Data processing device, data processing method, and data processing program

Abstract

This technology provides a way to identify the relationship between the operating conditions of manufacturing equipment and the target variable without requiring a large amount of data. [Solution] The data processing device comprises: an acquisition unit that acquires a dataset; a structuring unit that uses the dataset to generate structured data representing the causal structure between a condition variable and a target variable by connecting leaf nodes representing condition variables, root nodes representing target variables, and intermediate nodes located between the leaf nodes and the root nodes, which represent influencing variables as explanatory variables that are elements that affect the target variable but are different from the operating conditions, with directed edges; and a modeling unit that uses the dataset to model a manufacturing device by sequentially calculating a function from the leaf nodes to the root node that expresses the variables of the higher-level nodes that are the endpoints of specific edges using the variables of the lower-level nodes that are the starting points of specific edges, and generating a model function that includes the calculated function.

Description

Technical Field

[0001] The present disclosure relates to the technology of a data processing device, a data processing method, and a data processing program.

Background Art

[0002] Conventionally, when the quality of a product manufactured by operating a molding machine changes due to external factors or slight differences in the contained components of the molding material, a device that can modify the operating conditions of the molding machine so that the quality of the product approaches a reference value is known (Patent Document 1). This device estimates the quality of a product using a learned model that outputs the quality of the product when input with a feature amount extracted from detection data when manufacturing the product using the molding machine. Then, the conventional device acquires a quality transition by accumulating the estimated quality of the product. Then, the conventional device evaluates the quality change tendency with respect to the reference value of the quality using the acquired quality transition. Then, the conventional device generates a matrix representing the relationship between the quality change tendency and the operating conditions of the molding machine for returning the quality to the reference value for each type of molten state of the material. Then, the conventional device determines the content of the modification of the operating conditions of the molding machine using the matrix corresponding to the molten state of the material.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Various factors, such as individual differences in the molding machine itself, new materials, molds used for molding, and the installation environment of the molding machine, can affect the target variable related to the manufactured product, such as the quality of the manufactured product. By taking into account the influence of these various factors on the target variable, the relationship between the operating conditions of the molding machine and the target variable can be identified more accurately. When using conventional technology, it is necessary to prepare a matrix that represents the relationship between the operating conditions of the molding machine and the target variable for each of the various factors that affect the target variable, and then identify the relationship between the operating conditions of the molding machine and the target variable. However, preparing a matrix for each of the various factors that affect the target variable requires a large amount of data. This challenge is common not only to molding machines but also to other manufacturing equipment. [Means for solving the problem]

[0005] (1) According to the first embodiment of the present disclosure, a data processing device is provided. The data processing device includes: an acquisition unit that acquires a dataset containing two or more acquired data sets, which are acquired data obtained by operating a manufacturing device under predetermined operating conditions to manufacture a product, and which associate an objective variable relating to the product with an explanatory variable of the objective variable, which is a condition variable that is an explanatory variable representing the operating conditions; a structuring unit that uses the dataset to generate structured data representing the causal structure between the condition variable and the objective variable by connecting a leaf node representing the condition variable, a root node representing the objective variable, and an intermediate node located between the leaf node and the root node, which is an intermediate node that is an element that affects the objective variable and represents an element different from the operating conditions, with directed edges; and a modeling unit that uses the dataset to model the manufacturing device by sequentially calculating an explanatory function, which is a function that expresses the variable of the upper node that is the endpoint of a specific edge using the variable of the lower node that is the starting point of the specific edge, from the leaf node to the root node, and generating a model function that is a function that includes the calculated explanatory function. (2) According to a second embodiment of the present disclosure, a data processing method is provided. The data processing method comprises: an acquisition step of acquiring a dataset which includes two or more acquired data sets that associate an objective variable relating to the product with an explanatory variable of the objective variable, which is an explanatory variable representing the operating conditions; a structuring step of using the dataset to generate structured data that represents the causal structure between the condition variable and the objective variable by connecting a leaf node representing the condition variable, a root node representing the objective variable, and an intermediate node located between the leaf node and the root node, which is an explanatory variable representing an element that affects the objective variable and is different from the operating conditions, with directed edges; and a modeling step of using the dataset to model the manufacturing apparatus by sequentially calculating an explanatory function, which is a function that represents the variable of the upper node that is the endpoint of a specific edge using the variable of the lower node that is the starting point of the specific edge, from the leaf node to the root node, and generating a model function which is a function that includes the calculated explanatory function. (3) According to the third form of this disclosure, a data processing program is provided. The data processing program causes a computer to execute the following functions: an acquisition function that acquires a dataset containing two or more acquired data sets, each dataset relating an objective variable to the manufactured product and an explanatory variable to the objective variable, which is a condition variable representing the operating conditions; a structuring function that uses the dataset to generate structured data representing the causal structure between the condition variable and the objective variable by connecting a leaf node representing the condition variable, a root node representing the objective variable, and an intermediate node located between the leaf node and the root node, which represents an influence variable representing an element that affects the objective variable and is different from the operating conditions, with directed edges; and a modeling function that uses the dataset to model the manufacturing equipment by sequentially calculating explanatory functions, which are functions that represent the variables of the higher-level node that is the endpoint of a specific edge using the variables of the lower-level node that is the starting point of the specific edge, from the leaf node to the root node, and generating a model function that includes the calculated explanatory functions. [Brief explanation of the drawing]

[0006] [Figure 1] A block diagram showing the general configuration of the manufacturing system. [Figure 2] A flowchart illustrating the manufacturing method of a product using an injection molding machine. [Figure 3] A block diagram showing the configuration of the data processing device in the first embodiment. [Figure 4] A flowchart illustrating a method for modeling injection molding machines. [Figure 5] Conceptual diagram of structured data. [Figure 6] Figure 1 illustrates the method for calculating the explanatory function. [Figure 7] Figure 2 illustrates the method for calculating the explanatory function. [Figure 8]Figure 3 illustrates the method for calculating the explanatory function. [Figure 9] A diagram illustrating a concrete example of modeling an injection molding machine using a model function. [Figure 10] A block diagram showing the configuration of the data processing device in the second embodiment. [Figure 11] A flowchart illustrating the control method for an injection molding machine. [Figure 12] A block diagram showing the configuration of the data processing device in the third embodiment. [Figure 13] Conceptual diagram of structured data generated in the third embodiment. [Figure 14] A block diagram showing the configuration of the data processing device in the fourth embodiment. [Figure 15] Conceptual diagram of structured data generated in the fourth embodiment. [Figure 16] A block diagram showing the configuration of the data processing device in the fifth embodiment. [Modes for carrying out the invention]

[0007] A. First Embodiment: Figure 1 is a block diagram showing the schematic configuration of manufacturing system 1. Manufacturing system 1 is a system for determining the operating conditions of manufacturing equipment 10 and operating the manufacturing equipment 10 according to the determined operating conditions to manufacture products. Manufacturing system 1 comprises manufacturing equipment 10 and data processing equipment 50.

[0008] The manufacturing apparatus 10 manufactures products. In this embodiment, the manufacturing apparatus 10 is an injection molding machine 100. The injection molding machine 100 comprises a bed 110, an injection device 120, a mold 130, a clamping device 140, sensors 151 to 153, and a control device 160.

[0009] The injection device 120 is arranged on the bed 110. The injection device 120 includes a hopper 121, a heating cylinder 122, a screw 123, a nozzle 124, a heater 125, and an injection driving device 126. The hopper 121 is a material tank having an inlet for charging a molding material M1 such as resin. The heating cylinder 122 is movable along the axial direction D with respect to the bed 110. The heating cylinder 122 pressurizes the molten material M2 generated by heating and melting the molding material M1 charged into the hopper 121. The heating cylinder 122 is also referred to as a barrel. The screw 123 is arranged inside the heating cylinder 122. The screw 123 is movable along the axial direction D while rotating around the axis AX of the heating cylinder 122. The screw 123 supplies the molten material M2 inside the heating cylinder 122 to the nozzle 124 by moving along the axial direction D of the heating cylinder 122. The nozzle 124 is provided at the tip of the heating cylinder 122. The nozzle 124 is an injection port for injecting the molten material M2 inside the heating cylinder 122 toward the mold 130. The heater 125 generates the molten material M2 by heating the molding material M1 inside the heating cylinder 122. The injection driving device 126 moves the heating cylinder 122 along the axial direction D, rotates the screw 123 around the axis AX, or moves the screw 123 along the axial direction D. The injection driving device 126 is, for example, a cylinder device.

[0010] The mold 130 includes a fixed mold 131, a movable mold 132, a cavity 133, and a supply path 134. The cavity 133 is formed between the fixed mold 131 and the movable mold 132 when the fixed mold 131 and the movable mold 132 are clamped. The supply path 134 is provided across the inside of the fixed mold 131 and the inside of a fixed platen 141 described later so that the cavity 133 and the nozzle 124 communicate with each other. The supply path 134 is a flow path for supplying the molten material M2 injected from the nozzle 124 to the cavity 133.

[0011] The clamping device 140 is provided at a position facing the injection device 120 on the bed 110. The clamping device 140 performs the opening and closing operations of the mounted mold 130. Further, the clamping device 140 clamps the mold 130 so that the mold 130 does not open due to the pressure of the molten material M2 injected into the cavity 133. The clamping device 140 includes a fixed platen 141, a movable platen 142, a tie bar 143, and a mold driving device 144. The fixed platen 141 is fixed to the fixed mold 131. The movable platen 142 is fixed to the movable mold 132. The movable platen 142 can approach and separate from the fixed platen 141. The tie bar 143 supports the movable platen 142 movably. The mold driving device 144 moves the movable platen 142. The mold driving device 144 is, for example, a cylinder device.

[0012] Sensors 151 to 153 acquire various physical quantities at a predetermined time cycle during the operation of the injection molding machine 100. In the present embodiment, the injection molding machine 100 includes an injection device sensor 151, a mold sensor 152, and a clamping device sensor 153 as sensors 151 to 153. The injection device sensor 151 acquires physical quantities related to the injection device 120. The injection device sensor 151 acquires, for example, the injection speed of the molten material M2, the position of the heating cylinder 122 at the end of injection, the holding pressure, the holding time, and the state of the injection driving device 126. The mold sensor 152 acquires physical quantities related to the mold 130. The mold sensor 152 acquires, for example, the filling speed, the filling time, and the filling temperature of the molten material M2 in the supply path 134. The clamping device sensor 153 acquires physical quantities related to the clamping device 140. The clamping device sensor 153 acquires, for example, the clamping force of the mold 130, the temperature of the mold 130, and the state of the mold driving device 144. Note that the types of the sensors 151 to 153 are not limited to the above.

[0013] The control device 160 controls the operation of each part of the injection molding machine 100. The control device 160 is equipped with a communication device 161 for communicating with various devices such as the data processing device 50. The control device 160 obtains operation control signals from the data processing device 50 via the communication device 161. The operation control signals are control signals that represent the operating conditions of the injection molding machine 100. The operating conditions of the injection molding machine 100 are conditions that can be set directly when operating the injection molding machine 100. The operation control signals include parameters that represent the operating conditions of the injection molding machine 100, such as the temperature of the heating cylinder 122, the injection pressure of the molten material M2, the injection speed of the molten material M2, the holding pressure, the holding pressure time, and the rotational speed of the screw 123. The operation control signal may include parameters representing the operating conditions of the injection molding machine 100, such as the temperature of the nozzle 124 during injection, the position of the screw 123 when switching from injection to holding pressure, the clamping force of the mold 130, the temperature of the mold 130, and the cooling time of the mold 130. The control device 160 uses the operation control signal to control the operation of the injection drive unit 126 and the mold drive unit 144, etc. In this way, the control device 160 operates the injection molding machine 100 according to the operating conditions expressed in the operation control signal.

[0014] Figure 2 is a flowchart illustrating the manufacturing method of a product using an injection molding machine 100. In the preparation step S1, the molding material M1 is melted by heating with the heater 125 and shear heat generated by the rotation of the screw 123, thereby generating molten material M2. In the clamping step S2, the movable platen 142 is moved to align the fixed mold 131 and the movable mold 132, thereby clamping the mold 130. Furthermore, the heating cylinder 122 is moved along the axial direction D to approach the clamping device 140. This connects the nozzle 124 to the fixed platen 141 of the clamping device 140. In the injection step S3, with the rotation of the screw 123 stopped, the screw 123 is moved toward the nozzle 124 with a predetermined pushing force. This causes the molten material M2 to be injected into the cavity 133 of the mold 130 at high pressure. In the holding pressure step of step S4, a predetermined pressing force is applied to the screw 123 for a predetermined holding time, thereby holding the molten material M2 in the cavity 133 with a predetermined holding pressure for a predetermined holding time. This causes the molten material M2 to be pressed into the cavity 133. In the cooling step of step S5, the mold 130 is cooled with the pressing of the molten material M2 stopped, causing the molten material M2 in the cavity 133 to solidify. In the demolding step of step S6, the fixed mold 131 and the movable mold 132 are separated, and the manufactured product is removed.

[0015] Figure 3 is a block diagram showing the configuration of the data processing device 50 in the first embodiment. The data processing device 50 models an injection molding machine 100. The data processing device 50 is composed of a computer comprising a CPU 501, a storage device 502, a communication device 505, an input device 506, and a display device 507. The CPU 501 functions as an acquisition unit 511, a structuring unit 512, and a modeling unit 513 by executing a program PG stored in the storage device 502. Each function of the CPU 501 will be explained later in conjunction with the data processing method. The communication device 505 can communicate with external devices by wired communication or wireless communication. The input device 506 receives input from the user. The input device 506 has, for example, a keyboard or a mouse. The display device 507 displays various information. The display device 507 has, for example, a liquid crystal display.

[0016] Figure 4 is a flowchart illustrating a method for modeling the injection molding machine 100 as a data processing method. The injection molding machine 100 modeling method shown in Figure 4 is performed, for example, when modeling the injection molding machine 100.

[0017] In the modeling method, first, the acquisition process in step S100 is performed. The acquisition process is a process of collecting various data used when modeling the injection molding machine 100 and obtaining a dataset containing the collected data. The dataset contains at least two or more acquired data sets. The acquired data is obtained by operating the injection molding machine 100 under predetermined operating conditions to manufacture a product. The acquired data is data that associates a target variable related to the product with a condition variable, which is an explanatory variable representing the operating conditions of the injection molding machine 100 when manufacturing the product corresponding to the target variable. The target variable is, for example, quality information representing the quality of the product. The quality information is, for example, the weight of the product. The quality information may also be the dimensions of the product. The quality information is obtained, for example, by inspecting the product after it has been manufactured. The target variable may also be the cycle time when manufacturing the product, or it may be the variation in the quality of the product. The condition variable is, for example, the setting value of a parameter representing the operating conditions of the injection molding machine 100.

[0018] Here, there may be factors other than the operating conditions of the injection molding machine 100 that affect the objective variable. Factors other than operating conditions that affect the objective variable include, for example, the result conditions of the injection molding machine 100. The result conditions of the injection molding machine 100 are conditions that cannot be directly set when operating the injection molding machine 100. The result conditions of the injection molding machine 100 are determined according to the operating conditions of the injection molding machine 100, the external environment when operating the injection molding machine 100, the degree of deterioration of the movable parts of the injection molding machine 100, the material properties of the molding material M1, etc. Parameters that represent the result conditions of the injection molding machine 100 include, for example, the filling time of the molten material M2, the filling pressure of the molten material M2, the gate seal position, and the position of the heating cylinder 122 at the end of injection. Parameters that represent the result conditions of the injection molding machine 100 can be represented, for example, by sensing data output from sensors 151 to 153. Therefore, in order to take into account factors other than operating conditions that affect the objective variable when modeling the injection molding machine 100, in this embodiment, the dataset further includes sensing data output from sensors 151 to 153. The sensing data is, for example, numerical data output from sensors 151 to 153 at specific timings during the period in which the injection molding machine 100 is in operation. The sensing data may be time-series data generated by arranging two or more numerical data acquired at different timings in chronological order, or it may be feature quantities such as the mean, maximum, and minimum values ​​of the time-series data.

[0019] In the acquisition process, for example, the user determines factors and levels and creates an experimental design using an orthogonal array. Factors are parameters representing the operating conditions of the injection molding machine 100. Levels are the set values ​​of the parameters representing the operating conditions of the injection molding machine 100. The user sets the operating conditions based on the experimental design and acquires the manufactured product by operating the injection molding machine 100. The user acquires quality information such as the weight and dimensions of the manufactured product by inspecting it. The user acquires sensing data via a display device 507 or the like. The user performs this series of operations a predetermined number of times for each level. Then, the user associates the target variable such as quality information, the set values ​​of the parameters representing the operating conditions, and the sensing data for each manufactured product and inputs them into the data processing device 50 via the input device 506. As a result, the acquisition unit 511 acquires a dataset in which sensing data is associated with each of two or more acquired data sets that associate the target variable and the condition variable. Note that at least some of the processing in the acquisition process may be performed by the acquisition unit 511. Furthermore, the methods for obtaining the dataset are not limited to those described above.

[0020] Following the acquisition process, the structuring process in step S200 is executed. The structuring process is a process of generating structured data that represents the causal structure between the condition variable and the target variable using the dataset. Figure 5 is a conceptual diagram of structured data SD. Structured data SD is, for example, a directed acyclic graph (DAG). Structured data SD is generated by connecting the leaf node LN, the root node RN, and the intermediate node MN with directed edges E. The leaf node LN is a node N that has no input from other nodes N. The leaf node LN represents the condition variable CV. The root node RN is a node N that has no output to other nodes N. The root node RN represents the target variable OV. The intermediate node MN is a node N located between the leaf node LN and the root node RN. In other words, the intermediate node MN is a node N that has both input from other nodes N and output to other nodes N. The intermediate node MN represents the influencing variable IV. Influence variable IV is an explanatory variable that represents an element that influences the objective variable OV, and is different from the operating conditions. Influence variable IV is, for example, an measured value of a parameter that represents the result conditions of the injection molding machine 100. Structured data SD may also include other information such as mutual information.

[0021] In the structuring process, the structuring unit 512 generates structured data SD. The structuring unit 512 generates structured data SD by inputting the dataset into a causal estimation model MD1 pre-stored in the storage device 502, for example. The causal estimation model MD1 is a machine learning model that explores the causal structure between variables V included in the dataset. When a dataset is input, the causal estimation model MD1 outputs structured data SD. The causal estimation model MD1 sets up a node N for each variable V included in the dataset. The causal estimation model MD1 connects all nodes N with directed edges E that move from lower to higher levels in the hierarchy, according to the causal relationships between the nodes N. The causal estimation model MD1 calculates the correlation coefficient between the nodes N connected by each edge E. By deleting edges E whose correlation coefficient is below a predetermined threshold, the causal estimation model MD1 generates structured data SD that has edges E whose correlation coefficient is above a predetermined threshold. Note that at least some of the processing in the structuring process may be performed by the user. For example, the user may pre-specify, via the input device 506, which variable V to place in which node N at which hierarchical level. In this case, the structuring unit 512 generates structured data SD in which the variable V specified by the user is placed in node N at the hierarchical level specified by the user.

[0022] As shown in Figure 4, the evaluation step S300 is performed after the structuring step. The evaluation step is a process in which the structured data SD is evaluated by experts. Experts are people who have knowledge and experience regarding manufacturing using the injection molding machine 100. Experts have insights such as the tendency for the weight of the manufactured product to decrease when the setting value of a parameter representing a specific operating condition is increased. Experts are, for example, workers who operate the injection molding machine 100 and are routinely engaged in manufacturing using the injection molding machine 100. Experts may also be technicians who maintain the injection molding machine 100. Experts compare their own views with the structured data SD and extract any deficiencies or shortcomings in the structured data SD.

[0023] Following the evaluation process, the reflection process in step S350 is executed. The reflection process is a process in which measures are taken to reflect the evaluation results by experts in the structured data SD. If no deficiencies in the structured data SD are identified by the experts, that is, if step S351 is "No", then step S352 is executed. On the other hand, if deficiencies in the structured data SD are identified by the experts, that is, if step S351 is "Yes", then the process returns to the acquisition process, and the acquisition process, structuring process, and evaluation process are executed again. In this case, in the acquisition process, the user reviews the factors and levels, creates a new experimental design using an orthogonal array, determines the operating conditions based on the new experimental design, and operates the injection molding machine 100 to acquire a new product. The user then associates the sensing data with the acquired data related to the new product and inputs it into the data processing device 50 via the input device 506. As a result, the acquisition unit 511 acquires a new dataset. In the structuring process, the structuring unit 512 generates a new structured data SD using the new dataset. In the evaluation process, experts evaluate the new structured data SD. The acquisition process, the structuring process, and the evaluation process are repeatedly executed until the experts can no longer identify any deficiencies in the structured data SD, that is, until step S351 becomes "No".

[0024] If an expert identifies a deficiency in the structured data SD, i.e., if step S352 is "Yes", then step S353 is executed. If additional data is needed to correct the deficiency in the structured data SD, i.e., if step S353 is "Yes", then the process returns to the acquisition step, and the acquisition step, structuring step, and evaluation step are executed again. In this case, the user may take measures to resolve the deficiency in the structured data SD when generating the structured data SD next time using a new dataset containing the additional data. On the other hand, if no additional data is needed to correct the deficiency in the structured data SD, i.e., if step S353 is "No", then step S354 is executed. In step S354, the user takes measures to correct the structured data SD. As a result, the structuring unit 512 corrects the structured data SD according to the user's instructions, thereby correcting the deficiency in the structured data SD. The following are examples of measures that the user may take when an expert identifies a deficiency in the structured data SD.

[0025] For example, if the cause and effect are reversed in the structured data SD when considering the temporal changes in the internal state of the injection molding machine 100, the user takes the following action. In this case, the user takes action to swap node N so that the structuring unit 512 generates structured data SD in which node N representing the cause is placed lower than node N representing the effect. For example, if causal structure exploration reveals a causal relationship that the injection pressure increased because the injection stroke distance increased, then in the structured data SD, node N representing the injection stroke distance is placed lower than node N representing the injection pressure. However, in the behavior of the injection molding machine 100, the injection stroke distance increases with increasing injection pressure. In other words, when considering the temporal changes in the internal state of the injection molding machine 100, the increase in injection pressure occurs temporally before the determination of the injection stroke distance, and the cause and effect are reversed in the structured data SD. Therefore, the user takes measures to rearrange node N so that the structuring unit 512 generates structured data SD in which node N representing injection pressure is placed lower than node N representing injection stroke distance.

[0026] In this case, if no additional data is required to replace node N in the structured data SD, the user retrospectively specifies, for example, the content of the variable V represented by node N to be modified and the hierarchy of node N to be modified, via input device 506. This allows the structuring unit 512 to modify the structured data SD by replacing node N so that the node N specified by the user represents the variable V specified by the user, thereby correcting the causal structure of the structured data SD. On the other hand, if additional data is required to replace node N in the structured data SD, the user specifies, for example, the following information via input device 506 as prior information for when the structured data SD is generated next time using a new dataset. In this case, the user pre-specifies the content of the variable V represented by node N to be modified and the hierarchy of node N to be modified. This allows the structuring unit 512 to generate the structured data SD with node N replaced using the new dataset, thereby correcting the causal structure of the structured data SD.

[0027] Furthermore, for example, in structured data SD, if there is no edge E between nodes N that were assumed to be causally related, and the causal relationship cannot be confirmed, the user shall take the following measures: When operating the injection molding machine 100 for the next time or later to acquire additional data, the user shall modify the injection molding machine 100 or add sensors 151 to 153 so that sensing data that can confirm the causal relationship between nodes N that were assumed to be causally related can be acquired. For example, at least one of two or more nodes N that were assumed to be causally related may be represented by the features of the time series data when the manufactured product was produced. In this case, by manufacturing two or more manufactured products consecutively, two or more time series data may be acquired as a single waveform data. If a synchronization signal is not provided to each time series data that makes up the waveform data, when extracting time series data for each manufactured product from the waveform data, an incorrect section of the waveform data may be extracted as time series data. If an incorrect section of the waveform data is extracted as time series data, the correct features may not be acquired for at least one of the two or more manufactured products. In this case, the correlation coefficient between nodes N that were assumed to be causally related may not be calculated correctly, and structured data SD may be generated in which there is no edge E between nodes N that were assumed to be causally related. In such cases, the user modifies the injection molding machine 100 so that a synchronization signal can be obtained when the injection molding machine 100 is operated for the next time to acquire additional data. In this way, the synchronization signal can be used to correctly extract time-series data from waveform data, and thus the correct features of the time-series data can be obtained. As a result, the structured data SD is corrected when the structuring unit 512 generates structured data SD with edges E added between nodes N that were assumed to be causally related using the new dataset.

[0028] Furthermore, for example, in structured data SD, if an edge E exists between nodes N that were not expected to have a causal relationship, the user takes measures to remove the edge E between the nodes N that were not expected to have a causal relationship. For example, if the first heater and the second heater are controlled independently, and there is no correlation between node N representing the variable V related to the first heater and node N representing the variable V related to the second heater, the structured data SD is expected to take the following form. In this case, the structured data SD is expected to take the form in which there is no edge E between node N representing the variable V related to the first heater and node N representing the variable V related to the second heater. However, because the first heater and the second heater are close together, a correlation may occur between the first heater and the second heater. Thus, due to the correlation between the first heater and the second heater, an edge E may exist between node N representing the variable V related to the first heater and node N representing the variable V related to the second heater in the structured data SD. In such a case, the user takes measures to remove the edge E between node N representing the variable V for the first heater and node N representing the variable V for the second heater.

[0029] In this case, if no additional data is required to delete some edges E in the structured data SD, the user can retrospectively specify the edges E to be deleted, for example, via the input device 506. This corrects the causal relationships in the structured data SD, as the structuring unit 512 modifies the structured data SD by deleting the edges E specified by the user. On the other hand, if additional data is required to delete some edges E in the structured data SD, the user can, for example, change the hardware configuration of the injection molding machine 100, specifically the configuration of the node N that was assumed to have no causal relationship with the variable V. For example, if the distance between the first heater and the second heater is short, and an edge E exists between the node N representing the variable V related to the first heater and the node N representing the variable V related to the second heater in the structured data SD, the user can do the following: In this case, the user changes the arrangement of at least one of the first heater and the second heater in the injection molding machine 100 to increase the distance between the first heater and the second heater. As a result, the structuring unit 512 generates structured data SD using a new dataset in which there are no edges E between the nodes N specified by the user, thereby correcting the causal relationships in the structured data SD.

[0030] Furthermore, for example, if an edge E exists between nodes N that were assumed to have no causal relationship, the user may take the following measures: The user may take measures to add an intermediate node MN to the structured data SD that represents a new variable V correlated with a variable V represented by at least one of the two or more nodes N that were assumed to have no causal relationship. For example, if the distance between the first heater and the second heater is close and an edge E exists in the structured data SD between a node N representing a variable V related to the first heater and a node N representing a variable V related to the second heater, the user may take the following measures: In this case, the user may take measures to add an intermediate node MN to the structured data SD that represents an influencing variable IV indicating the distance between the first heater and the second heater.

[0031] In this case, if no additional data is required to add the intermediate node MN to the structured data SD, the user retrospectively specifies, for example, the content of the variable V represented by the node N to be added and the hierarchy of the node N to be added, via the input device 506. As a result, the structuring unit 512 generates structured data SD in which the node N representing the variable V specified by the user is added to the hierarchy specified by the user, thereby correcting the causal relationships of the structured data SD. On the other hand, if additional data is required to add the intermediate node MN to the structured data SD, the user specifies, for example, the following information via the input device 506 as prior information for when the structured data SD is generated next time using a new dataset. In this case, the user pre-specifies the content of the variable V represented by the node N to be added and the hierarchy of the node N to be added. As a result, the structuring unit 512 generates structured data SD with the node N added using the new dataset, thereby correcting the causal relationships of the structured data SD.

[0032] If experts do not identify any deficiencies or shortcomings in the structured data (SD), that is, if both steps S351 and S352 are answered "No", the structured data (SD) is considered complete, and the modeling process in step S400 is executed.

[0033] The modeling process is the process of modeling the injection molding machine 100 using a dataset and structured data SD. In structured data SD, an edge E means that it has some influence from the starting point to the ending point. Therefore, in the modeling process, the modeling unit 513 sequentially calculates an explanatory function, which is a function that expresses the variable V of the higher-level node that is the endpoint of a particular edge E using the lower-level node that is the starting point of that particular edge E, from the leaf node LN to the root node RN. In other words, the modeling unit 513 calculates an explanatory function for each level in structured data SD. Then, the modeling unit 513 models the injection molding machine 100 by generating a model function, which is a function that includes the calculated explanatory functions. In other words, the model function is a function that views the edge E as the propagation of numerical values.

[0034] Figure 6 illustrates how to calculate the explanatory function for a hierarchy having an edge E connecting one lower node UN to one upper node AN. The first upper variable Z1, as the variable V represented by the upper node AN, can be expressed using the first lower variable A, as the variable V represented by the lower node UN, for example, as shown in equation (1) below. Z1=kA×A+t1 Formula (1) In equation (1) above, the first independent coefficient kA is a coefficient that represents the influence that the first sub-variable A has on the first superordinate variable Z1 on its own. The first constant t1 is a constant that represents the influence that the first sub-variable A has on the first superordinate variable Z1 on its own.

[0035] The modeling unit 513 first calculates the first single coefficient kA and the first constant t1. The modeling unit 513 calculates the first single coefficient kA and the first constant t1 using, for example, the function calculation model MD2 pre-stored in the memory device 502. The function calculation model MD2 is a machine learning model that calculates the coefficients and constants of a function that represents the causal relationship between the lower node UN and the upper node AN as shown in equation (1) above. When a variable V represented by the lower node UN and a variable V represented by the upper node AN are input, the function calculation model MD2 outputs coefficients and constants that represent the influence that the variable V of the lower node UN has on the variable V of the upper node AN. The function calculation model MD2 calculates the coefficients and constants that represent the influence that the variable V of the lower node UN has on the variable V of the upper node AN by, for example, approximating the variable V using the least squares method. The modeling unit 513 inputs the first lower variable A and the first upper variable Z1 included in the dataset into the function calculation model MD2. Based on this, the modeling unit 513 calculates the first single coefficient kA and the first constant t1.

[0036] Next, the modeling unit 513 calculates an explanatory function representing the first superordinate variable Z1 using the first subordinate variable A, the calculated coefficient kA, and the constant t1. The modeling unit 513 calculates an explanatory function such as that shown in equation (2) below. Specifically, the modeling unit 513 generates a term obtained by multiplying the first individual coefficient kA by the first subordinate variable A. Then, the modeling unit 513 calculates an explanatory function representing the first relation value F1 using a formula obtained by adding the first constant t1 to this term. F1=kA×A+t1 Formula (2) In equation (2) above, the first relation value F1 is an index that represents the relationship between the first sub-variable A and the first super-variable Z1.

[0037] Figure 7 illustrates how to calculate the explanatory function of a hierarchy having an edge E connecting two lower nodes UN to one upper node AN. The second upper variable Z2, as the variable V represented by the upper node AN, can be expressed, for example, as shown in equation (3) below, using the first lower variable A and the second lower variable B, as the variables V represented by the two lower nodes UN, respectively. Z2=kA×A+kB×B+kAB×A×B+t2 Formula (3) In equation (3) above, the first single coefficient kA is the coefficient representing the influence of the first subvariate A on the second supervariate Z2 on its own. The second single coefficient kB is the coefficient representing the influence of the second subvariate B on the second supervariate Z2 on its own. The first reciprocal coefficient kAB is the coefficient representing the influence of the interaction between the first subvariate A and the second subvariate B on the second supervariate Z2. The second constant t2 is a constant representing the influence of the first subvariate A and the second subvariate B on the second supervariate Z2.

[0038] The modeling unit 513 first inputs the first sub-variable A, the second sub-variable B, and the second supernumerator Z2 included in the dataset into the function calculation model MD2. Based on this, the modeling unit 513 calculates the first single coefficient kA, the second single coefficient kB, the first mutual coefficient kAB, and the second constant t2. Next, the modeling unit 513 calculates an explanatory function representing the second supernumerator Z2 using the first sub-variable A and the second sub-variable B, the calculated coefficients kA, kB, kAB, and the constant t2. For example, the modeling unit 513 calculates an explanatory function as shown in equation (4) below. Specifically, the modeling unit 513 generates a term obtained by multiplying the first single coefficient kA by the first sub-variable A, and a term obtained by multiplying the second single coefficient kB by the second sub-variable B. Furthermore, the modeling unit 513 generates a term obtained by multiplying the first mutual coefficient kAB by the first sub-variable A and the second sub-variable B. Then, the modeling unit 513 calculates an explanatory function representing the second relation value F2 using a mathematical formula that adds a second constant t2 to these terms. F2=kA×A+kB×B+kAB×A×B+t2 Formula (4) In equation (4) above, the second relation value F2 is an index that represents the relationship between the first subvariate A and the second subvariate B and the second supervariate Z2.

[0039] Figure 8 illustrates how to calculate the explanatory function of a hierarchy having an edge E connecting three subnodes UN to one supernode AN. The third super-major variable Z3, as the variable V represented by the supernode AN, can be expressed, for example, as shown in equation (5) below, using the first sub-variable A, the second sub-variable B, and the third sub-variable C, as the variables V represented by the three subnodes UN, respectively. Z3=kA×A+kB×B+kC×C +kAB×A×B+kBC×B×C+kCA×C×A+t3 Formula (5) In equation (5) above, the first single coefficient kA is the coefficient representing the influence of the first subvariable A on the third supervariable Z3 on its own. The second single coefficient kB is the coefficient representing the influence of the second subvariable B on the third supervariable Z3 on its own. The third single coefficient kC is the coefficient representing the influence of the third subvariable C on the third supervariable Z3 on its own. The first mutual coefficient kAB is the coefficient representing the influence of the interaction between the first subvariable A and the second subvariable B on the third supervariable Z3. The second mutual coefficient kBC is the coefficient representing the influence of the interaction between the second subvariable B and the third subvariable C on the third supervariable Z3. The third mutual coefficient kCA is the coefficient representing the influence of the interaction between the third subvariable C and the first subvariable A on the third supervariable Z3. The third constant t3 is a constant representing the influence of the first subvariable A, the second subvariable B, and the third subvariable C on the third supervariable Z3.

[0040] The modeling unit 513 first inputs the first sub-variable A, second sub-variable B, third sub-variable C, and third supernumerator Z3 included in the dataset into the function calculation model MD2. Based on this, the modeling unit 513 calculates the first single coefficient kA, the second single coefficient kB, the third single coefficient kC, the first mutual coefficient kAB, the second mutual coefficient kBC, the third mutual coefficient kCA, and the third constant t3. Next, the modeling unit 513 calculates an explanatory function representing the third supernumerator Z3 using the first sub-variable A, the second sub-variable B, and the third sub-variable C, along with the calculated coefficients kA, kB, kC, kAB, kBC, kCA, and the constant t3. For example, the modeling unit 513 calculates an explanatory function as shown in equation (6) below. Specifically, the modeling unit 513 generates a term obtained by multiplying the first individual coefficient kA by the first sub-variable A, a term obtained by multiplying the second individual coefficient kB by the second sub-variable B, and a term obtained by multiplying the third individual coefficient kC by the third sub-variable C. Furthermore, the modeling unit 513 generates a term obtained by multiplying the first mutual coefficient kAB by the first sub-variable A and the second sub-variable B, a term obtained by multiplying the second mutual coefficient kBC by the second sub-variable B and the third sub-variable C, and a term obtained by multiplying the third mutual coefficient kCA by the third sub-variable C and the first sub-variable A. Then, the modeling unit 513 calculates an explanatory function representing the third relation value F3 using a formula that adds a third constant t3 to these terms. F3 = kA × A + kB × B + kC × C +kAB×A×B+kBC×B×C+kCA×C×A+t3 Formula (6) In equation (6) above, the third relation value F3 is an index that represents the relationship between the first sub-variable A, the second sub-variable B, and the third sub-variable C, and the third supernumerator variable Z3.

[0041] Figure 9 shows a specific example of modeling an injection molding machine 100 using a model function. Figure 9 illustrates an example of structured data SD related to the injection molding machine 100. In the example shown in Figure 9, the condition variable CV represented by the leaf node LN is the temperature P of the heating cylinder 122, the temperature Q of the nozzle 124, the position R of the screw 123 when switching from injection to holding pressure, and the injection speed U of the molten material M2. The influence variable IV represented by the intermediate node MN is the position W of the heating cylinder 122 at the end of injection. The objective variable OV represented by the root node RN is the weight O of the manufactured product.

[0042] The position W of the heating cylinder 122 at the end of injection can be expressed, for example, as shown in equation (7) below, using the temperature Q of the nozzle 124 connected by the edge E, the position R of the screw 123 when switching from injection to holding pressure, and the injection speed U of the molten material M2. W = kQ1 × Q + kR1 × R + kU × U +kQR1×Q×R+kRU×R×U+kUQ×U×Q+t4 Equation (7) In equation (7) above, the fourth individual coefficient kQ1 is a coefficient representing the effect of the temperature Q of the nozzle 124 on the position W of the heating cylinder 122 at the end of injection. The fifth individual coefficient kR1 is a coefficient representing the effect of the position R of the screw 123 when switching from injection to holding pressure on the position W of the heating cylinder 122 at the end of injection. The sixth individual coefficient kU is a coefficient representing the effect of the injection speed U of the molten material M2 on the position W of the heating cylinder 122 at the end of injection. The fourth mutual coefficient kQR1 is a coefficient representing the effect of the interaction between the temperature Q of the nozzle 124 and the position R of the screw 123 when switching from injection to holding pressure on the position W of the heating cylinder 122 at the end of injection. The fifth mutual coefficient kRU is a coefficient representing the effect of the interaction between the position R of the screw 123 when switching from injection to holding pressure and the injection speed U of the molten material M2 on the position W of the heating cylinder 122 at the end of injection. The sixth reciprocal coefficient kUQ is a coefficient that represents the effect of the interaction between the injection velocity U of the molten material M2 and the temperature Q of the nozzle 124 on the position W of the heating cylinder 122 at the end of injection. The fourth constant t4 is a constant that represents the effect of the temperature Q of the nozzle 124, the position R of the screw 123 when switching from injection to holding pressure, and the injection velocity U of the molten material M2 on the position W of the heating cylinder 122 at the end of injection.

[0043] Furthermore, the weight O of the manufactured product can be expressed, for example, as shown in equation (8) below, using the temperature P of the heating cylinder 122 connected by the edge E, the temperature Q of the nozzle 124, the position R of the screw 123, and the position W of the heating cylinder 122. O = kP × P + kQ² × Q + kR² × R + kW × W +kPQ×P×Q+kQR2×Q×R+kRW×R×W+kWP×W×P+t5 Formula (8) In equation (8) above, the seventh individual coefficient kP is a coefficient representing the effect of the temperature P of the heating cylinder 122 on the weight O of the product. The eighth individual coefficient kQ2 is a coefficient representing the effect of the temperature Q of the nozzle 124 on the weight O of the product. The ninth individual coefficient kR2 is a coefficient representing the effect of the position R of the screw 123 when switching from injection to holding pressure on the weight O of the product. The tenth individual coefficient kW is a coefficient representing the effect of the position W of the heating cylinder 122 at the end of injection on the weight O of the product. The seventh mutual coefficient kPQ is a coefficient representing the effect of the interaction between the temperature P of the heating cylinder 122 and the temperature Q of the nozzle 124 on the weight O of the product. The eighth mutual coefficient kQR2 is a coefficient representing the effect of the interaction between the temperature Q of the nozzle 124 and the position R of the screw 123 when switching from injection to holding pressure on the weight O of the product. The ninth mutual coefficient kRW is a coefficient representing the effect of the interaction between the position R of the screw 123 when switching from injection to holding pressure and the position W of the heating cylinder 122 at the end of injection on the weight O of the product. The tenth interaction coefficient kWP is a coefficient that represents the effect of the interaction between the position W of the heating cylinder 122 and the temperature P of the heating cylinder 122 at the end of injection on the weight O of the product. The fifth constant t5 is a constant that represents the effect of the temperature P of the heating cylinder 122, the temperature Q of the nozzle 124, the position R of the screw 123 when switching from injection to holding pressure, and the position W of the heating cylinder 122 at the end of injection on the weight O of the product.

[0044] The modeling unit 513 inputs the temperature P of the heating cylinder 122, the temperature Q of the nozzle 124, the position R of the screw 123 when switching from injection to holding pressure, the injection speed U of the molten material M2, and the weight O of the manufactured product, all included in the dataset, into the function calculation model MD2. Based on this, the modeling unit 513 calculates the fourth individual coefficient kQ1, the fifth individual coefficient kR1, the sixth individual coefficient kU, the seventh individual coefficient kP, the eighth individual coefficient kQ2, the ninth individual coefficient kR2, and the tenth individual coefficient kW. Furthermore, the modeling unit 513 calculates the fourth mutual coefficient kQR1, the fifth mutual coefficient kRU, the sixth mutual coefficient kUQ, the seventh mutual coefficient kPQ, the eighth mutual coefficient kQR2, the ninth mutual coefficient kRW, and the tenth mutual coefficient kWP. Furthermore, the modeling unit 513 calculates the fourth constant t4 and the fifth constant t5. In this way, by calculating the coefficients kQ1, kR1, kU, kP, kQ2, kR2, kW, kQR1, kRU, kUQ, kPQ, kQR2, kRW, kWP, and constants t4, t5, the modeling unit 513 generates the following model function MF. The modeling unit 513 generates a model function MF that represents the weight O of the manufactured product using the temperature P and position R of the heating cylinder 122, the temperature Q of the nozzle 124, the position W of the screw 123, and the calculated coefficients kP, kQ2, kR2, kW, kPQ, kQR2, kRW, kWP, and constant t5. For example, the modeling unit 513 generates a model function MF as shown in equation (9) below. Specifically, the modeling unit 513 generates a term obtained by multiplying the seventh single coefficient kP by the temperature P of the heating cylinder 122, and a term obtained by multiplying the eighth single coefficient kQ2 by the temperature Q of the nozzle 124. Furthermore, the modeling unit 513 generates a term obtained by multiplying the ninth individual coefficient kR2 by the position R of the screw 123 when switching from injection to holding pressure, and a term obtained by multiplying the tenth individual coefficient kW by the position W of the heating cylinder 122 at the end of injection. Furthermore, the modeling unit 513 generates a term obtained by multiplying the seventh mutual coefficient kPQ by the temperature P of the heating cylinder 122 and the temperature Q of the nozzle 124. Furthermore, the modeling unit 513 generates a term obtained by multiplying the eighth mutual coefficient kQR2 by the temperature Q of the nozzle 124 and the position R of the screw 123 when switching from injection to holding pressure.Furthermore, the modeling unit 513 generates a term obtained by multiplying the ninth reciprocal coefficient kRW by the position R of the screw 123 when switching from injection to holding pressure and the position W of the heating cylinder 122 at the end of injection. Furthermore, the modeling unit 513 generates a term obtained by multiplying the tenth reciprocal coefficient kWP by the position W of the heating cylinder 122 at the end of injection and the temperature P of the heating cylinder 122. Finally, the modeling unit 513 generates a mathematical formula by adding the fifth constant t5 to these terms as the model function MF. MF=kP×P+kQ2×Q+kR2×R+kW×W+kPQ×P×Q +kQR2×Q×R+kRW×R×W+kWP×W×P+t5 Equation (9)

[0045] Furthermore, the position W of the heating cylinder 122 shown in equation (9) above can be expressed by the temperature Q of the nozzle 124, the position R of the screw 123, the injection velocity U of the molten material M2, and the coefficients kQ1, kR1, kU, kQR1, kRU, kUQ, and the constant t4, as shown in equation (7) above. Therefore, the model function MF shown in equation (9) above includes the explanatory function shown in equation (10) below. F4 = kQ1 × Q + kR1 × R + kU × U +kQR1×Q×R+kRU×R×U+kUQ×U×Q+t4 Equation (10) In equation (10) above, the fourth relation value F4 is an index representing the relationship between the temperature Q of the nozzle 124, the position R of the screw 123 when switching from injection to holding pressure, the injection speed U of the molten material M2, and the position W of the heating cylinder 122 at the end of injection. In calculating the fourth relation value F4, the modeling unit 513 generates a term obtained by multiplying the fourth individual coefficient kQ1 by the temperature Q of the nozzle 124, a term obtained by multiplying the fifth individual coefficient kR1 by the position R of the screw 123 when switching from injection to holding pressure, and a term obtained by multiplying the sixth individual coefficient kU by the injection speed U of the molten material M2. Furthermore, the modeling unit 513 generates a term obtained by multiplying the fourth mutual coefficient kQR1 by the temperature Q of the nozzle 124 and the position R of the screw 123 when switching from injection to holding pressure. Furthermore, the modeling unit 513 generates a term obtained by multiplying the fifth reciprocal coefficient kRU by the position R of the screw 123 when switching from injection to holding pressure and the injection speed U of the molten material M2. Furthermore, the modeling unit 513 generates a term obtained by multiplying the sixth reciprocal coefficient kUQ by the injection speed U of the molten material M2 and the temperature Q of the nozzle 124. Then, the modeling unit 513 calculates an explanatory function representing the fourth relation value F4 by adding the fourth constant t4 to these terms in a mathematical formula.

[0046] According to the first embodiment described above, the data processing device 50 can acquire a dataset containing two or more acquired data sets that associate a condition variable CV representing the operating conditions of the injection molding machine 100 with a target variable OV relating to the manufactured product. The data processing device 50 then performs a causal structure search using the dataset and generates structured data SD representing the causal structure between the condition variable CV and the target variable OV by connecting nodes N with directed edges E. The data processing device 50 then uses the dataset and structured data SD to calculate explanatory functions that express the variable V of the upper node AN using the variable V of the lower node UN, sequentially from the leaf node LN to the root node RN, moving from the lower to the upper levels of the hierarchy. The data processing device 50 then generates a model function MF, which is a function that includes the calculated explanatory functions. As a result, the data processing device 50 can identify the relationship between the operating conditions of the injection molding machine 100 and the target variable OV by representing the relationship between the condition variable CV and the target variable OV using the model function MF. At this time, the data processing device 50 calculates explanatory functions sequentially from the leaf node LN to the upper levels of the hierarchy. Therefore, as shown in equation (10) above, the data processing device 50 can use the condition variable CV to represent the influencing variable IV, which represents the elements that affect the target variable OV. In other words, the data processing device 50 can quantitatively represent various elements that affect the target variable OV using the condition variable CV. As a result, the data processing device 50 can identify the relationship between the operating conditions of the injection molding machine 100 and the target variable OV without acquiring a large amount of data and preparing a matrix for each of the various elements that affect the target variable OV.

[0047] Furthermore, according to the first embodiment described above, as shown in equation (10) above, the influencing variable IV can be expressed using the conditional variable CV. In other words, the influencing variable IV does not necessarily have to exist as an actual measured value such as sensing data when generating the model function MF. Therefore, even if the influencing variable IV cannot be uniquely represented by sensing data or the like for a node N that is expected to exist as an intermediate node MN, the data processing device 50 can generate the model function MF in the following way, for example. In the reflection step shown in Figure 4, the user takes measures to add a hypothetical intermediate node MN to a node N that is expected to exist as an intermediate node MN. Furthermore, the user takes measures to connect the edges E that are input and output to the added intermediate node MN from all expected nodes N. In the structuring step, the data processing device 50 deletes unnecessary edges E from among the edges E that are input and output to the added intermediate node MN by deleting edges E whose correlation coefficient is less than a predetermined threshold. Using the structured data SD generated in this way, the data processing device 50 sequentially calculates an explanatory function that expresses the variable V of the upper node AN, which is the endpoint of a specific edge E, using the lower node UN, which is the starting point of that specific edge E, from the leaf node LN to the root node RN. This allows the data processing device 50 to generate a model function MF even when the influencing variable IV cannot be uniquely represented.

[0048] Furthermore, according to the first embodiment described above, the data processing device 50 can identify the causal relationship between the condition variable CV, the influence variable IV, and the target variable OV, and then, based on the identified causal relationship, identify the relationship between the operating conditions of the injection molding machine 100 and the target variable OV. Therefore, the data processing device 50 can identify the relationship between the operating conditions of the injection molding machine 100 and the target variable OV without acquiring and learning from a large amount of data, as is the case with deep learning.

[0049] Furthermore, according to the first embodiment described above, the data processing device 50 can calculate an explanatory function of a predetermined order or less. Specifically, the data processing device 50 can generate a first-order explanatory function consisting of addition and multiplication, for example, as shown in equation (4) above. In this way, by calculating an explanatory function of a predetermined order or less, the data processing device 50 can represent the relationship between the operating conditions of the injection molding machine 100 and the target variable OV using a simpler function, without having to perform simulations that reproduce physical phenomena.

[0050] In the first embodiment described above, the data processing device 50 calculated a first-order explanatory function, but it is not limited to this. The data processing device 50 may calculate a second-order or higher explanatory function representing the fifth relation value F5, for example, as shown in equation (11) below. F5 = kA² × A 2 +kA1×A formula (11) In equation (11) above, the 11th individual coefficient kA2 is the value obtained by squaring the first sub-variable A. 2 The coefficient represents the influence that the first lower variable A has on the higher variables Z1 to Z3. The twelfth single coefficient kA1 represents the influence that the first lower variable A has on the higher variables Z1 to Z3. The fifth relationship value F5 is an index that represents the relationship between the first lower variable A and the higher variables Z1 to Z3. In this form, the data processing device 50 can more accurately represent the relationship between the operating conditions of the injection molding machine 100 and the target variable OV.

[0051] Furthermore, according to the first embodiment described above, when there are two or more subnodes UN with respect to a higher node AN in the structured data SD, the data processing device 50 can generate the following model function MF. As shown in equation (9) above, the data processing device 50 can generate a model function MF that includes an explanatory function representing the influence that each of the two or more subnodes UN has on the higher node AN, and an explanatory function representing the influence that the interaction between the two or more subnodes UN has on the higher node AN. In this form, the data processing device 50 can represent both the influence that one subnode UN has on the higher node AN and the influence that the interaction between the two or more subnodes UN has on the higher node AN using the model function MF.

[0052] Furthermore, according to the first embodiment described above, the data processing device 50 can generate a model function MF that includes an explanatory function representing the effect that the interaction between a predetermined number of lower nodes UN has on the upper node AN. In this configuration, the data processing device 50 can suppress an increase in processing load during the process of generating the model function MF, and prevent the model function MF from becoming more complex.

[0053] Furthermore, the model function MF may include explanatory functions that represent the influence of the interaction between lower-node UNs on the upper-node AN, without including explanatory functions that represent the influence of each of the two or more lower-node UNs on the upper-node AN. Alternatively, it may include explanatory functions that represent the influence of each of the two or more lower-node UNs on the upper-node AN, without including explanatory functions that represent the influence of the interaction between lower-node UNs on the upper-node AN.

[0054] B. Second Embodiment: Figure 10 is a block diagram showing the configuration of the data processing device 50a in the second embodiment. When external environmental conditions such as temperature, humidity, and atmospheric pressure change over time during the manufacturing of a product, these changes in the external environment may affect the target variable OV. For example, changes in ambient temperature may affect water temperature, causing the weight of the product to change over time. Also, when the movable parts of the injection molding machine 100 deteriorate over time due to wear, or when the movable parts are replaced or repaired, the condition of the movable parts of the injection molding machine 100 may affect the target variable OV. For example, when the movable parts of the injection molding machine 100 are replaced with new ones or maintained through an overhaul, clogging may occur in proportion to a logarithmic function as the number of products manufactured increases. When clogging occurs in proportion to a logarithmic function as the number of products manufactured increases, the number of products manufactured per unit time and the cycle time may change over time. Furthermore, when material properties change over time, differences in material properties may affect the target variable OV. For example, if the lot or storage conditions of the molding material M1 used differ, the viscosity and moisture content of the molding material M1 may differ, which may cause variations in the quality of the manufactured product over time. In this way, it may become impossible to manufacture a product that satisfies the predetermined threshold range of the target variable OV during the manufacturing process. If the target variable OV deviates from the threshold range including the predetermined reference value during the manufacturing process, it is preferable to correct the deviation of the target variable OV from the reference value by modifying the operating conditions of the injection molding machine 100 so that the target variable OV falls within the threshold range. Therefore, in this embodiment, the data processing device 50a further has a function for modifying the operating conditions of the injection molding machine 100. The other configurations are the same as in the first embodiment unless otherwise specified. Components identical to those in the first embodiment are denoted by the same reference numerals and their descriptions are omitted.

[0055] The data processing device 50a is composed of a computer comprising a CPU 501a, a storage device 502a, a communication device 505, an input device 506, and a display device 507. The CPU 501a functions as an acquisition unit 511, a structuring unit 512, a modeling unit 513, a condition calculation unit 514, and a display control unit 516 and a manufacturing control unit 517, which are function units 515, by executing a program PG stored in the storage device 502a. Each function of the CPU 501a will be explained in conjunction with the data processing method shown below.

[0056] Figure 11 is a flowchart illustrating the control method of the injection molding machine 100 as a data processing method. For example, the user detects when the target variable OV deviates from a threshold range, based on the change in the target variable OV when two or more products are manufactured. The user then inputs information indicating that the target variable OV has deviated from the threshold range via the input device 506. The data processing device 50 then detects that the target variable OV has deviated from the threshold range by receiving input from the user. The data processing device 50 may also automatically detect that the target variable OV has deviated from the threshold range using various data, such as sensing data, acquired from the injection molding machine 100, without receiving input from the user. When it is detected that the target variable OV has deviated from the threshold range, the control method of the injection molding machine 100 shown in Figure 11 is executed.

[0057] In the control method for the injection molding machine 100, first, the condition calculation step in step S500 is performed. The condition calculation step is a process of calculating new operating conditions for the injection molding machine 100 in order to bring the time-varying target variable OV closer to a reference value, using a model function MF. In the condition calculation step, the condition calculation unit 514 substitutes a condition variable CV, which represents an arbitrary operating condition different from the current operating conditions of the injection molding machine 100, into the model function MF. As a result, the condition calculation unit 514 predicts the target variable OV for the product manufactured by operating the injection molding machine 100 under the arbitrary operating conditions. While repeating this process, the condition calculation unit 514 searches for operating conditions that can bring the predicted value of the target variable OV closer to a predetermined value related to the reference value. As a result, the condition calculation unit 514 calculates new operating conditions for the injection molding machine 100 that can bring the time-varying target variable OV closer to the reference value, that is, return the time-varying target variable OV to the reference value.

[0058] The condition calculation unit 514 calculates new operating conditions for the injection molding machine 100, for example, using the condition calculation model MD3 pre-stored in the memory device 502. The condition calculation model MD3 is a machine learning model that calculates optimal operating conditions according to a reference value of the target variable OV. When a predetermined value related to the reference value of the target variable OV is input, the condition calculation model MD3 outputs a new set value for the operating conditions of the injection molding machine 100. Various optimization algorithms for calculating the optimal solution can be used as the algorithm for the condition calculation model MD3, such as Newton's method, steepest descent method, amoeba method, swarm intelligence, and evolutionary algorithms.

[0059] For example, if the dependent variable OV is the weight of the product and the baseline weight of the product is 10g, and the weight of the product decreases to 8g over time, then to bring the weight of the product closer to the baseline, it is necessary to increase the weight of the product by 2g. Also, for example, if the dependent variable OV is the weight of the product and the baseline weight of the product is 10g, and the weight of the product increases to 12g over time, then to bring the weight of the product closer to the baseline, it is necessary to decrease the weight of the product by 2g. In reality, if the weight of the product changes over time without changing the operating conditions of the injection molding machine 100, and the condition calculation model MD3 uses the model function MF which does not take into account the elements representing the change over time, then the condition calculation model MD3 will only output the following information. In this case, even if the baseline weight of the product, 10g, is input to the condition calculation model MD3, the condition calculation model MD3 will only output the current operating condition setting. In other words, as long as the reference value of the manufactured product's weight remains the same, the operating conditions of the injection molding machine 100, as determined using the model function MF, are constant. Therefore, when the condition calculation model MD3 uses the model function MF, which does not take into account elements representing changes over time, the condition calculation unit 514 inputs a value into the condition calculation model MD3 that reflects the difference between the target variable OV that has changed over time and the reference value, as a predetermined value related to the reference value of the target variable OV. Specifically, if the weight of the manufactured product decreases over time from the reference value of 10g to 8g, the condition calculation unit 514 inputs 12g into the condition calculation model MD3 by adding the difference of 2g from the reference value to the reference value of 10g. If the weight of the manufactured product increases over time from the reference value of 10g to 12g, the condition calculation unit 514 inputs 8g into the condition calculation model MD3 by subtracting the difference of 2g from the reference value to the reference value of 10g. In this way, the condition calculation unit 514 can calculate new operating conditions for the injection molding machine 100 using a model function MF that does not take into account elements representing changes over time.

[0060] Furthermore, the operating conditions of the injection molding machine 100 may have upper and lower limits that can be set as set values. Therefore, if the newly calculated operating conditions fall outside the acceptable range determined according to the upper and lower limits that can be set as operating conditions of the injection molding machine 100, the condition calculation unit 514 may output error information via the display device 507 or the like. In this way, the user can detect abnormalities during manufacturing.

[0061] Furthermore, the condition calculation unit 514 may calculate new operating conditions that fall within the acceptable range. This prevents the condition calculation unit 514 from calculating unrealistic operating conditions.

[0062] Furthermore, the condition calculation unit 514 may output information representing the priority of the operating conditions to be changed among multiple operating conditions as supplementary information for the new operating conditions. The priority of the operating conditions is determined, for example, according to the time required for the measured value to change in response to a change in the setting value of the operating conditions. For example, if the target variable OV can be brought closer to the reference value by changing either the setting value of the injection pressure or the temperature of each part of the injection molding machine 100 as operating conditions, the condition calculation unit 514 will do the following: When manufacturing two or more products in succession, the injection pressure can be changed for each shot. This allows the injection pressure to be changed immediately from the time the next product is manufactured. On the other hand, it may take time for the temperature of each part of the injection molding machine 100 to change to the desired temperature. Therefore, the condition calculation unit 514 outputs supplementary information to change the setting value of the injection pressure. In this way, the waiting time required until a product with the target variable OV within the threshold range can be manufactured can be shortened. This helps to suppress a decrease in manufacturing efficiency.

[0063] Following the condition calculation process, the functional process S600 is executed. The functional process is executed when new operating conditions for the injection molding machine 100 are calculated.

[0064] If the new operating conditions for the injection molding machine 100 calculated in the condition calculation step are within a preset tolerance range, that is, if step S610 is "YES", the manufacturing control step of step S630 is executed. The manufacturing control step is the process of operating the injection molding machine 100 with the new operating conditions calculated in the condition calculation step. In the manufacturing control step, the manufacturing control unit 517 generates an operating control signal that represents the new operating conditions calculated in the condition calculation step. The manufacturing control unit 517 then transmits the generated operating control signal to the injection molding machine 100. As a result, the manufacturing control unit 517 operates the injection molding machine 100 with the new operating conditions.

[0065] If the new operating conditions for the injection molding machine 100 calculated in the condition calculation step are outside the acceptable range, that is, if step S610 is "No", the display control step of step S650 is executed. The display control step is a process that prompts the user of the injection molding machine 100 to change the operating conditions of the injection molding machine 100 by displaying the new operating conditions calculated in the condition calculation step on the display device 507. In the display control step, the display control unit 516 prompts the user of the injection molding machine 100 to change the operating conditions of the injection molding machine 100 by displaying the new operating conditions calculated in the condition calculation step on the display device 507. The user, for example, sets a value within the acceptable range according to the new operating conditions displayed on the display device 507. As a result, the display control unit 516 operates the injection molding machine 100 with the new operating conditions.

[0066] In other embodiments, the data processing device 50a may include a manufacturing control unit 517 instead of a display control unit 516. In other words, the control method for the injection molding machine 100 may include a manufacturing control step instead of a display control step. Furthermore, in other embodiments, the data processing device 50a may include a display control unit 516 instead of a manufacturing control unit 517. In other words, the control method for the injection molding machine 100 may include a display control step instead of a manufacturing control step.

[0067] According to the second embodiment described above, when it is detected that the target variable OV has deviated from the threshold range, the data processing device 50a can use a pre-generated model function MF to calculate new operating conditions to bring the target variable OV closer to the reference value. The data processing device 50a then displays the new operating conditions on the display device 507 and prompts the user of the injection molding machine 100 to change the operating conditions, thereby enabling the injection molding machine 100 to be operated under the new operating conditions. Furthermore, if the new operating conditions are within the acceptable range, the data processing device 50a can automatically operate the injection molding machine 100 under the new operating conditions without receiving input from the user. In this way, the data processing device 50a can correct the deviation of the target variable OV from the reference value. Therefore, the user can manufacture the desired product.

[0068] Furthermore, according to the second embodiment described above, the data processing device 50a can predict the target variable OV by substituting an arbitrary condition variable CV into the model function MF. The data processing device 50a can then calculate new operating conditions for the injection molding machine 100 by searching for operating conditions that can bring the predicted value of the target variable OV closer to a predetermined value determined according to the difference from a reference value. In other words, the data processing device 50a can calculate new operating conditions for the injection molding machine 100 through calculations within the data processing device 50a itself, without having to acquire new data and search for new operating conditions through trial and error, as in Bayesian optimization. As a result, the data processing device 50a can avoid an increase in the defect rate of manufactured products while calculating new operating conditions. Therefore, the data processing device 50a can avoid unstable conditions such as an increase in the defect rate of manufactured products during the period in which new operating conditions are calculated, resulting in loss of molding material M1 or a decrease in manufacturing efficiency.

[0069] Furthermore, even when the reference value of the objective variable OV itself changes, such as when the weight of the manufactured product changes due to a change in the specifications of the manufactured product, the data processing device 50a can use the model function MF to calculate the new operating conditions for the injection molding machine 100.

[0070] Furthermore, according to the second embodiment described above, the model function MF can include explanatory functions of a predetermined order or less. In this configuration, the data processing device 50a can reduce the processing load when calculating new operating conditions for the injection molding machine 100. As a result, the data processing device 50a can shorten the time required to calculate new operating conditions for the injection molding machine 100.

[0071] Furthermore, according to the second embodiment described above, the target variable OV is, for example, quality information. In this configuration, the data processing device 50a can use the model function MF to predict the quality of the manufactured product according to the operating conditions of the injection molding machine 100. When it is detected that the quality of the manufactured product has deviated from the threshold range, the data processing device 50a can calculate operating conditions to bring the quality of the manufactured product closer to the standard value and operate the injection molding machine 100. This allows the user to manufacture products of the desired quality.

[0072] Furthermore, according to the second embodiment described above, the target variable OV may be the cycle time when manufacturing the product. In this configuration, the data processing device 50a can predict the cycle time according to the operating conditions of the injection molding machine 100 using the model function MF. When it is detected that the cycle time has deviated from the threshold range, the data processing device 50a can calculate operating conditions to bring the cycle time closer to the reference value and operate the injection molding machine 100. As a result, the user can manufacture products with the desired cycle time.

[0073] Furthermore, according to the second embodiment described above, the objective variable OV may be the variation in the quality of the manufactured product. In this configuration, the data processing device 50a can use the model function MF to predict the variation in the quality of the manufactured product that occurs according to the operating conditions of the injection molding machine 100. The data processing device 50a can then calculate the operating conditions necessary to suppress the variation in the quality of the manufactured product and operate the injection molding machine 100 accordingly. As a result, the user can manufacture two or more manufactured products with less variation in quality.

[0074] C. Third Embodiment: Figure 12 is a block diagram showing the configuration of the data processing device 50b in the third embodiment. Figure 13 is a conceptual diagram of the structured data SD generated in the third embodiment. In this embodiment, the data processing device 50b generates a model function MF that takes into account elements representing changes over time. When it is detected that the target variable OV has deviated from the threshold range, the data processing device 50b uses the model function MF that takes into account elements representing changes over time to calculate new operating conditions for the injection molding machine 100. The other configurations are the same as in the second embodiment unless otherwise specified. Components identical to those in the second embodiment are denoted by the same reference numerals and their descriptions are omitted.

[0075] The data processing device 50b is composed of a computer comprising a CPU 501b, a storage device 502b, a communication device 505, an input device 506, and a display device 507. The CPU 501b executes a program PG stored in the storage device 502b, thereby functioning as an acquisition unit 511b, a structuring unit 512b, a modeling unit 513b, a condition calculation unit 514b, and a display control unit 516 and a manufacturing control unit 517 as functional units 515.

[0076] The acquisition unit 511b acquires a dataset including the time-series variable TV. The time-series variable TV is an explanatory variable that represents the change over time when two or more products are manufactured. The time-series variable TV is, for example, an actual value of a parameter that changes over time. The parameter that changes over time is, for example, the number of products manufactured.

[0077] The structuring unit 512b generates structured data SD using a dataset that includes the time-series variable TV, with node N representing the time-series variable TV positioned below the root node RN.

[0078] The modeling unit 513b generates a model function MF using a dataset containing the time-series variable TV and structured data SD containing nodes N representing the time-series variable TV.

[0079] When it is detected that the target variable OV has deviated from the threshold range, the condition calculation unit 514b calculates new operating conditions for the injection molding machine 100 using a model function MF that incorporates elements representing changes over time. In this embodiment, the condition calculation model MD3 can utilize the model function MF that incorporates elements representing changes over time. Therefore, by inputting the reference value itself as a predetermined value related to the reference value of the target variable OV into the condition calculation model MD3, the condition calculation model MD3 can output new operating condition settings. Thus, for example, if the target variable OV is the weight of the product and the reference value of the product weight is 10g, when the weight of the product changes over time, the condition calculation unit 514b inputs the reference value of 10g into the condition calculation model MD3. As a result, the condition calculation unit 514b calculates new operating conditions for the injection molding machine 100.

[0080] According to the third embodiment described above, the data processing device 50b can further acquire the time-series variable TV. The data processing device 50b can generate structured data SD, which includes node N representing the time-series variable TV below the root node RN. The data processing device 50b can then use the dataset containing the time-series variable TV and the structured data SD, which includes node N representing the time-series variable TV, to generate a model function MF that incorporates elements representing changes over time. As a result, when it is detected that the target variable OV has deviated from the threshold range, the data processing device 50b can calculate new operating conditions for the injection molding machine 100 by searching for operating conditions that can bring the predicted value of the target variable OV closer to the reference value itself. In other words, the data processing device 50b can calculate new operating conditions for the injection molding machine 100 without calculating a predetermined value related to the reference value of the target variable OV using the difference from the reference value. Therefore, the data processing device 50b can reduce the processing load when calculating new operating conditions for the injection molding machine 100. This allows the data processing device 50b to reduce the time required to calculate new operating conditions for the injection molding machine 100.

[0081] Furthermore, according to the third embodiment described above, the data processing device 50b can generate a model function MF that takes into account elements representing changes over time. As a result, the data processing device 50b can use the model function MF to detect abnormalities during manufacturing or to predict changes in the target variable OV over time.

[0082] D. Fourth Embodiment: Figure 14 is a block diagram showing the configuration of the data processing device 50c in the fourth embodiment. Figure 15 is a conceptual diagram of the structured data SD1 to SD3 generated in the fourth embodiment. In this embodiment, the data processing device 50c generates a model function MF after concatenating the two structured data SDs. The other configurations are the same as in the first embodiment unless otherwise specified. Components identical to those in the first embodiment are denoted by the same reference numerals and their descriptions are omitted.

[0083] The data processing device 50c is composed of a computer comprising a CPU 501c, a storage device 502c, a communication device 505, an input device 506, and a display device 507. The CPU 501c functions as an acquisition unit 511c, a structuring unit 512c, a structural connection unit 518, and a modeling unit 513c by executing a program PG stored in the storage device 502c.

[0084] The acquisition unit 511c acquires a first dataset and a second dataset that includes the dependent variable OV in the first dataset as an explanatory variable. For example, the first dataset includes two or more acquired data from the first step of the multiple steps shown in Figure 2. The second dataset includes two or more acquired data from the second step, which is executed after the first step, of the multiple steps shown in Figure 2.

[0085] As shown in Figures 14 and 15, the structuring unit 512c generates first structured data SD1 using the first dataset and second structured data SD2 using the second dataset. For example, if the first dataset includes data acquired in the first process, the generated first structured data SD1 represents the causal structure between the condition variable CV and the target variable OV in the first process. If the second dataset includes data acquired in the second process, the generated second structured data SD2 represents the causal structure between the condition variable CV and the target variable OV in the second process.

[0086] The structural connection unit 518 generates composite structured data SD3 by connecting the first structured data SD1 and the second structured data SD2. For example, if the root node RN of the first structured data SD1 and the leaf node LN of the second structured data SD2 represent the same element, the structural connection unit 518 connects the first structured data SD1 and the second structured data SD2 by connecting the root node RN of the first structured data SD1 and the leaf node LN of the second structured data SD2, which represent the same element, to form an intermediate node MN. In this way, the structural connection unit 518 generates composite structured data SD3. For example, if the first structured data SD1 represents the causal structure in the first process and the second structured data SD2 represents the causal structure in the second process, the generated composite structured data SD3 represents the causal structure in the two processes spanning from the first process to the second process.

[0087] The modeling unit 513c generates a model function MF using the first dataset, the second dataset, and the composite structured data SD3. For example, if the composite structured data SD3 represents the causal structure in two processes spanning from the first to the second process, the generated model function MF is a function that models the state of the injection molding machine 100 in the two processes spanning from the first to the second process.

[0088] According to the fourth embodiment described above, the data processing device 50c can generate a composite structured data SD3 by linking two structured data SDs together. In this form, the data processing device 50c can easily generate a model function MF even when the causal structure between the condition variable CV and the target variable OV is complex. As a result, even when the causal structure between the condition variable CV and the target variable OV is complex, the data processing device 50c can model the injection molding machine 100 by representing the state of the injection molding machine 100 with a single model function MF.

[0089] E. Fifth Embodiment: Figure 16 is a block diagram showing the configuration of the data processing device 50d in the fifth embodiment. In this embodiment, the data processing device 50d generates a composite model function by concatenating two model functions MF. The other configurations are the same as in the fourth embodiment unless otherwise specified. Components identical to those in the fourth embodiment are denoted by the same reference numerals and their descriptions are omitted.

[0090] The data processing device 50d is composed of a computer comprising a CPU 501d, a storage device 502d, a communication device 505, an input device 506, and a display device 507. The CPU 501d functions as an acquisition unit 511c, a structuring unit 512c, a modeling unit 513d, and a model linking unit 519 by executing a program PG stored in the storage device 502d.

[0091] The modeling unit 513d generates a first model function using the first structured data SD1 and a second model function using the second structured data SD2. For example, if the first structured data SD1 represents the causal structure in the first process, the generated first model function is a function that models the state of the injection molding machine 100 in the first process. If the second structured data SD2 represents the causal structure in the second process, the generated second model function is a function that models the state of the injection molding machine 100 in the second process.

[0092] The model coupling unit 519 generates a composite model function by coupling the first model function and the second model function. For example, if the first model function is a function that models the state of the injection molding machine 100 in the first process, and the second model function is a function that models the state of the injection molding machine 100 in the second process, the generated composite model function is as follows. In this case, the generated composite model function is a function that models the state of the injection molding machine 100 in two processes spanning from the first process to the second process.

[0093] According to the fifth embodiment described above, the data processing device 50d can generate a composite model function by linking two model functions MF together. In this configuration, the data processing device 50d can easily generate a composite model function even when the causal structure between the condition variable CV and the target variable OV is complex. As a result, even when the causal structure between the condition variable CV and the target variable OV is complex, the data processing device 50d can easily model the injection molding machine 100 by representing the state of the injection molding machine 100 with a single composite model function.

[0094] F. Other embodiments: F-1. Other Embodiments 1: When the relationship between a specific lower node UN and a higher node AN is known, and a physical quantity representing the relationship between the specific lower node UN and the higher node AN can be calculated using a known function, the modeling units 513, 513b to 513d may be as follows. In this case, the modeling units 513, 513b to 513d may generate a model function MF that includes a physical quantity representing the relationship between the specific lower node UN and the higher node AN, instead of an explanatory function for which the relationship between the specific lower node UN and the higher node AN is known. That is, the modeling units 513, 513b to 513d include a physical quantity that fixedly represents the edge E connecting the specific lower node UN and the higher node AN in the model function MF. The known function is a function that can represent the relationship between the specific lower node UN and the higher node AN without using a condition variable CV. The known function is, for example, a physically trivial function. The known function may also be a chemically trivial function. Furthermore, the known function may be a function that is trivial based on evidence such as simulations. For example, if the higher-level variables Z1 to Z3 can be explained by the displacement x in uniformly accelerated linear motion, the modeling unit 513 generates a model function MF that includes the displacement x as a physical quantity calculated by substituting the measured values ​​of initial velocity v0, time t, and acceleration a into equation (12) shown below. x = v0 × t + a × 1 / 2 × t 2 Formula (12) In this configuration, the data processing devices 50, 50b to 50d can generate a model function MF that includes physical quantities calculated using known functions, without having to calculate an explanatory function in which the relationship between a specific lower node UN and an upper node AN is known.

[0095] F-2. Other Embodiments 2: When the relationship between a specific lower node UN and a higher node AN is known, and it is not possible to calculate a physical quantity representing the relationship between the specific lower node UN and the higher node AN using a known function, the modeling units 513, 513b to 513d may do the following. In this case, the modeling units 513, 513b to 513d may generate a model function MF that includes a known function instead of an explanatory function for which the relationship between the specific lower node UN and the higher node AN is known. In this case, the modeling units 513, 513b to 513d may generate a model function MF that includes a known function KF of a higher order than a predetermined order such as first or second order. For example, a term including time t and acceleration a with time t squared t 2 a × t 2 When the superordinate variables Z1 to Z3 can be explained by adding the terms containing and , the modeling units 513, 513b to 513d generate a model function MF that includes the known function KF shown in equation (13) below. KF = kT × t + a × kAT × t 2 Formula (13) In equation (13) above, the first unknown coefficient kT is a coefficient that represents the effect of time t on the higher variables Z1~Z3. The second unknown coefficient kAT is the acceleration a multiplied by the square of time t. 2 a × t 2 This coefficient represents the influence of the supernumerator variables Z1 to Z3. In this configuration, the data processing devices 50, 50b to 50d can generate a model function MF that includes the known function KF without having to calculate an explanatory function in which the relationship between a specific lower node UN and a higher node AN is known.

[0096] F-3. Other Embodiments 3: The modeling units 513, 513b to 513d may generate explanatory functions that include various functions such as logarithmic functions, exponential functions, and trigonometric functions, as well as model functions MF that include known functions KF. In this form, if it is known empirically that the variable V of a particular node N exhibits nonlinear behavior, the data processing units 50, 50a to 50d can more accurately represent the relationship between the operating conditions of the injection molding machine 100 and the target variable OV.

[0097] F-4. Other Embodiments 4: The manufacturing apparatus 10 may be a molding machine other than the injection molding machine 100, such as a blow molding machine or a compression molding machine, or it may be a device other than a molding machine. If the manufacturing apparatus 10 is a device other than the injection molding machine 100, the expression "injection molding machine" in this disclosure may be replaced with "manufacturing apparatus" as appropriate.

[0098] F-5. Other Embodiments 5: At least some of the functions of the data processing devices 50, 50a to 50d may be implemented as a function of the manufacturing apparatus 10. At least some of the functions of the manufacturing apparatus 10 may be implemented as a function of the data processing devices 50, 50a to 50d. The manufacturing apparatus 10 and the data processing devices 50, 50a to 50d may be configured as an integrated unit or as separate units.

[0099] G. Other forms: This disclosure is not limited to the embodiments described above, and can be implemented in various configurations without departing from its spirit. For example, the technical features of the embodiments corresponding to the technical features in each form described in the summary of the invention can be replaced or combined as appropriate in order to solve some or all of the above-described problems, or to achieve some or all of the above-described effects. Furthermore, if a technical feature is not described as essential in this specification, it can be deleted as appropriate.

[0100] (1) According to the first embodiment of the present disclosure, a data processing device is provided. The data processing device includes: an acquisition unit that acquires a dataset containing two or more acquired data sets, which are acquired data obtained by operating a manufacturing device under predetermined operating conditions to manufacture a product, and which associate an objective variable relating to the product with an explanatory variable of the objective variable, which is a condition variable that is an explanatory variable representing the operating conditions; a structuring unit that uses the dataset to generate structured data representing the causal structure between the condition variable and the objective variable by connecting a leaf node representing the condition variable, a root node representing the objective variable, and an intermediate node located between the leaf node and the root node, which is an intermediate node that is an element that affects the objective variable and represents an element different from the operating conditions, with directed edges; and a modeling unit that uses the dataset to model the manufacturing device by sequentially calculating an explanatory function, which is a function that expresses the variable of the upper node that is the endpoint of a specific edge using the variable of the lower node that is the starting point of the specific edge, from the leaf node to the root node, and generating a model function that is a function that includes the calculated explanatory function. In this configuration, the data processing device can acquire a dataset containing two or more acquired data sets that associate condition variables representing the operating conditions of the manufacturing equipment with target variables relating to the manufactured product. The data processing device can then use the dataset to generate structured data representing the causal structure between the condition variables and the target variables. The data processing device then uses the dataset and structured data to sequentially calculate explanatory functions from the leaf nodes to the root node, expressing the variables of higher-level nodes using the variables of lower-level nodes. The data processing device then generates a model function, which is a function that includes the calculated explanatory functions. In this way, the data processing device can identify the relationship between the operating conditions of the manufacturing equipment and the target variable by representing the relationship between the condition variables and the target variable using the model function. At this time, the data processing device calculates explanatory functions sequentially from the leaf nodes upwards in the hierarchy. Therefore, the data processing device can use the condition variables to represent influencing variables that affect the target variable.In other words, a data processing device can quantitatively represent various factors that influence the target variable using condition variables. Therefore, a data processing device can identify the relationship between the operating conditions of the manufacturing equipment and the target variable without acquiring a large amount of data and preparing a matrix for each of the various factors that influence the target variable.

[0101] (2) In the above configuration, two or more products are manufactured, the acquisition unit further acquires time-series variables as explanatory variables that represent the changes over time when the two or more products are manufactured, and the structuring unit generates structured data that includes nodes representing the time-series variables below the root node. In this configuration, the data processing device can further acquire time-series variables that represent changes over time. The data processing device can then generate structured data that includes nodes representing time-series variables below the root node. In this way, the data processing device can use the dataset containing time-series variables and the structured data containing nodes representing time-series variables to generate a model function that incorporates elements representing changes over time. As a result, the data processing device can use the model function to predict the changes over time of the target variable.

[0102] (3) In the above configuration, the modeling unit may calculate the explanatory function of a predetermined order or less. According to this configuration, the data processing unit can calculate an explanatory function of a predetermined order or less. This allows the data processing unit to represent the relationship between the operating conditions of the manufacturing equipment and the objective variable using a simpler model function.

[0103] (4) In the above embodiment, if the relationship between a particular subordinate node and the superior node is known, and a physical quantity representing the relationship can be calculated using a known function which is a function that represents the relationship without using the condition variable, the modeling unit may generate the model function which includes the physical quantity representing the relationship instead of the known explanatory function which is known for the relationship, without calculating the explanatory function which is known for the relationship. According to this embodiment, the data processing device can generate a model function which includes the physical quantity calculated using the known function without calculating the explanatory function which is known for the relationship between a particular subordinate node and the superior node.

[0104] (5) In the above embodiment, if the relationship between a particular subordinate node and the superior node is known, and the physical quantity representing the relationship cannot be calculated using a known function which is a function that represents the relationship without using the condition variable, the modeling unit may generate the model function which includes the known function instead of the explanatory function which is known for the relationship, without calculating the explanatory function which is known for the relationship. According to this embodiment, the data processing device can generate a model function which includes the known function without calculating the explanatory function which is known for the relationship between a particular subordinate node and the superior node.

[0105] (6) In the above embodiment, if there are two or more subnodes with respect to the upper node in the structured data, the modeling unit may generate a model function that includes an explanatory function representing the influence that each of the two or more subnodes has on the upper node, and an explanatory function representing the influence that the interaction between a predetermined number of subnodes has on the upper node. According to this embodiment, the data processing unit can generate a model function that includes an explanatory function representing the influence that each of the two or more subnodes has on the upper node, and an explanatory function representing the influence that the interaction between the two or more subnodes has on the upper node. As a result, the data processing unit can represent both the influence that one subnode has on the upper node and the influence that the interaction between two or more subnodes has on the upper node using the model function. Furthermore, according to this embodiment, the data processing unit can generate a model function that includes an explanatory function representing the influence that the interaction between a predetermined number of subnodes has on the upper node. As a result, the data processing unit can suppress an increase in processing load and complexity of the model function during the process of generating the model function.

[0106] (7) In the above embodiment, the acquisition unit acquires a first dataset and a second dataset that includes the target variable in the first dataset as an explanatory variable, the structuring unit generates first structured data using the first dataset and second structured data using the second dataset, and the data processing device may further include a structure linking unit that generates composite structured data by linking the first structured data and the second structured data. According to this embodiment, the data processing device can generate composite structured data by linking the two structured data, thereby combining the two structured data into one. As a result, the data processing device can easily generate a model function even when the causal structure between the condition variable and the target variable is complex.

[0107] (8) In the above embodiment, the acquisition unit acquires a first dataset and a second dataset that includes the target variable in the first dataset as an explanatory variable; the structuring unit generates first structured data using the first dataset and second structured data using the second dataset; the modeling unit generates a first model function using the first structured data and a second model function using the second structured data; and the data processing device may further include a model linking unit that generates a composite model function by linking the first model function and the second model function. According to this embodiment, the data processing device can generate a composite model function that combines the two model functions into one by linking the two model functions. As a result, even when the causal structure between the condition variable and the target variable is complex, the data processing device can easily model the manufacturing device by representing the state of the manufacturing device with a single composite model function.

[0108] (9) In the above embodiment, the data processing device may further include at least one of the following: a condition calculation unit that, when it is detected that the target variable has deviated from a threshold range including a predetermined reference value, uses the model function to calculate new operating conditions for bringing the target variable closer to the reference value; a display control unit that functions when the new operating conditions have been calculated, and prompts the user of the manufacturing device to change the operating conditions by displaying the new operating conditions on a display device; and a manufacturing control unit that operates the manufacturing device under the new operating conditions when the new operating conditions are within a preset tolerance range. In this embodiment, when it is detected that the target variable has deviated from a threshold range, the data processing device can calculate new operating conditions for bringing the target variable closer to the reference value using a pre-generated model function. The data processing device can then operate the manufacturing device under the new operating conditions by displaying the new operating conditions on a display device and prompting the user of the manufacturing device to change the operating conditions. Furthermore, if the new operating conditions are within a tolerance range, the data processing device can automatically operate the manufacturing device under the new operating conditions without receiving input from the user. This allows the data processing device to correct deviations of the target variable from the reference value. Therefore, the data processing device can use the manufacturing equipment to produce the desired product.

[0109] (10) A second embodiment of the present disclosure provides a data processing method. The data processing method includes: an acquisition step of acquiring a dataset which includes two or more acquired data sets that associate an objective variable relating to the product with an explanatory variable of the objective variable, which is an explanatory variable representing the operating conditions; a structuring step of using the dataset to generate structured data that represents the causal structure between the condition variable and the objective variable by connecting a leaf node representing the condition variable, a root node representing the objective variable, and an intermediate node located between the leaf node and the root node, which is an explanatory variable representing an element that affects the objective variable and is different from the operating conditions, with directed edges; and a modeling step of using the dataset to model the manufacturing apparatus by sequentially calculating an explanatory function from the leaf node to the root node, which represents an explanatory function that expresses the variable of the upper node that is the endpoint of a specific edge using the variable of the lower node that is the starting point of the specific edge, and generating a model function which is a function that includes the calculated explanatory function. This configuration allows for the acquisition of a dataset containing two or more data points that associate condition variables representing the operating conditions of the manufacturing equipment with target variables relating to the manufactured product. Using this dataset, structured data representing the causal structure between the condition variables and the target variables can be generated. Then, using the dataset and structured data, explanatory functions are calculated sequentially from the leaf nodes to the root node, expressing the variables of higher-level nodes using the variables of lower-level nodes. Finally, a model function is generated, which includes the calculated explanatory functions. This allows for the identification of the relationship between the operating conditions of the manufacturing equipment and the target variable by representing the relationship between the condition variables and the target variable using the model function. In this case, the explanatory functions are calculated sequentially from the leaf nodes upwards in the hierarchy. Therefore, influencing variables representing elements that affect the target variable can be expressed using the condition variables. In other words, various elements that affect the target variable can be quantitatively represented using the condition variables.Therefore, it is possible to identify the relationship between the operating conditions of the manufacturing equipment and the target variable without acquiring a large amount of data and preparing a matrix for each of the various factors that affect the target variable.

[0110] (11) In the above embodiment, the system may further include at least one of the following functional steps: a condition calculation step which, when it is detected that the target variable has deviated from a threshold range including a predetermined reference value, uses the model function to calculate new operating conditions for bringing the target variable closer to the reference value; a display control step which is executed when the new operating conditions have been calculated, by displaying the new operating conditions on a display device to prompt the user of the manufacturing equipment to change the operating conditions; and a manufacturing control step which, when the new operating conditions are within a preset tolerance range, operates the manufacturing equipment under the new operating conditions. According to this embodiment, when it is detected that the target variable has deviated from a threshold range, new operating conditions for bringing the target variable closer to the reference value can be calculated using a pre-generated model function. The new operating conditions can then be displayed on a display device to prompt the user of the manufacturing equipment to change the operating conditions, thereby enabling the manufacturing equipment to operate under the new operating conditions. Furthermore, if the new operating conditions are within a tolerance range, the manufacturing equipment can be automatically operated under the new operating conditions without receiving input from the user. This allows for correction of deviations from the baseline value of the target variable. Therefore, the manufacturing equipment can be used to produce the desired product.

[0111] (12) According to the third form of this disclosure, a data processing program is provided. The data processing program causes a computer to execute the following functions: an acquisition function that acquires a dataset containing two or more acquired data sets, each set associating an objective variable with the operational conditions, which is acquired by operating a manufacturing apparatus under predetermined operating conditions to produce a product; a structuring function that uses the dataset to generate structured data representing the causal structure between the condition variable and the objective variable by connecting a leaf node representing the condition variable, a root node representing the objective variable, and an intermediate node located between the leaf node and the root node, which represents an influence variable that affects the objective variable and represents an element different from the operational conditions, using directed edges; and a modeling function that uses the dataset to model the manufacturing apparatus by sequentially calculating an explanatory function from the leaf node to the root node, which represents an explanatory function that expresses the variable of the higher node that is the endpoint of a specific edge using the variable of the lower node that is the starting point of the specific edge, and generating a model function that includes the calculated explanatory function. This configuration allows for the acquisition of a dataset containing two or more data points that associate condition variables representing the operating conditions of the manufacturing equipment with target variables relating to the manufactured product. Using this dataset, structured data representing the causal structure between the condition variables and the target variables can be generated. Then, using the dataset and structured data, explanatory functions are calculated sequentially from the leaf nodes to the root node, expressing the variables of higher-level nodes using the variables of lower-level nodes. Finally, a model function is generated, which includes the calculated explanatory functions. This allows for the identification of the relationship between the operating conditions of the manufacturing equipment and the target variable by representing the relationship between the condition variables and the target variable using the model function. In this case, the explanatory functions are calculated sequentially from the leaf nodes upwards in the hierarchy. Therefore, influencing variables representing elements that affect the target variable can be expressed using the condition variables. In other words, various elements that affect the target variable can be quantitatively represented using the condition variables.Therefore, it is possible to identify the relationship between the operating conditions of the manufacturing equipment and the target variable without acquiring a large amount of data and preparing a matrix for each of the various factors that affect the target variable.

[0112] Not all of the components of each form of the present disclosure described above are essential, and it is possible to modify, delete, replace with other new components, or delete some of the limitations of some of the components as appropriate in order to solve some or all of the problems described above or to achieve some or all of the effects described herein. Furthermore, it is also possible to combine some or all of the technical features included in one form of the present disclosure described above with some or all of the technical features included in another form of the present disclosure described above to form an independent form of the present disclosure in order to solve some or all of the problems described above or to achieve some or all of the effects described herein.

[0113] This disclosure can also be implemented in various forms other than data processing devices, data processing methods, and data processing programs. For example, it can be implemented in the form of a manufacturing system comprising a data processing device and a manufacturing apparatus, a method for manufacturing a data processing device and a manufacturing system, a method for controlling a data processing device and a manufacturing system, a computer program that implements the control method, and a non-temporary recording medium on which the computer program is stored. [Explanation of symbols]

[0114] 1...Manufacturing system, 10...Manufacturing equipment, 50, 50a~50d...Data processing equipment, 100...Injection molding machine, 110...Bed, 120...Injection device, 121...Hopper, 122...Heating cylinder, 123...Screw, 124...Nozzle, 125...Heater, 126...Injection drive device, 130...Mold, 131...Fixed mold, 132...Movable mold, 133...Cavity, 134...Feed path, 140...Clamping device, 141...Fixed platen, 142...Movable platen, 143...Diver, 144...Mold drive device, 15 1...Sensor for injection molding machine, 152...Sensor for mold, 153...Sensor for clamping device, 160...Control device, 161...Communication device for injection molding machine, 501, 501a~501d...CPU of data processing device, 502, 502a~502d...Storage device of data processing device, 505...Communication device for data processing device, 506...Input device, 507...Display device, 511, 511b, 511c...Acquisition unit, 512, 512b, 512c...Structuring unit, 513, 513b~513d...Modeling unit, 514,514b...Condition calculation unit, 515...Function unit, 516...Display control unit, 517...Manufacturing control unit, 518...Structural connection unit, 519...Model connection unit, a...Acceleration, A...First sub-variable, AN...Upper node, AX...Axis, B...Second sub-variable, C...Third sub-variable, CV...Condition variable, D...Axis direction, E...Edge, F1...First relation value, F2...Second relation value, F3...Third relation value, F4...Fourth relation value, F5...Fifth relation value, IV...Influence variable, kA...First individual coefficient, kA1...Twelfth individual coefficient kA2...11th individual coefficient, kAB...1st reciprocal coefficient, kAT...2nd unknown coefficient, kB...2nd individual coefficient, kBC...2nd reciprocal coefficient, kC...3rd individual coefficient, kCA...3rd reciprocal coefficient, KF...Known function, kP...7th individual coefficient, kPQ...7th reciprocal coefficient, kQ1...4th individual coefficient, kQ2...8th individual coefficient, kQR1...4th reciprocal coefficient, kQR2...8th reciprocal coefficient, kR1...5th individual coefficient, kR2...9th individual coefficient, kRU...5th reciprocal coefficient, kRW...9th reciprocal coefficient, kT...1st unknown Coefficients: kU...6th individual coefficient, kUQ...6th reciprocal coefficient, kW...10th individual coefficient, kWP...10th reciprocal coefficient, LN...leaf node, M1...molding material, M2...molten material, MD1...causal estimation model, MD2...function calculation model, MD3...condition calculation model, MF...model function, MN...intermediate node, N...node, O...weight, OV...dependent variable, P...temperature of heating cylinder, PG...program, Q...nozzle temperature, R...screw position when switching from injection to holding pressure RN...Root node, SD...Structured data, SD1...First structured data, SD2...Second structured data, SD3...Composite structured data, t...Time, t1...First constant, t2...Second constant, t3...Third constant, t4...Fourth constant, t5...Fifth constant, TV...Time-dependent variable, U...Injection velocity of molten material, UN...Sub-node, v0...Initial velocity, V...Variable, W...Position of heating cylinder at the end of injection, x...Displacement, Z1...First higher variable, Z2...Second higher variable, Z3...Third higher variable,

Claims

1. A data processing device, An acquisition unit acquires a dataset that includes two or more data sets, each containing an acquisition data set that associates an objective variable related to the product with a condition variable that is an explanatory variable for the objective variable and represents the operating conditions. A structuring unit generates structured data representing the causal structure between the condition variable and the target variable by connecting, with directed edges, leaf nodes representing the condition variable, root nodes representing the target variable, and intermediate nodes located between the leaf node and the root node, which represent influencing variables as explanatory variables that are elements that affect the target variable and are different from the operating conditions. A data processing device comprising: a modeling unit that models the manufacturing apparatus by using the dataset described above, calculating an explanatory function sequentially from the leaf node to the root node, which is a function that expresses the variables of the upper node that is the endpoint of a particular edge using the variables of the lower node that is the starting point of the particular edge, and generating a model function that includes the calculated explanatory function.

2. A data processing device according to claim 1, The aforementioned manufactured product is produced in two or more quantities. The acquisition unit further acquires the time-dependent variables, which are explanatory variables, that represent the changes over time when the two or more manufactured products are manufactured. The structuring unit is a data processing device that generates the structured data, in which nodes representing the time-series variables are located below the root node.

3. A data processing device according to claim 1, The modeling unit is a data processing device that calculates the explanatory function of a predetermined order or less.

4. A data processing device according to claim 1, A data processing device in which, when the relationship between a specific subordinate node and the superior node is known, and a physical quantity representing the relationship can be calculated using a known function which is a function that represents the relationship without using the condition variable, the modeling unit generates a model function that includes a physical quantity representing the relationship instead of calculating the explanatory function for which the relationship is known.

5. A data processing device according to claim 1, A data processing device in which, when the relationship between a specific subordinate node and the superior node is known, and a physical quantity representing the relationship cannot be calculated using a known function which is a function that represents the relationship without using the condition variable, the modeling unit generates a model function that includes the known function instead of the known explanatory function for the known relationship, without calculating the explanatory function for the known relationship.

6. A data processing device according to claim 1, In the structured data, if there are two or more subnodes with respect to the upper node, the modeling unit generates a model function that includes an explanatory function representing the influence each of the two or more subnodes has on the upper node, and an explanatory function representing the influence that a predetermined number of interactions between the subnodes have on the upper node.

7. A data processing device according to claim 1, The acquisition unit acquires a first dataset and a second dataset that includes the target variable in the first dataset as the explanatory variable. The structuring unit generates first structured data using the first dataset and generates second structured data using the second dataset. The data processing device further includes a structure linking unit that generates composite structured data by linking the first structured data and the second structured data.

8. A data processing device according to claim 1, The acquisition unit acquires a first dataset and a second dataset that includes the target variable in the first dataset as the explanatory variable. The structuring unit generates first structured data using the first dataset and generates second structured data using the second dataset. The modeling unit generates a first model function using the first structured data, and generates a second model function using the second structured data. The data processing device further comprises a model linking unit that generates a composite model function by linking the first model function and the second model function.

9. A data processing device according to any one of claims 1 to 8, further, A condition calculation unit that, when it is detected that the target variable has deviated from a threshold range including a predetermined reference value, uses the model function to calculate new operating conditions to bring the target variable closer to the reference value. A functional unit that operates when the aforementioned new operating conditions are calculated, A display control unit that prompts the user of the manufacturing apparatus to change the operating conditions by displaying the new operating conditions on a display device, A data processing device comprising a manufacturing control unit that operates the manufacturing apparatus under the new operating conditions when the new operating conditions are within a preset tolerance range, and at least one of the functional units of a manufacturing control unit.

10. A data processing method, An acquisition step involves acquiring a dataset that includes two or more data sets, each set containing data obtained by operating a manufacturing apparatus under predetermined operating conditions to produce a product, wherein the data sets include data obtained by associating a target variable related to the product with a condition variable that is an explanatory variable for the target variable and represents the operating conditions. A structuring step of generating structured data representing the causal structure between the condition variable and the target variable by connecting, with directed edges, leaf nodes representing the condition variable, root nodes representing the target variable, and intermediate nodes located between the leaf node and the root node, which represent influencing variables as explanatory variables that are elements that affect the target variable and are different from the operating conditions, using the aforementioned dataset; A data processing method comprising: a modeling step of modeling the manufacturing apparatus by using the dataset described above, calculating an explanatory function sequentially from the leaf node to the root node, which is a function that expresses the variables of the upper node that is the endpoint of a particular edge using the variables of the lower node that is the starting point of the particular edge, and generating a model function that includes the calculated explanatory function.

11. A data processing method according to claim 10, further, When it is detected that the target variable has deviated from a threshold range that includes a predetermined reference value, a condition calculation step is performed to calculate new operating conditions to bring the target variable closer to the reference value using the model function. A functional process that is executed when the aforementioned new operating conditions are calculated, A display control step that prompts the user of the manufacturing apparatus to change the operating conditions by displaying the new operating conditions on a display device, A data processing method comprising: a manufacturing control step of operating the manufacturing apparatus under the new operating conditions when the new operating conditions are within a preset tolerance range; and at least one of the following functional steps.

12. A data processing program, An acquisition function that acquires a dataset containing two or more data sets, which are acquired by operating a manufacturing device under predetermined operating conditions to produce a product, and which associate an objective variable related to the product with an explanatory variable of the objective variable, which is a condition variable representing the operating conditions. A structuring function that generates structured data representing the causal structure between the condition variable and the target variable by connecting, with directed edges, leaf nodes representing the condition variable, root nodes representing the target variable, and intermediate nodes located between the leaf node and the root node, which represent influencing variables as explanatory variables that are elements that affect the target variable and are different from the operating conditions, using the aforementioned dataset. A data processing program that causes a computer to perform a modeling function, which models the manufacturing apparatus by using the aforementioned dataset to sequentially calculate an explanatory function, which is a function that expresses the variables of the higher-level node that is the endpoint of a specific edge using the variables of the lower-level node that is the starting point of the specific edge, from the leaf node to the root node, and then generating a model function, which is a function that includes the calculated explanatory function.

Images