next generation.

Operation m4 is point mutation. Point mutation is an operation that reintroduces extinct relations and operators into a population in order to maintain diversity. In step M231, processor 31 randomly selects a node in the winning tree. In step M232, processor 31 replaces the node selected in step M231 with another node. As a result, the relational expression represented by the winning tree is replaced with another relational expression that requires the same number of arguments as the original node. The winning tree obtained by the replacement becomes the offspring (individual) of the next generation.

Operation m5 is regeneration, in which the winning tree is replicated and included in the next generation without modification.

Figures 7 to 10 are diagrams illustrating the operations performed on the winning tree. Figure 7 is a diagram illustrating the operation m1 (crossover). In the example of Figure 7, subtree tr111 of winning tree tr11 is replaced with subtree tr121 of donor tree tr12, resulting in winning tree tr13. Winning tree tr13 becomes the descendant (individual) of the next generation.

Figure 8 is a diagram illustrating the contents of operation m2 (subtree mutation). In the example in Figure 8, subtree tr111 of winner tree tr11 is replaced with subtree tr22, resulting in winner tree tr23. Winner tree tr23 becomes the descendant (individual) of the next generation.

Figure 9 is a diagram illustrating the contents of operation m3 (hoist mutation). In the example in Figure 9, subtree tr1121 of winner tree tr11 is hoisted up to the position of subtree tr112, becoming winner tree tr31. Winner tree tr31 becomes the descendant (individual) of the next generation.

Figure 10 is a diagram illustrating the contents of operation m4 (point mutation). In the example of Figure 10, nodes n21 and n34 included in winner tree tr11 are replaced with nodes n421 and n434, resulting in winner tree tr41. Winner tree tr41 becomes the descendant (individual) of the next generation.

(Validity of the relational expression)
To verify the validity of the relational equation identified in the relational equation identifying step M11, mixing parameters were periodically and repeatedly collected. Then, a relational equation z = f(x1, x2, ..., xn, y1, y2, ..., ym) was identified using the method described above as a specific example of the relational equation identifying step M11. The relational equation z = f(x1, x2, ..., xn, y1, y2, ..., ym) was identified using the objective variable z, the controllable explanatory variables x1, x2, ..., xn representing the amount of molding sand charged, the amount of additives charged, the mixing conditions, and the amount of water charged, and the uncontrollable explanatory variables y1, y2, ..., ym representing the external environmental variables, the properties of the mixed sand, and the properties of the molding sand. The actual measured values of compactability (CB value) and the predicted values of compactability (CB value) obtained by substituting the actual measured values of the mixing parameters other than compactability into the relational equation z = f(x1, x2, ..., xn, y1, y2, ..., ym) were plotted. The results are shown in FIG. 11 .

In Figure 11, the solid line graph represents predicted values of compactability, and the dotted plots represent actual measured values of compactability. Figure 11 shows that the predicted values of compactability closely approximate the actual measured values of compactability. This indicates that the relational expression identified in relational expression identification step M11 is valid.

Second Embodiment
(Configuration of sand property control system)
The configuration of a sand property control system S2 according to a second embodiment of the present invention will be described with reference to Fig. 12. Fig. 12 is a block diagram showing the configuration of the sand property control system S2. For ease of explanation, members having the same functions as those described in the above embodiment will be denoted by the same reference numerals, and their description will not be repeated.

The sand property control system S2 includes a mixing system 1, a parameter identification device 3B, a data logger 2, a group of external environment sensors 4, and a machine learning device 5. Of these, the mixing system 1 and the data logger 2 have the same configuration as in the first embodiment described above.

The machine learning device 5 is a device for implementing the machine learning method M2. The machine learning method M2 is a method for constructing a kneading parameter set PS, which is a learning dataset, using data provided by the data logger 2, and for constructing a learned model LM1 through machine learning (supervised learning) using the kneading parameter set PS. The learned model LM1 can be based on algorithms such as neural network models such as convolutional neural networks and recurrent neural networks, regression models such as linear regression, or tree models such as regression trees. Details of the configuration of the machine learning device 5 and the flow of the machine learning method M2 will be described later with reference to the accompanying drawings.

The trained model LM1 is a trained model that has machine-learned the correlation between input data and output data. The input of the trained model LM1 is a subset of multiple blending parameters. The input of the trained model LM1 includes, for example, controllable blending parameters x1, x2, ..., xn, and uncontrollable blending parameters y1, y2, ..., ym.

The output of the trained model LM1 includes the remaining mixing parameters from among the multiple mixing parameters. The output of the trained model LM1 includes, for example, the mixed sand property z.

[Parameter Identification Device]
The parameter identification device 3B includes a data collection device 311. The data collection device 311 temporarily stores input data to the learned model LM1 and output data from the learned model LM1, and inputs the stored input data and output data to a nonlinear regression algorithm (e.g., a genetic algorithm GA) in the next stage. The data collection device 311 is, for example, a buffer equipped with information storage means (e.g., secondary memory 33) such as a semiconductor memory or a hard disk, and input/output means. Note that, although the example in FIG. 12 illustrates a configuration in which the data collection device 311 is included in the parameter identification device 3B, the data collection device 311 may be configured as a device separate from the parameter identification device 3B. Furthermore, the data collection device 311 may be configured as a hardware unit, or may be configured as a combination of hardware and software.

When the data collection device 311 is configured as a device separate from the parameter identification device 3B, the data collection device 311 includes, for example, a memory (not shown) and a processor (not shown).

The parameter identification device 3B executes the parameter identification method MB1. The flow of the parameter identification method MB1 will be described in detail later with reference to the accompanying drawings. In addition to the parameter identification program P2 and the kneading parameter set PS, the secondary memory 33 of the parameter identification device 3B stores the learned model LM1. The processor 31 expands the learned model LM1 stored in the secondary memory 33 onto the primary memory 32. The learned model LM1 expanded onto the primary memory 32 is used when the processor 31 executes the estimation step MB11 (described below) of the parameter identification method MB1. Note that the learned model LM1 being stored in the secondary memory 33 means that the parameters defining the learned model LM1 are stored in the secondary memory 33.

[Flow of parameter identification method]
The flow of the parameter identification method MB1 will be described with reference to FIG. 13. FIG. 13 is a flowchart showing the flow of the parameter identification method MB1. The parameter identification method MB1 includes an estimation step MB11, a relational equation identification step MB12, a relational equation output step M12, a determination step M13, a parameter identification step M14, and a control step M15. Of these steps, the processes of the relational equation output step M12, the determination step M13, the parameter identification step M14, and the control step M15 are the same as the processes of the respective steps described in FIG. 4 in the first embodiment.

Estimation step MB11 is a step in which processor 31 estimates output data from input data using trained model LM1. In estimation step MB11, processor 31 acquires mixing parameters collected by data logger 2 and inputs input data including some of the acquired mixing parameters into trained model LM1 to acquire output data. The mixing parameters acquired by processor 31 are some of the mixing parameters collected by data logger 2 and are selected, for example, randomly. The input data input to trained model LM1 includes, for example, controllable mixing parameters x1, x2, ..., xn and uncontrollable mixing parameters y1, y2, ..., ym. The output data output by trained model LM2 includes, for example, mixed sand properties z.

The set of blending parameters PS1, PS2, ..., PSk, which is a combination of input data input to the learned model LM1 and output data output from the learned model LM1, is temporarily stored in the data collection device 311.

The relational equation identification step MB12 is a step in which the processor 31 identifies, by nonlinear regression using a genetic algorithm GA, a nonlinear relational equation that represents the relationship between the numerical values representing the input data input to the learned model LM1 and the numerical values representing the output data estimated in the estimation step MB11. The data collection device 311 inputs the temporarily stored input and output data of the learned model LM1 into the genetic algorithm GA in a time-synchronized manner. In this way, the data collection device 311 acts as a kind of buffer, collecting the data of the learned model LM1 (in certain information units) and inputting it to the next-stage genetic algorithm GA.

Figure 14 is a diagram illustrating a trained model LM1 and a genetic algorithm GA. The trained model LM1 is a deep neural network composed of multiple layers. The trained model LM1 includes multiple layers, including an input layer LX to which input data is input, a hidden layer LY, and an output layer LZ that outputs output data. Each layer has a structure in which multiple nodes are connected by edges.

Each layer has a function called an activation function, and edges can have weights. The output value of each node is calculated from the output value of the node connected to it in the previous layer. In other words, the output value of each node is calculated from the output value of the node in the previous layer, the weight value of the connecting edge, and the activation function of the layer.

In the example of Figure 14, the input layer LX includes nodes X1, X2, X3, and X4. That is, in this example, the input data input to the trained model LM1 includes output values x1, x2, x3, and x4 of nodes X1, X2, X3, and X4.

The hidden layer LY includes nodes Y1, Y2, and Y3. The output values y1, y2, and y3 of nodes Y1, Y2, and Y3 are calculated from the output values x1, x2, x3, and x4 of the nodes in the input layer LX, which is the layer preceding the hidden layer LY.

The output layer LZ includes node Z1. The output value z1 of node Z1 is calculated from the output values y1, y2, y3, and y4 of the nodes in the hidden layer LY, which is the layer preceding the output layer LZ. That is, in this example, the output data of the trained model LM1 is the output value z1 of node Z1. The output value z1 of node Z1 is, for example, the mixed sand property z.

The data collection device 311 temporarily stores input data to the learned model LM1 and output data from the learned model LM1, and inputs the stored data into the next stage of the genetic algorithm GA.

The method for identifying the relational equation in relational equation identification step MB12 is the same as the method for identifying the relational equation in relational equation identification step M11 described in the first embodiment. Once the relational equation has been identified, processor 31 executes the processes from relational equation output step M12 onwards. In particular, in parameter identification step M14, processor 31 identifies the remaining mixing parameters by substituting some of the mixing parameters, including the sand property parameters, into the identified relational equation and performing calculations. Furthermore, in control step M15, processor 31 controls mixing system 1 using the identified mixing parameters.

According to this embodiment, the parameter identification device 3 estimates the output of the learned model LM1 using a nonlinear relational expression identified by nonlinear regression using a genetic algorithm (GA), and identifies some or all of the controllable mixing parameters for maintaining the sand properties using the estimated output of the learned model LM1. In other words, the parameter identification device 3 can estimate the controllable mixing parameters from the input data without using the learned model LM1. Therefore, for example, even if the calculation device of the parameter identification device 3 does not support advanced calculation processing, the parameter identification device 3 can estimate the controllable mixing parameters.

In addition, in this embodiment, the data collection device 311 buffers the input data and output data of the learned model LM1 and inputs them to the genetic algorithm GA in a time-synchronized manner. Because the output from the learned model LM1 and the input to the genetic algorithm GA can be time-synchronized, it is possible to predict the time until the next-stage genetic algorithm GA outputs the final solution.

In addition, in this embodiment, the data collection device 311 acts as a kind of buffer that collects data from the learned model LM1 (in certain information units) and inputs it to the next-stage genetic algorithm GA. This reduces wasted memory space and unnecessary calculations in the generation calculation processing of the next-stage genetic algorithm GA, making it possible to improve the efficiency of the generation calculation processing of the genetic algorithm.

(Configuration of machine learning device)
The configuration of the machine learning device 5 will be described with reference to Fig. 15. Fig. 15 is a block diagram showing the configuration of the machine learning device 5.

The machine learning device 5 is implemented using a general-purpose computer and includes a processor 51, a primary memory 52, a secondary memory 53, an input/output interface 54, a communication interface 55, and a bus 56. The processor 51, the primary memory 52, the secondary memory 53, the input/output interface 54, and the communication interface 55 are interconnected via the bus 56.

The secondary memory 53 stores a machine learning program P5 and a training dataset DS. The training dataset DS is a collection of training data DS1, DS2, etc. The training data DS1, DS2, etc. are sets of blending parameters. The processor 51 expands the machine learning program P5 stored in the secondary memory 53 onto the primary memory 52. The processor 51 then executes each step of the machine learning method M2 in accordance with the instructions contained in the machine learning program P5 expanded onto the primary memory 52. The training dataset DS stored in the secondary memory 53 is constructed in a training dataset construction step M21 (described below) of the machine learning method M2 and is used in a trained model construction step M22 (described below) of the machine learning method M2. The trained model LM1 constructed in the trained model construction step M22 of the machine learning method M2 is also stored in the secondary memory 53. Note that the machine learning program P5 being stored in the secondary memory 53 refers to the source code or an executable file obtained by compiling the source code being stored in the secondary memory 53. Furthermore, the learned model LM1 being stored in the secondary memory 53 means that the parameters that define the learned model LM1 are stored in the secondary memory 53.

Devices that can be used as the processor 51 include, for example, a CPU (Central Processing Unit), GPU (Graphics Processing Unit), DSP (Digital Signal Processor), MPU (Micro Processing Unit), FPU (Floating Point Number Processing Unit), PPU (Physics Processing Unit), microcontroller, or a combination of these. The processor 51 is sometimes called a "computing device."

Devices that can be used as the primary memory 52 include, for example, semiconductor RAM (Random Access Memory). The primary memory 52 is sometimes referred to as a "main storage device." Devices that can be used as the secondary memory 53 include, for example, flash memory, HDD (Hard Disk Drive), SSD (Solid State Drive), ODD (Optical Disk Drive), FDD (Floppy Disk Drive), or a combination thereof. The secondary memory 53 is sometimes referred to as an "auxiliary storage device." The secondary memory 53 may be built into the machine learning device 5, or may be built into another computer (e.g., a computer constituting a cloud server) connected to the machine learning device 5 via the input/output interface 54 or the communication interface 55. In this embodiment, the storage in the machine learning device 5 is realized using two memories (primary memory 52 and secondary memory 53), but this is not limited to this. In other words, the storage in the machine learning device 5 may be realized using a single memory. In this case, for example, one storage area of the memory can be used as primary memory 52, and another storage area of the memory can be used as secondary memory 53.

An input device and/or an output device are connected to the input/output interface 54. Examples of the input/output interface 54 include interfaces such as USB (Universal Serial Bus), ATA (Advanced Technology Attachment), SCSI (Small Computer System Interface), and PCI (Peripheral Component Interconnect). An example of an input device connected to the input/output interface 54 is a data logger 2. Data acquired in the machine learning method M2 is input to the machine learning device 5 via the data logger 2 and stored in the primary memory 52. Examples of input devices connected to the input/output interface 54 include a keyboard, mouse, touchpad, microphone, or a combination thereof. Examples of output devices connected to the input/output interface 54 include a display, projector, printer, speakers, headphones, or a combination thereof. Information provided to the user in the machine learning method M2 is output from the machine learning device 5 via these output devices. Note that the machine learning device 5 may have a built-in keyboard that functions as an input device and a built-in display that functions as an output device, like a laptop computer. Alternatively, the machine learning device 5 may have a built-in touch panel that functions as both an input device and an output device, like a tablet computer.

Other computers are connected to the communication interface 55 via a network, either wired or wirelessly. Examples of the communication interface 55 include interfaces such as Ethernet (registered trademark) and Wi-Fi (registered trademark). Available networks include a PAN (Personal Area Network), LAN (Local Area Network), CAN (Campus Area Network), MAN (Metropolitan Area Network), WAN (Wide Area Network), GAN (Global Area Network), or an internetwork including these networks. The internetwork may be an intranet, an extranet, or the Internet. Data (e.g., the trained model LM1) provided by the machine learning device 5 to other computers (e.g., the parameter identification device 3) is sent and received via these networks.

Note that while this embodiment employs a configuration in which machine learning method M2 is executed using a single processor (processor 51), the present invention is not limited to this. That is, a configuration in which machine learning method M2 is executed using multiple processors may also be employed. In this case, the multiple processors that cooperate to execute machine learning method M2 may be provided on a single computer and configured to be able to communicate with each other via a bus, or may be distributed across multiple computers and configured to be able to communicate with each other via a network. As an example, a processor built into a computer that constitutes a cloud server and a processor built into a computer owned by a user of that cloud server may cooperate to execute machine learning method M2.

Furthermore, while this embodiment employs a configuration in which the training dataset DS is stored in memory (secondary memory 53) built into the same computer as the processor (processor 51) that executes machine learning method M2, the present invention is not limited to this. That is, a configuration in which the training dataset DS is stored in memory built into a computer different from the processor that executes machine learning method M2 may also be employed. In this case, the computer incorporating the memory that stores the training dataset DS is configured to be able to communicate with a computer incorporating the processor that executes machine learning method M2 via a network. As an example, the training dataset DS may be stored in memory built into a computer that constitutes a cloud server, and the machine learning method M2 may be executed by a processor built into a computer owned by a user of the cloud server.

Furthermore, while this embodiment employs a configuration in which the training dataset DS is stored in a single memory (secondary memory 53), the present invention is not limited to this. That is, a configuration in which the training dataset DS is distributed and stored in multiple memories may also be employed. In this case, the multiple memories that store the training dataset DS may be provided in a single computer (which may or may not be a computer with a built-in processor that executes machine learning method M2), or may be distributed and stored in multiple computers (which may or may not include a computer with a built-in processor that executes machine learning method M2). As an example, a configuration in which the training dataset DS is distributed and stored in memory built into each of multiple computers that make up a cloud server is conceivable.

Furthermore, in this embodiment, a configuration is adopted in which parameter identification method M1 and machine learning method M2 are executed using different processors (processor 31 and processor 51), but the present invention is not limited to this. That is, parameter identification method M1 and machine learning method M2 may be executed using the same processor. In this case, by executing machine learning method M2, the learned model LM1 is stored in memory built into the same computer as this processor. Then, when executing parameter identification method M1, this processor uses the learned model LM1 stored in this memory to identify relational expressions and parameters.

[Machine learning method flow]
The flow of the machine learning method M2 will be described with reference to Fig. 16. Fig. 16 is a flowchart showing the flow of the machine learning method M2.

The machine learning method M2 includes a learning dataset construction step M21 and a trained model construction step M22.

The learning dataset construction step M21 is a step in which the processor 51 constructs a learning dataset DS, which is a collection of training data DS1, DS2, .... Each training data DSi (i = 1, 2, ...) includes multiple blending parameters. In the learning dataset construction step M21, the processor 51 acquires the blending parameters in the same manner as the parameter identification device 3 and stores them in the secondary memory 33. The processor 31 repeats the above process to construct the learning dataset DS.

The trained model construction step M22 is a step in which the processor 51 constructs a trained model LM1. In the trained model construction step M22, the processor 51 constructs the trained model LM1 through supervised learning using the training dataset DS. The processor 31 then stores the constructed trained model LM1 in the secondary memory 53.

〔summary〕
The parameter identification device of aspect 1 includes one or more processors that execute a parameter identification step of identifying the remaining parameters from among a plurality of parameters related to a mixing system that mixes foundry sand, water, and additives to produce mixed sand, including parameters that represent the sand properties of the mixed sand.

With the above configuration, the parameter identification device can identify parameters for maintaining the sand properties of mixed sand.

The parameter identification device according to Aspect 2 has the following features in addition to the features of the parameter identification device according to Aspect 1. That is, in the parameter identification device according to Aspect 2, the processor further executes a relational equation identification step in which, by referencing the set of parameters, a nonlinear relational equation expressing the relationship between the parameters is identified using a nonlinear regression algorithm, and in the parameter identification step, the remaining parameters are identified by substituting some of the parameters, including parameters expressing the sand properties of the mixed sand, into the relational equation and performing a calculation.

With the above configuration, the parameter identification device can identify parameters for maintaining the sand properties of mixed sand.

The parameter specifying device according to Aspect 3 has the following features in addition to the features of the parameter specifying device according to Aspect 1 or 2. That is, in the parameter specifying device according to Aspect 3, the processor executes a control step of controlling the kneading system using the parameters specified in the parameter specifying step.

With the above configuration, the parameter identification device can maintain the sand properties of the mixed sand by controlling the mixing system using the identified parameters.

The parameter specifying device according to Aspect 4 has the following characteristics in addition to the characteristics of the parameter specifying device according to any one of Aspects 1 to 3. That is, in the parameter specifying device according to Aspect 4, the plurality of parameters include, in addition to a parameter representing the sand properties of the mixed sand, parameters representing at least one of the sand properties of the foundry sand, the amount of foundry sand charged, the amount of additives charged, the amount of water charged, the mixing conditions, and the external environment of the mixing system.

With the above configuration, the parameter identification device can maintain the sand properties of the mixed sand by controlling the mixing system using the identified parameters.

The parameter specifying device according to Aspect 5 has the following features in addition to the features of the parameter specifying device according to Aspect 2. That is, in the parameter specifying device according to Aspect 5, in the relational expression specifying step, the processor specifies the nonlinear relational expression by nonlinear regression using a genetic algorithm.

With the above configuration, the parameter identification device can identify parameters for maintaining the sand properties of mixed sand.

A parameter identification device according to Aspect 6 has the following features in addition to the features of the parameter identification device according to Aspect 2 or 5. That is, in the parameter identification device according to Aspect 6, the processor further executes an estimation step of estimating the output data from the input data using a trained model that has learned by machine learning the correlation between input data that includes some of the plurality of parameters and output data that includes the remaining parameters of the plurality of parameters, and in the relational expression identification step, the processor identifies the nonlinear relational expression that represents the relationship between numerical values that represent the input data input to the trained model and numerical values that represent the output data estimated in the estimation step.

With the above configuration, the parameter identification device can identify parameters for maintaining the sand properties of mixed sand without using a trained model.

The parameter identification method according to aspect 7 includes a parameter identification step in which one or more processors identify, from among a plurality of parameters related to a mixing system that mixes foundry sand, water, and additives to produce mixed sand, the remaining parameters.

The above configuration makes it possible to identify parameters for maintaining the sand properties of mixed sand.

[Additional Note 1]
In each of the above-described embodiments, in the parameter identification step M14, the processor 31 obtains an equation with the controllable mixing parameters x1, x2, ..., xn as unknowns by substituting the target value of the mixed sand property z and the measured values of the uncontrollable mixing parameters y1, y2, ..., ym into the relational equation z=f(x1, x2, ..., xn, y1, y2, ..., ym). The processor 31 then solves this equation to obtain the set values of the controllable mixing parameters x1, x2, ..., xn. The mixing parameters substituted into the relational equation are not limited to those described in the above-described embodiments. For example, in addition to the target value of the mixed sand property z and the measured values of the uncontrollable mixing parameters y1, y2, ..., ym, some set values of the controllable mixing parameters x1, x2, ..., xn may be substituted into the relational equation. In this case, the processor 31 obtains the remaining set values of the controllable kneading parameters x1, x2, . . . , xn by solving an equation obtained by substituting each value into the relational expression.

[Additional Note 2]
In the second embodiment described above, a trained model trained by supervised learning is used as the trained model LM1, but a trained model trained by unsupervised learning may also be used. For a trained model trained by unsupervised learning, the processor 31 also identifies a nonlinear relational expression that represents the relationship between the input and output of the trained model by nonlinear regression using a genetic algorithm.

[Appendix 3]
In the second embodiment described above, the machine learning device 5 constructs the trained model LM1, but the trained model LM1 may be constructed in advance by a device other than the machine learning device 5. In this case, the parameter identifying device 3 performs the parameter identifying method MB1 described above using the trained model constructed in advance by the other device.

[Appendix 4]
In the second embodiment described above, the processor 31 identifies a nonlinear relational equation representing the relationship between input data and output data of a trained model by nonlinear regression using a genetic algorithm, with the nonlinear relational equation being a candidate solution. However, the method for identifying the relational equation representing the relationship between input data and output data is not limited to the method described in the second embodiment. For example, the processor 31 may identify the first relational equation and the second relational equation by nonlinear regression using a genetic algorithm, with the candidate solutions being a first nonlinear relational equation representing the relationship between numerical values representing input data and output values of some or all nodes belonging to a hidden layer, and a second nonlinear relational equation representing the relationship between the output values of the nodes and numerical values representing output data. In this case, the processor 31 identifies a nonlinear relational equation representing the relationship between numerical values representing input data and output data, for example, by solving a simultaneous equation of the identified first relational equation and second relational equation.

Furthermore, the algorithm for identifying the nonlinear relational equation that represents the relationship between input data and output data of the trained model is not limited to a genetic algorithm, and may be another nonlinear algorithm. As one example, processor 11 may identify the nonlinear relational equation that represents the relationship between input data and output data using the Monte Carlo method. In this case, as one example, processor 11 creates multiple relational equations by randomly selecting the length of the relational equation and elements of the relational equation (variables, operators, etc.), and constructs a relational equation that derives output data from input data by selecting the relational equation with the smallest error from the multiple relational equations created.

Furthermore, in the second embodiment described above, the processor 11 identifies a nonlinear relational equation representing the relationship between the input and output of the learned model LM1 by nonlinear regression, and identifies the remaining multiple parameters by substituting some of the parameters into the identified relational equation and performing a calculation. The method of identifying the remaining multiple parameters is not limited to that described in the above embodiment. As an example, the processor 11 may identify the remaining multiple parameters by inputting some of the multiple parameters into the learned model LM1.

[Appendix 5]
The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Other embodiments obtained by appropriately combining the technical means disclosed in the above-described embodiments are also included in the technical scope of the present invention.

1 Mixing system 2 Data logger 3, 3B Parameter identification device 5 Machine learning device 10 Molding sand property measurement device 11 Molding sand injection device 12 Additive injection device 13 Water injection device 14 Mixer 15 Mixed sand property measurement device 16 Sensor group 31, 51 Processor 32, 52 Primary memory 33, 53 Secondary memory 34, 54 Input/output interface 35, 55 Communication interface 36, 56 Bus M1, MB1 Parameter identification method M11, MB12 Relational equation identification step M12 Relational equation output step M13 Judgment step M14 Parameter identification step M15 Control step M2 Machine learning method M21 Learning dataset construction step M22 Learned model construction step MB11 Estimation step S1 Sand property control system

Claims (6)

a parameter specifying step of specifying, from among a plurality of parameters related to a mixing system that mixes foundry sand, water, and additives to produce mixed sand, the plurality of parameters including parameters that represent sand properties of the mixed sand, the remaining parameters;
The remaining parameters include parameters representing mixing conditions in the mixing system, including at least one of the rotation speed and rotation time of a mixing blade of a mixer that mixes the mixed sand,
the plurality of parameters include, in addition to a parameter representing the sand properties of the mixed sand, a parameter representing at least one of the amount of the foundry sand to be mixed by the mixing system, the amount of the additives to be added, the amount of water to be added, the mixing conditions, and the external environment of the mixing system;
the processor controls the mixer so as to mix under the mixing conditions represented by the parameters identified in the parameter identification step.
A parameter specifying device characterized by: The processor:
a relational expression specifying step of specifying a nonlinear relational expression representing a relationship between the plurality of parameters by using a nonlinear regression algorithm with reference to the set of the plurality of parameters;
In the parameter specifying step, a part of the plurality of parameters including a parameter representing the sand properties of the mixed sand is substituted into the relational expression for calculation, thereby specifying the remaining part of the plurality of parameters.
2. The parameter specifying device according to claim 1. 3. The parameter specifying device according to claim 1, wherein the processor executes a control step of controlling the kneading system using the parameters specified in the parameter specifying step. In the relational expression specifying step, the processor specifies the nonlinear relational expression by nonlinear regression using a genetic algorithm.
3. The parameter specifying device according to claim 2. the processor further executes an estimation step of estimating the output data from the input data using a trained model that has undergone machine learning to determine a correlation between input data including a portion of the plurality of parameters and output data including the remaining parameters of the plurality of parameters;
In the relational expression identification step, the processor identifies the nonlinear relational expression that represents a relationship between a numerical value representing input data input to the trained model and a numerical value representing output data estimated in the estimation step.
5. The parameter specifying device according to claim 2 or 4. a parameter identification step in which one or more processors identify, from among a plurality of parameters related to a mixing system that mixes foundry sand, water, and additives to produce mixed sand, the plurality of parameters including parameters that represent sand properties of the mixed sand, the remaining parameters;
Including,
The remaining parameters include parameters representing mixing conditions in the mixing system, including at least one of the rotation speed and rotation time of a mixing blade of a mixer that mixes the mixed sand,
the plurality of parameters include, in addition to a parameter representing the sand properties of the mixed sand, a parameter representing at least one of the amount of the foundry sand to be mixed by the mixing system, the amount of the additives to be added, the amount of water to be added, the mixing conditions, and the external environment of the mixing system;
the processor controls the mixer so as to mix under the mixing conditions represented by the parameters identified in the parameter identification step.
A parameter specifying method comprising: