Information processing device, information processing method, and information processing program

Abstract

An information processing device (2) comprises: a variable value prediction unit (202) that predicts the value of an explanatory variable that cannot be manipulated or is difficult to manipulate in a quality prediction model (212) that predicts the quality of a resin molded article; a quality prediction unit (203) that predicts the quality of the resin molded article from the value of the variable; and an optimization calculation unit (204) that optimizes the manufacturing condition on the basis of the prediction result.

Description

Information processing device, information processing method, and information processing program

[0001] The present invention relates to an information processing device, an information processing method, and an information processing program.

[0002] Conventionally, there are known inventions for optimizing manufacturing conditions for products manufactured by injection molding (for example, Patent Document 1). The invention described in Patent Document 1 predicts the quality of products manufactured by injection molding using a structural analysis constitutive model, which is a machine learning model, and optimizes the manufacturing conditions using the predicted quality.

[0003] Japan Special Table No. 2006-523351

[0004] In order to accurately predict product quality, not only in injection molding, but in any other field, it is necessary to ensure that the explanatory variables of the machine learning model used to predict product quality are highly correlated with quality. However, explanatory variables highly correlated with quality may include variables that are inoperable or difficult to manipulate in product manufacturing. Even if optimal values ​​for such explanatory variables are determined through optimization calculations, it is difficult or impossible to reflect the optimal values ​​in manufacturing. Therefore, there is a problem in that values ​​that are inoperable or difficult to manipulate in product manufacturing cannot be used as explanatory variables, which makes it impossible to improve the accuracy of product quality prediction, and therefore makes it impossible to improve the accuracy of optimization of manufacturing conditions. Patent Document 1 does not mention this problem at all.

[0005] An object of one aspect of the present invention is to optimize manufacturing conditions while performing quality prediction using explanatory variables that cannot be manipulated or are difficult to manipulate.

[0006] In order to solve the above problems, an information processing device according to one embodiment of the present invention includes a variable value prediction unit that predicts the values ​​of explanatory variables that cannot be manipulated or are difficult to manipulate in a quality prediction model that predicts the quality of a resin molded product from the manufacturing conditions of the resin molded product; a quality prediction unit that uses the quality prediction model to predict the quality of a resin molded product manufactured under the manufacturing conditions from the values ​​of the variables predicted by the variable value prediction unit; and an optimization calculation unit that optimizes the manufacturing conditions based on the prediction results of the quality prediction unit, wherein the variable value prediction unit predicts the values ​​of some explanatory variables selected from a plurality of explanatory variables that cannot be manipulated or are difficult to manipulate in the quality prediction model based on the degree of their contribution to the prediction result, from the manufacturing conditions.

[0007] In order to solve the above-mentioned problems, an information processing method according to one aspect of the present invention is an information processing method executed by one or more information processing devices, and includes: a variable value prediction step of predicting, from the manufacturing conditions of a resin molded product, the values ​​of explanatory variables that cannot be manipulated or are difficult to manipulate in a quality prediction model that predicts the quality of the resin molded product; a quality prediction step of predicting, using the quality prediction model, the quality of a resin molded product manufactured under the manufacturing conditions from the values ​​of the variables predicted in the variable value prediction step; and an optimization calculation step of optimizing the manufacturing conditions based on the prediction results of the quality prediction step, wherein, in the variable value prediction step, values ​​of some explanatory variables selected from a plurality of explanatory variables that cannot be manipulated or are difficult to manipulate in the quality prediction model are predicted from the manufacturing conditions.

[0008] The information processing device according to each aspect of the present invention may be realized by a computer. In this case, a control program that causes the computer to operate as each part (software element) of the information processing device to realize the information processing device on a computer, and a computer-readable recording medium on which the control program is recorded, also fall within the scope of the present invention.

[0009] According to one aspect of the present invention, it is possible to optimize manufacturing conditions while performing quality prediction using explanatory variables that cannot be manipulated or are difficult to manipulate.

[0010] FIG. 1 is a block diagram showing an example of the configuration of the main parts of an information processing device according to a first embodiment of the present invention. FIG. 2 is a diagram showing an overview of an information processing system according to a first embodiment of the present invention. FIG. 3 is a graph showing the behavior of a screw position during the manufacture of a resin molded product and the filling pressure during one shot when filling a cavity with molten resin. FIG. 4 is a schematic diagram showing an example of the flow of each piece of data, such as an operable value and an inoperable or difficult-to-operate value, according to the first embodiment of the present invention. FIG. 5 is a flowchart showing an example of a method for predicting the quality of a resin molded product and a method for optimizing manufacturing conditions, according to a second embodiment of the present invention. FIG. 6 is a schematic diagram showing an example of the flow of each piece of data, such as an operable value and an inoperable or difficult-to-operate value, according to a third embodiment of the present invention.

[0011] [First Embodiment] A first embodiment of the present invention will be described in detail below with reference to the drawings. In the description of the drawings, the same parts are given the same reference numerals and the description thereof will be omitted.

[0012] (Outline of Information Processing System) An outline of the information processing system 5 according to this embodiment will be described with reference to Fig. 2. Fig. 2 is a diagram showing an outline of the information processing system 5. Fig. 2 also shows an outline of an injection molding machine. The information processing system 5 is a system that can be used to support injection molding, and includes information processing devices 1 and 2. Below, the injection molding to be supported will be described, followed by a description of the information processing device 1 and the information processing device 2.

[0013] Injection molding uses an injection molding machine that includes an injection unit, a mold, and a mold clamping unit. In injection molding, resin pellets are first fed from a hopper into a cylinder, and the resin in the cylinder is heated by a heater to form molten resin. Next, a screw is advanced by a motor, and the molten resin is injected into the mold. The injected molten resin is cooled and solidified inside the mold. After the molten resin has solidified, the crosshead is moved to open the mold along the tie bars, and the molded resin product is removed from the mold by an ejector mechanism.

[0014] The removed resin molded product is then confirmed for quality issues before being used as a finished product. In other words, in this embodiment, the target of quality confirmation is the resin molded product, and resin molded products that meet quality standards are considered finished products. On the other hand, resin molded products that do not meet quality standards are considered defective. Figure 2 shows an example in which the resin molded product is a spoon, and its quality is determined using a visual inspection machine. The information processing system 5 can support any injection molding process. While the configuration of the injection molding machine used for the injection molding and the resin molded product to be injection-molded are not particularly limited, the information processing system 5 is particularly suitable for molding resins with low decomposition temperatures. For example, the information processing system 5 can be used effectively for molding biodegradable plastics. Specifically, the resin may be a polyhydroxyalkanoate resin, such as poly(3-hydroxyalkanoate)-based resin or poly(3-hydroxybutyrate-co-3-hydroxyhexanoate). Polyhydroxyalkanoate resins are biodegradable aliphatic polyesters.

[0015] The information processing device 1 generates a variable value prediction model and a quality prediction model through machine learning using training data, and the information processing device 2 predicts the quality of the resin molded product from the prediction data using the variable value prediction model and the quality prediction model. The information processing device 2 then uses the prediction results to optimize the manufacturing conditions for the resin molded product.

[0016] (Configuration example of information processing device 1 and information processing device 2) Fig. 1 is a block diagram showing an example of the main configuration of information processing device 1 and information processing device 2. Below, each configuration of information processing device 1 and information processing device 2 will be described in order based on Fig. 1.

[0017] 1, the information processing device 1 includes a control unit 10 that controls each unit of the information processing device 1, and a storage unit 11 that stores various data used by the information processing device 1. The information processing device 1 also includes a communication unit 12 that enables the information processing device 1 to communicate with other devices, an input unit 13 that accepts various data input to the information processing device 1, and an output unit 14 that enables the information processing device 1 to output various data. The control unit 10 also includes a training data acquisition unit 101 and a learning unit 102. The storage unit 11 stores training data sets 111 and 112.

[0018] The information processing device 1 may be configured as a general-purpose computer. When the information processing device 1 is configured as a general-purpose computer, the control unit 10 may be configured to include a processor, a read-only memory (ROM), a random access memory (RAM), etc. The processor can read programs from the ROM and execute various programs using the RAM as a working area. The storage unit 11 may be configured as a hard disk drive (HDD), a solid state drive (SSD), etc.

[0019] The input unit 13 is typically a keyboard or a mouse, but is not limited to these and may be any configuration that allows various data to be input. For example, the input unit 13 may be a microphone, a touch panel, etc. The output unit 14 is typically a display, but is not limited to these and may be any configuration that allows various data to be output. For example, the output unit 14 may be a speaker, a warning light, etc.

[0020] The training data sets 111 and 112 stored in the storage unit 11 will be described. The training data sets 111 and 112 are data sets including multiple pieces of training data (labeled data used in machine learning). Machine learning is a technique for learning features included in input data and generating a "model" that predicts results corresponding to newly input data. A variable value prediction model 211 and a quality prediction model 212, which will be described later, are trained models generated by machine learning using the training data sets 111 and 112, respectively.

[0021] As will be described in detail later, the quality prediction model 212 uses manufacturing conditions, etc. of a resin molded product as input data and predicts the quality of the resin molded product that will be manufactured when those manufacturing conditions, etc. are applied. The quality prediction results from the quality prediction model 212 are used to optimize the manufacturing conditions of the resin molded product. Here, the explanatory variables of this quality prediction model 212 include values ​​that cannot or are difficult to manipulate in order to improve the accuracy of the quality prediction. However, even if the optimal values ​​of such values ​​are determined, it is impossible or difficult to manipulate such values, and therefore it is impossible or difficult to reflect them in the manufacture of resin molded products.

[0022] Therefore, the information processing device 2 predicts the values ​​of explanatory variables that are inoperable or difficult to operate from the manufacturing conditions of the resin molded product. The prediction model used for this prediction is the variable value prediction model 211. In this way, by predicting the values ​​of explanatory variables that are inoperable or difficult to operate from the manufacturing conditions of the resin molded product, it becomes possible to optimize the manufacturing conditions while making quality predictions using such explanatory variables.

[0023] The training data set 111 includes a plurality of training data used for machine learning of the variable value prediction model 211. The training data included in the training data set 111 is obtained by associating input data input to the variable value prediction model 211 with values ​​to be output by the variable value prediction model 211 as correct answer data. This input data can also be referred to as explanatory variables, and the correct answer data as objective variables.

[0024] As described above, the predicted values ​​predicted by the variable value prediction model 211 are values ​​that are inoperable or difficult to manipulate during the manufacture of a resin molded product, and are used to predict quality by the quality prediction model 212. Furthermore, the variable value prediction model 211 predicts the values ​​that are inoperable or difficult to manipulate during the manufacture of a resin molded product, based on the manufacturing conditions of the resin molded product. Therefore, the training data included in the training dataset 111 is data in which the above-mentioned values ​​that are inoperable or difficult to manipulate when the manufacturing conditions are applied are associated as "correct answer data" with "input data" related to the manufacturing conditions of the resin molded product.

[0025] An example of the "input data" is data indicating the manufacturing conditions of a resin molded product, specifically, values ​​that can be manipulated by an operator engaged in the manufacture of the resin molded product. The values ​​that can be manipulated by the operator are not particularly limited, but may include, for example, various setting values ​​(SVs) applied during the manufacture of the resin molded product, various actual measurements taken during the manufacture of the resin molded product, or index values ​​calculated from these values. Specific examples include the cumulative number of injections, the output values ​​of heaters installed in various parts of the cylinder, the temperature near the heaters, the output values ​​of heaters installed in the flow paths leading to the cavity, the temperature of the flow paths, the ratio between the current and normal values ​​of each heater output value, the injection pressure, screw position data, and the slope of the injection pressure relative to the screw position.

[0026] To explain the gradient of injection pressure relative to screw position, during injection molding, the injection pressure rises as the screw position moves toward the mold. In the initial state, the injection pressure rises sharply due to the effects of static friction, but as the screw position moves toward the mold, the increase in injection pressure becomes gentler. This increase in injection pressure with changes in screw position is a phenomenon common to both normal operations and when the air vent is blocked. However, when the air vent is blocked, gas is less likely to escape from the mold, so the injection pressure rises earlier than normal.

[0027] In this way, the presence or absence of air vent blockage is reflected in the manner of change in injection pressure. The "input data" may also include the slope of injection pressure relative to screw position (in other words, the ratio of the amount of change in injection pressure to the amount of change in screw position), which is a parameter that indicates the manner of change in injection pressure. This makes it possible to make highly accurate predictions that reflect subtle changes such as the presence or absence of air vent blockage.

[0028] Regarding the screw position, in addition to positions that the operator can operate, such as the initial position when filling with molten resin and the position at the end of injection, there are also positions that the operator cannot operate or that are difficult to operate. The "holding pressure completion position" described below is a position that the operator cannot operate or that is difficult to operate.

[0029] Although the "input data" has been described as a value that can be manipulated by an operator, the present invention is not limited to this. For example, the "input data" may be a value that can be manipulated by a PLC (Programmable Logic Controller), which is a control device. When a PLC is used to control an injection molding machine, the PLC can also control the cumulative number of injections, the heater output value, etc. Note that, although a loop controller or a microcomputer may be used as a control device other than a PLC, the present embodiment will be described using a PLC as an example. For convenience of explanation, hereinafter, unless otherwise specified, values ​​that can be manipulated by an operator or a PLC will be simply referred to as "manipulable values."

[0030] Next, an example of "correct answer data" will be described. The correct answer data is data indicating the objective variable of the variable value prediction model 211, and the value of a predetermined variable is the correct answer data. More specifically, when certain "input data" is input to the variable value prediction model 211, the value of a predetermined variable that the variable value prediction model 211 should output is the "correct answer data" corresponding to that "input data." As described above, "input data" is a "manipulable value" such as the cumulative number of injections or the heater output value, whereas "correct answer data" is a "value that cannot be manipulated or is difficult to manipulate" by an operator or a PLC. Details of values ​​that cannot be manipulated or are difficult to manipulate by an operator or a PLC will be described later. Hereinafter, for convenience of explanation, unless otherwise specified, a value that cannot be manipulated or is difficult to manipulate by an operator or a PLC will be simply referred to as a "value that cannot be manipulated or is difficult to manipulate."

[0031] The training data is generated by associating the above-described input data with the correct answer data, and is stored as a training data set 111 in the storage unit 11. The process of generating the training data set 111 may be performed by the information processing device 1 or another device. Note that a variable value prediction model 211 is generated for each variable value to be predicted. Therefore, when predicting multiple variable values, a training data set 111 corresponding to each variable value is stored.

[0032] When generating the variable value prediction model 211, the training data acquisition unit 101 refers to the storage unit 11 and acquires the training data set 111. The training data acquisition unit 101 outputs the acquired training data set 111 to the learning unit 102.

[0033] The learning unit 102 generates a variable value prediction model 211 for predicting an inoperable or difficult-to-operate value from an operable value through machine learning using the training data set 111 acquired from the training data acquisition unit 101. The variable value prediction model 211 is a computation model that uses an operable value as an explanatory variable and an inoperable or difficult-to-operate value as a target variable. The machine learning algorithm is not particularly limited. For example, the learning unit 102 may generate the variable value prediction model 211 using a neural network, or may generate the variable value prediction model 211 using regression analysis, random forest, or the like.

[0034] Furthermore, the learning unit 102 generates a quality prediction model 212 in addition to the variable value prediction model 211. Like the variable value prediction model 211, the quality prediction model 212 is also a model generated by machine learning. However, when generating the quality prediction model 212, the training data used for machine learning (data in which input data is associated with correct answer data) is different from the training data used for machine learning of the variable value prediction model 211. Specifically, the training dataset 112 is used when generating the quality prediction model 212. The training data included in the training dataset 112 will be described below.

[0035] The training data included in the training dataset 112 is a correspondence between "input data (which can also be referred to as explanatory variables)" and "ground truth data (which can also be referred to as objective variables)." This "input data" includes various data related to the quality to be predicted. Specifically, the "input data" includes predicted values ​​(i.e., values ​​that cannot be manipulated or are difficult to manipulate) output from the variable value prediction model 211. In addition, the "input data" includes manipulated values ​​(e.g., physical property data of the raw material resin) among data indicating the manufacturing conditions of the resin molded product. Examples of physical property data of the resin include resin viscosity and molecular weight distribution.

[0036] The "correct answer data" in the training data included in the training data set 112 may be data indicating the quality of the resin molded product. The "quality of the resin molded product" can be expressed by evaluation values ​​for evaluating the quality of the product, such as the defect rate, strength, dimensions, etc.

[0037] A data set generated by associating such input data with the correct answer data serves as training data for generating the quality prediction model 212. The training data generated in this manner is stored in the storage unit 11 as a training data set 112. The process of generating the training data set 112 may be performed by the information processing device 1 or another device.

[0038] When generating the quality prediction model 212 , the training data acquisition unit 101 acquires the training data set 112 and outputs it to the learning unit 102 .

[0039] The variable value prediction model 211 and the quality prediction model 212 generated by the learning unit 102 are transmitted to the information processing device 2 and stored in the storage unit 21 of the information processing device 2. Note that the variable value prediction model 211 and the quality prediction model 212 generated by the learning unit 102 may be stored in the storage unit 11.

[0040] On the other hand, the information processing device 2 includes a control unit 20 that controls each unit of the information processing device 2 in an integrated manner, and a storage unit 21 that stores various data used by the information processing device 2. The information processing device 2 also includes a communication unit 22 that enables the information processing device 2 to communicate with other devices, an input unit 23 that accepts various data input to the information processing device 2, and an output unit 24 that enables the information processing device 2 to output various data. The control unit 20 also includes a data acquisition unit 201, a variable value prediction unit 202, a quality prediction unit 203, an optimization calculation unit 204, and an output control unit 205. The storage unit 21 stores a variable value prediction model 211 and a quality prediction model 212.

[0041] The information processing device 2 may be configured as a general-purpose computer like the information processing device 1, or may be configured as a PLC.

[0042] The data acquisition unit 201 acquires input data to be input to the variable value prediction model 211 and the quality prediction model 212. For example, when a plurality of manipulable values, such as the cumulative number of injections and the heater output value, are input data to the variable value prediction model 211, the data acquisition unit 201 acquires those values. The method for acquiring the manipulable values ​​is not particularly limited. For example, the data acquisition unit 201 may acquire data input by an operator to the input unit 23 as the manipulable values, or may acquire data set in the PLC as the manipulable values. The data acquisition unit 201 outputs the acquired manipulable values ​​to the variable value prediction unit 202. Note that the manipulable values ​​acquired by the data acquisition unit 201 can be said to be data for predicting values ​​that cannot be manipulated or are difficult to manipulate.

[0043] The variable value prediction unit 202 predicts the values ​​of inoperable or difficult-to-operate explanatory variables in a quality prediction model 212 that predicts the quality of a resin molded product, from the manufacturing conditions of the resin molded product. Specifically, the variable value prediction unit 202 predicts inoperable or difficult-to-operate values ​​from the operable values ​​acquired by the data acquisition unit 201, using a variable value prediction model 211 that is generated by learning the relationship between operable values ​​and inoperable or difficult-to-operate values. The inoperable or difficult-to-operate values ​​predicted by the variable value prediction unit 202 are output to the quality prediction unit 203. Note that any method for predicting variable values ​​is possible and is not limited to the example using the variable value prediction model 211.

[0044] Furthermore, the variable value prediction unit 202 does not need to predict all of the multiple explanatory variables that are inoperable or difficult to operate in the quality prediction model 212. In other words, the variable value prediction unit 202 may predict the values ​​of some explanatory variables selected based on the degree of contribution to the quality prediction result from the multiple explanatory variables that are inoperable or difficult to operate in the quality prediction model 212. This point will be described in detail later.

[0045] The quality prediction unit 203 predicts the quality of the resin molded product using the variable values ​​predicted by the variable value prediction unit 202. Specifically, the quality prediction unit 203 predicts the quality of the resin molded product by inputting the inoperable or difficult-to-operate values ​​predicted by the variable value prediction unit 202 and other data (e.g., physical property data of the resin) that are explanatory variables of the quality prediction model 212 to the quality prediction model 212. The quality of the resin molded product predicted by the quality prediction unit 203 is output to the optimization calculation unit 204.

[0046] The optimization calculation unit 204 optimizes the manufacturing conditions for the resin molded product based on the prediction results of the quality prediction unit 203. A well-known optimization method may be used for the optimization calculation. Examples of well-known optimization methods include grid search, random search, genetic algorithm, and Bayesian optimization. For example, when using grid search, the optimization calculation unit 204 sets multiple manufacturing conditions and causes the quality prediction unit 203 to predict the quality corresponding to each set manufacturing condition. This allows the optimization calculation unit 204 to search for manufacturing conditions that achieve the desired quality. The set manufacturing conditions include input data for the variable value prediction unit 202 and input data for the quality prediction model 212. By inputting the variable values ​​calculated using the input data for the variable value prediction unit 202 and the input data for the quality prediction model 212 into the quality prediction model 212, a prediction result of the quality when the resin molded product is manufactured under the manufacturing conditions is obtained.

[0047] The output control unit 205 outputs the manufacturing conditions optimized by the optimization calculation unit 204 to the output unit 24. The output mode is arbitrary; for example, the output control unit 205 may output the optimized manufacturing conditions in at least one of display output, audio output, and print output. Furthermore, the device that outputs the optimized manufacturing conditions may be a device external to the information processing device 2. The operator can grasp the optimized manufacturing conditions, enabling the manufacturing of products with the desired quality. Furthermore, when a PLC is used to control the injection molding machine, the optimized manufacturing conditions are automatically applied, enabling the automatic manufacturing of products with the desired quality.

[0048] Furthermore, the output control unit 205 may output information according to the prediction result by the quality prediction unit 203. For example, when the prediction result by the quality prediction unit 203 indicates a sign of deviation from the standard, the output control unit 205 may output a notification urging correction of the manufacturing conditions.

[0049] As described above, the information processing device 2 includes the variable value prediction unit 202 that predicts the values ​​of explanatory variables that are inoperable or difficult to operate in a quality prediction model 212 that predicts the quality of a resin molded product from the manufacturing conditions of the resin molded product, the quality prediction unit 203 that predicts the quality of a resin molded product manufactured under the manufacturing conditions from the values ​​of the variables predicted by the variable value prediction unit 202 using the quality prediction model 212, and the optimization calculation unit 204 that optimizes the manufacturing conditions based on the prediction results of the quality prediction unit 203. The variable value prediction unit 202 predicts, from the manufacturing conditions, the values ​​of some explanatory variables that are selected from the multiple explanatory variables that are inoperable or difficult to operate in the quality prediction model 212 based on the magnitude of their contribution to the prediction result.

[0050] (Values ​​that are inoperable or difficult for the operator or PLC to operate) Next, an example of a value that is inoperable or difficult to operate will be described with reference to Fig. 3. Fig. 3 is a graph showing the behavior of the screw position and the filling pressure during one shot when molten resin is filled into a cavity during the production of a resin molded product. The horizontal axis of Fig. 3 represents time, and the vertical axis represents the numerical values ​​of the screw position and the filling pressure.

[0051] Figure 3 shows the behavior of the screw position and filling pressure in three steps of the manufacturing process for resin molded products: "filling," "holding pressure," and "next metering." The "peak filling pressure," "packing pressure," and "position data indicating the dwelling completion position" shown in Figure 3 are examples of values ​​that cannot be manipulated or are difficult to manipulate. The "peak filling pressure," "packing pressure," and "dwelling completion position" will be explained below.

[0052] When the molten resin starts to fill the cavity, the filling pressure starts to rise and reaches a peak. The filling pressure at the peak is called the "peak filling pressure."

[0053] In the final stage of filling, the screw speed is generally reduced for fine adjustment, and the filling pressure begins to decrease after reaching its peak. The "packing pressure" shown in Figure 3 refers to the filling pressure immediately before transitioning to the next process, "holding pressure." To ensure a smooth transition between processes, it is preferable that the packing pressure be close to the set pressure value for holding pressure, but since it is affected by factors such as the properties of the raw materials, there is often some difference between the packing pressure and the holding pressure.

[0054] "Holding" refers to the process of applying pressure after the entire cavity is filled with molten resin until the gate is sealed at a constant pressure to prevent the molten resin from flowing back from the gate. "Holding completion position" refers to the position of the screw when the holding process is complete. After the holding process is complete, the metering of raw resin material to manufacture the next resin molded product begins.

[0055] The position data indicating the peak filling pressure, pack pressure, and dwell end position are important variable values ​​that have a significant impact on the quality of resin molded products, but even experienced operators find it difficult to manipulate these variable values. By predicting such unmanipulable or difficult-to-manipulate values ​​using the variable value prediction model 211 that uses manipulable values ​​as explanatory variables, it becomes possible to reflect the results of the optimization calculation in the manufacturing conditions and improve the accuracy of the quality prediction of resin molded products.

[0056] As described above, the variable value prediction unit 202 may predict values ​​of some explanatory variables selected from among multiple explanatory variables in the quality prediction model 212 that are inoperable or difficult to operate based on the magnitude of their contribution to the quality prediction result. The position data indicating the peak filling pressure, the pack pressure, and the dwelling pressure completion position correspond to the values ​​of some explanatory variables selected based on the magnitude of their contribution to the quality prediction result. By using the variable value prediction model 211 to predict only important variable values ​​that have a large contribution to the quality prediction result, it is possible to reduce the cost and time required to generate the variable value prediction model 211. Furthermore, reducing the number of variable value prediction models 211 used also contributes to improving the flexibility and adaptability of the management of the variable value prediction model 211. Note that known methods such as variable importance and permutation importance can be used to calculate the contribution of each explanatory variable to the prediction result.

[0057] (Example of Data Flow During Manufacturing) Next, an example of the flow of each piece of data, such as operable values ​​and inoperable or difficult to operate values, during the manufacturing of a resin molded product will be described with reference to Fig. 4. Fig. 4 is a schematic diagram showing an example of the flow of each piece of data, such as operable values ​​and inoperable or difficult to operate values, during the manufacturing of a resin molded product.

[0058] 4 indicate manufacturing conditions that are set when manufacturing a resin molded product, and specifically, are values ​​that can be manipulated by an operator or PLC, such as the cumulative number of injections, heater output value, etc. The setting values ​​40, 41, 42, etc. are also values ​​acquired by the data acquisition unit 201.

[0059] The set values ​​40, 41, 42, ... are input to variable value prediction models 211A, 211B, 211C, ..., 211N. N variable value prediction models are shown in FIG. 4. When there is no need to distinguish between the models, they will simply be referred to as "variable value prediction models 211." The input data input to each variable value prediction model 211 may be different. From the perspective of the time and cost required for generation and maintenance, it is preferable that the number of variable value prediction models 211 be around 5 to 10.

[0060] The variable value prediction model 211A may be a model that predicts the above-mentioned peak filling pressure. In this case, values ​​from the set values ​​40, 41, 42, ... that are necessary for predicting the peak filling pressure are input to the variable value prediction model 211A. Furthermore, the variable value prediction model 211B may be a model that predicts the above-mentioned pack pressure. In this case, values ​​from the set values ​​40, 41, 42, ... that are necessary for predicting the pack pressure are input to the variable value prediction model 211B. Furthermore, the variable value prediction model 211C may be a model that predicts position data that indicates the above-mentioned pressure-maintaining completion position. In this case, values ​​from the set values ​​40, 41, 42, ... that are necessary for predicting the position data that indicates the pressure-maintaining completion position are input to the variable value prediction model 211C.

[0061] The variable value output from the variable value prediction model 211A is referred to as PV (Process Value). １Similarly, the variable value output from the variable value prediction model 211B is expressed as PV ２ The variable value output from the variable value prediction model 211C is expressed as PV ３ The variable value output from the variable value prediction model 211N is expressed as PV Ｎ Let's say.

[0062] These highly important variable values ​​PV １ , ..., PV Ｎ The data set 301 made up of the above is part of the data making up the input data set 300 input to the quality prediction model 212. Each piece of data included in the input data set 300 is an explanatory variable of the quality prediction model 212.

[0063] The actual measurement data 50 is a variable value of low importance among values ​​that cannot be manipulated or are difficult to manipulate. Variable values ​​of low importance have a small contribution to the quality prediction result, so for such variable values, the actual measurement data 50 is used instead of predicting using the variable value prediction model 211. The actual measurement data 50 may be, for example, the current value of the variable of low importance. However, without being limited to this, for variable values ​​of low importance, an average value obtained from past measurements may be used instead of the current value, or an estimated value may be used. Variable value PV of low importance ｍ１ , ..., PV ｍＮ is part of the data that makes up the input dataset 300 of the quality prediction model 212.

[0064] The physical property data 51 is data relating to the physical properties of the resin, which is the raw material, and is a value that does not change or cannot be changed. １ , . . . is part of the data that constitutes the input data set 300 of the quality prediction model 212.

[0065] As described above, the input data set 300 of the quality prediction model 212 includes three data sets: data sets 301, 302, and 303. The quality prediction unit 203 inputs the input data set 300 to the quality prediction model 212 and predicts a predicted value 213 of the quality of the resin molded product. As an example, the predicted quality value 213 is assumed to be the defect rate of the resin molded product.

[0066] The optimization calculation unit 204 optimizes the manufacturing conditions so as to minimize the defect rate indicated by the quality prediction value 213. For example, when grid search is used as the optimization method, the optimization calculation unit 204 updates each parameter of the setting values ​​40, 41, 42, ..., which are manufacturing conditions, as indicated by the arrows pointing from the quality prediction value 213 to the setting values ​​40, 41, 42, ..., and verifies all parameter combinations of the setting values ​​40, 41, 42, .... This allows the optimization calculation unit 204 to search for manufacturing conditions that minimize the defect rate.

[0067] If the predicted quality value 213 indicates the strength of the resin molded product, the optimization calculation unit 204 may optimize the manufacturing conditions so that the strength is maximized. If the predicted quality value 213 indicates the dimensions of the resin molded product, the optimization calculation unit 204 may optimize the manufacturing conditions so that the dimensions fall within a target range.

[0068] Although the input dataset 300 of the quality prediction model 212 has been described as including three datasets, datasets 301, 302, and 303, this is not limiting. For example, the input dataset 300 of the quality prediction model 212 may be composed of only dataset 301. The fewer the number of variables input to the model, the more likely it is to reduce the amount of required data. Furthermore, the input dataset 300 of the quality prediction model 212 may be composed of only datasets 301 and 302, or may be composed of only datasets 301 and 303.

[0069] (Example of Quality Prediction Method and Optimization Method) Next, an example of a method for predicting the quality of a resin molded product and a method for optimizing manufacturing conditions executed by the information processing device 2 will be described with reference to Fig. 5. Fig. 5 is a flowchart showing an example of a method for predicting the quality of a resin molded product and a method for optimizing manufacturing conditions. Note that, although an example of searching for optimal manufacturing conditions while updating the manufacturing conditions will be described below, it is also possible to predict quality for each of a plurality of manufacturing conditions and detect optimal manufacturing conditions from the prediction results.

[0070] In step S101, the data acquisition unit 201 acquires prediction data. The prediction data is data indicating manufacturing conditions applied when manufacturing a resin molded product, and specifically, values ​​that can be manipulated by an operator or a PLC, such as the cumulative number of injections and the heater output value. A portion of the prediction data is used as input data for a variable value prediction model 211, and the predicted values ​​output by the variable value prediction model 211 and another portion of the prediction data are used as input data for a quality prediction model 212. Referring to FIG. 4 , the prediction data used as input data for the variable value prediction model 211 are set values ​​40, 41, 42, ..., and the prediction data used as input data for the quality prediction model 212 are actual measurement data 50 and physical property data 51.

[0071] The processing proceeds to step S102, where the variable value prediction unit 202 inputs the prediction data acquired in the processing of step S101 into the variable value prediction models 211A, 211B, 211C, ..., 211N to predict values ​​that are highly important and cannot be operated or are difficult to operate, such as filling peak pressure, pack pressure, and position data indicating the pressure retention completion position.

[0072] The processing proceeds to step S103, and the quality prediction unit 203 inputs into the quality prediction model 212 the values ​​of high importance that are inoperable or difficult to operate predicted in the processing of step S102, and the prediction data obtained in the processing of step S101 that are used as input data for the quality prediction model 212, to predict the quality of the resin molded product.

[0073] The process proceeds to step S104, where it is determined whether the quality of the resin molded product predicted in the process of step S103 satisfies the termination condition. If it is determined that the quality of the resin molded product satisfies the termination condition (YES in step S104), the process proceeds to step S106. On the other hand, if it is determined that the quality of the resin molded product does not satisfy the termination condition (NO in step S104), the process proceeds to step S105. The termination condition can be set arbitrarily. For example, the termination condition may be set to be that the defect rate of the resin molded product is equal to or less than an allowable upper limit.

[0074] In step S105, the optimization calculation unit 204 updates the manufacturing conditions (e.g., set values ​​40, 41, 42, ...), and then returns to step S101. The method for updating the manufacturing conditions may be determined in advance. The processes of steps S101 to S105 are repeatedly executed until the termination condition is satisfied (YES is determined in step S104).

[0075] In step S106, the output control unit 205 outputs the manufacturing conditions optimized by repeating the processes of steps S101 to S105 to the output unit 24. This allows the operator to understand the optimized manufacturing conditions, making it possible to manufacture products of the desired quality. Furthermore, when a PLC is used to control the injection molding machine, the optimized manufacturing conditions are automatically applied, making it possible to automatically manufacture products of the desired quality. When a PLC is used to control the injection molding machine, the process of step S106 may be omitted. This ends the process of FIG. 5.

[0076] As described above, the information processing method according to this embodiment includes a variable value prediction step (S102) for predicting values ​​of explanatory variables that are inoperable or difficult to manipulate in a quality prediction model 212 for predicting the quality of a resin molded product based on manufacturing conditions for the resin molded product; a quality prediction step (S103) for predicting the quality of a resin molded product manufactured under the manufacturing conditions based on the values ​​of the variables predicted in the variable value prediction step using the quality prediction model 212; and an optimization calculation step (S101 to S105) for optimizing the manufacturing conditions based on the prediction results of the quality prediction step. The variable value prediction step predicts values ​​of some explanatory variables selected from among multiple explanatory variables that are inoperable or difficult to manipulate in the quality prediction model 212 based on the magnitude of their contribution to the prediction results. Note that the processing flow in the flowchart shown in FIG. 5 is merely an example, and steps may be deleted or new steps may be added without departing from the spirit of the method.

[0077] (Operational Effects) As described above, according to the first embodiment, the following operational effects can be obtained.

[0078] In this embodiment, the values ​​of explanatory variables that cannot be manipulated or are difficult to manipulate in the quality prediction model 212 that predicts the quality of a resin molded product are predicted from the manufacturing conditions of the resin molded product. Furthermore, the quality prediction model 212 is used to predict the quality of a resin molded product manufactured under the manufacturing conditions from the values ​​of the variables predicted by the variable value prediction unit 202. Furthermore, the manufacturing conditions are optimized based on the prediction results of the quality prediction unit 203.

[0079] The accuracy of quality prediction can be improved by using explanatory variables that cannot be manipulated or are difficult to manipulate. However, when optimizing manufacturing conditions using such explanatory variables, optimal values ​​of explanatory variables that cannot be manipulated or are difficult to manipulate are calculated, making it impossible or difficult to reflect the optimal values ​​in manufacturing.

[0080] In this regard, according to the above configuration, the values ​​of explanatory variables that cannot be operated or are difficult to operate are predicted from the manufacturing conditions, so it is possible to optimize the manufacturing conditions while making quality predictions using explanatory variables that cannot be operated or are difficult to operate.

[0081] Furthermore, in this embodiment, the values ​​of some explanatory variables selected from among the multiple explanatory variables in the quality prediction model 212 that cannot be manipulated or are difficult to manipulate based on the degree of their contribution to the prediction result are predicted from the manufacturing conditions.

[0082] According to the above configuration, the explanatory variables to be predicted are narrowed down based on the degree of their contribution to the prediction result, so that the amount of calculation required for the optimization can be reduced while maintaining the accuracy of the optimization. Furthermore, by using the model to predict only the variable values ​​with high importance, it is possible to reduce the cost and time required to generate the model.

[0083] Furthermore, in this embodiment, a two-stage prediction process is adopted in which output from the variable value prediction model 211 is input to the quality prediction model 212. In machine learning, a large number of variables are required to generate a computational model that performs quality prediction, as in this embodiment. However, by adopting such a two-stage prediction process, fewer variables are required for each model, contributing to a reduction in the amount of required data. Furthermore, generating models with fewer variables also contributes to improved flexibility and adaptability in model management.

[0084] Furthermore, since the quality prediction model 212 includes, as explanatory variables, values ​​that cannot be manipulated or are difficult to manipulate, which are predicted by the variable value prediction model 211, a so-called cascade effect occurs, improving the accuracy of quality prediction.

[0085] The resin molded product may also be a molded product containing a biodegradable resin. Generally, biodegradable resins tend to have large variations in the viscosity of pellets, and when such pellets are used to manufacture a resin molded product, the optimal manufacturing conditions are likely to fluctuate. In this regard, according to the above configuration, the values ​​of explanatory variables that reflect the variations in the viscosity of the pellets, etc., are predicted by the variable value prediction model 211, and the manufacturing conditions can be optimized using the prediction results, thereby enabling efficient production of biodegradable resin molded products.

[0086] The biodegradable resin may also be a polyhydroxyalkanoate-based resin. Polyhydroxyalkanoate-based resins have excellent degradability, and therefore can solve environmental problems caused by discarded plastics. Particularly preferably, the polyhydroxyalkanoate-based resin is a poly(3-hydroxyalkanoate)-based resin (hereinafter, sometimes referred to as a P3HA-based resin). More specifically, the P3HA-based resin preferably contains 3-hydroxybutyrate (3HB) units. The P3HA-based resin containing 3HB units is preferably selected from the group consisting of poly(3-hydroxybutyrate) (P3HB), poly(3-hydroxybutyrate-co-3-hydroxyvalerate) (P3HB3HV), poly(3-hydroxybutyrate-co-3-hydroxyhexanoate) (P3HB3HH), poly(3-hydroxybutyrate-co-3-hydroxyvalerate-co-3-hydroxyhexanoate) (P3HB3HV3HH), poly(3-hydroxybutyrate-co-4-hydroxybutyrate) (P3HB4HB), poly(3-hydroxybutyrate-co-3-hydroxyoctanoate), and poly(3-hydroxybutyrate-co-3-hydroxydecanoate). The P3HA-based resin may contain only one type, or two or more types.

[0087] Furthermore, the operator can also simulate the production of resin molded products by inputting various manufacturing conditions and having the variable value prediction unit 202 and the quality prediction unit 203 make predictions. In this way, the information processing device 2 can also be used as an auxiliary tool for determining the manufacturing conditions of resin molded products.

[0088] Although the present embodiment has been described with reference to injection molding as a method for processing resin, the present invention is not limited to this and can be applied to various methods for processing resin, such as extrusion molding, vacuum molding, and pressure molding.

[0089] Second Embodiment Next, a second embodiment of the present invention will be described with reference to Figures 6 and 7. For ease of explanation, the same reference numerals will be used to designate the same components as those described in the first embodiment, and the description thereof will not be repeated.

[0090] 6 is a schematic diagram showing an example of the flow of data such as operable values ​​and inoperable or difficult-to-operate values ​​during the manufacture of a resin molded product. The second embodiment differs from the first embodiment in that quality prediction using the quality prediction model 212 is performed using multiple input data sets, rather than just one input data set 300. Specifically, in the second embodiment, input data sets 310, 320, ... are used in addition to the input data set 300.

[0091] Generally, predicted values ​​obtained by machine learning models contain errors. Therefore, it can be said that the predicted values ​​of the variable value prediction models 211A, 211B, 211C, ..., 211N each contain errors, and the prediction results of the quality prediction model 212 using these predicted values ​​reflect the errors in each predicted value.

[0092] Therefore, when a first-stage prediction is made using a plurality of variable value prediction models 211, the errors in each first-stage prediction are accumulated in the second-stage prediction using the quality prediction model 212, resulting in an invalid prediction result, which raises the concern that it may not be possible to derive valid manufacturing conditions.

[0093] Therefore, in the second embodiment, the error range of each predicted value by the variable value prediction models 211A, 211B, 211C, ..., 211N is taken into consideration. The optimization calculation unit 204 optimizes the manufacturing conditions based on each prediction result obtained by inputting into the quality prediction model 212 the predicted value predicted by the variable value prediction unit 202 and the upper and lower end values ​​calculated from the predicted value based on the prediction error range of the variable value prediction models 211A, 211B, 211C, ..., 211N.

[0094] Input datasets 310 and 320 are input datasets used to prevent errors from being accumulated. Input dataset 310 includes dataset 311 instead of dataset 301 in input dataset 300. Similarly to input dataset 300, input dataset 310 also includes datasets 302 and 303.

[0095] Here, the predicted value output by the variable value prediction model 211A is PV１ Let the error range of this predicted value be ±e. In this case, PV １ The true value of is (PV １ -e) to (PV １ + e). In other words, the optimization calculation unit 204 uses the value of e to calculate the PV １ The upper limit of (PV １ +e), the lower limit value (PV １ The prediction error value e can be calculated from each predicted value obtained by inputting a part of the training data set 111 as test data into the variable value prediction model 211A.

[0096] If there are N variable value prediction models, for each of the N predicted values, there are three patterns: one in which the predicted value is used as is, one in which the upper end value of the predicted value is used, and one in which the lower end value of the predicted value is used. ｎ +1 input data set can be created. ｎ +1 possible quality prediction results can be output. The optimization calculation unit 204 then combines these prediction results to calculate the quality prediction value 213. For example, the optimization calculation unit 204 may calculate a representative value of each prediction result (e.g., an average value or a worst value) as the quality prediction value 213. By performing verification calculations in this manner that take into account the upper and lower end values ​​that are the two ends of the error range, it is possible to avoid a situation in which errors in each prediction in the first stage by the variable value prediction model are accumulated, resulting in the result of the second stage quality prediction by the quality prediction model 212 lacking validity.

[0097] Here, a supplementary explanation will be given of the prediction error of the quality prediction model 212 with reference to FIG. 7. FIG. 7 is a diagram for explaining the prediction error of the quality prediction model 212. In FIG. 7, the quality prediction model 212 that predicts the product defect rate is shown as a curved graph Y=f(X). The X axis of the figure indicates the value of the explanatory variable, and the Y axis indicates the defect rate. A point 62 on the graph indicates the point where the value of the explanatory variable is x １ This shows the defect rate when

[0098] This explanatory variable x １ is predicted by the variable value prediction model 211 and therefore contains an error. If this error is ±e, the defect rate predicted by the quality prediction model 212 is １ -e), the value of the explanatory variable (x １ +e). Specifically, the defect rate predicted by the quality prediction model 212 is minimum at point 62 and maximum at point 61. By using the predicted values ​​of the defect rates at these three points, x １ Even if the predicted value is not valid, the predicted result of the defect rate by the quality prediction model 212 can be set to a value within a valid range.

[0099] In reality, there are a plurality of explanatory variables predicted by the variable value prediction model 211. For example, assume that the number of explanatory variables is n. In this case, the quality prediction unit 203 predicts x predicted by the variable value prediction model 211. １ From x ｎ For each of the n explanatory variables up to , the quality prediction unit 203 inputs the predicted value of the explanatory variable into the quality prediction model 212 to calculate a predicted value. Furthermore, for each of the n explanatory variables, the quality prediction unit 203 inputs the upper end value of the error range of the explanatory variable into the quality prediction model 212 to calculate a predicted value, and for each of the n explanatory variables, inputs the lower end value of the error range of the explanatory variable into the quality prediction model 212 to calculate a predicted value. ｎ +1 predicted value is obtained. ｎ The manufacturing conditions can be optimized using the +1 predicted value.

[0100] [Modifications] The entity that executes each process described in each of the above-described embodiments is arbitrary and is not limited to the above-described examples. In other words, functions similar to those of information processing devices 1 and 2 can be realized by multiple information processing devices that can communicate with each other. For example, the processing of each step shown in FIG. 5 may be shared and executed by multiple information processing devices. Furthermore, an information processing device that combines the functions of information processing device 1 and information processing device 2 is also included in the scope of the present invention.

[0101] [Example of implementation by software] The functions of the information processing devices 1 and 2 (hereinafter simply referred to as "devices") can be realized by a program (information processing program) for causing a computer to function as the device, and a program for causing a computer to function as each control block of the device (particularly each part included in the control unit 10 and the control unit 20).

[0102] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program, and the computer executes the program to realize each function described in each embodiment.

[0103] The program may be stored non-transitory on one or more computer-readable storage media. The storage media may or may not be included in the device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

[0104] In addition, some or all of the functions of each control block can be realized by a logic circuit. For example, an integrated circuit in which a logic circuit that functions as each control block is formed is also included in the scope of the present invention. In addition, the functions of each control block can also be realized by, for example, a quantum computer.

[0105] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention.

[0106] 1, 2 Information processing device 202 Variable value prediction unit 203 Quality prediction unit 204 Optimization calculation unit 211 Variable value prediction model 212 Quality prediction model

Claims

1. An information processing device comprising: a variable value prediction unit that predicts the values ​​of explanatory variables that cannot be manipulated or are difficult to manipulate in a quality prediction model that predicts the quality of a resin molded product from the manufacturing conditions of the resin molded product; a quality prediction unit that uses the quality prediction model to predict the quality of a resin molded product manufactured under the manufacturing conditions from the values ​​of the variables predicted by the variable value prediction unit; and an optimization calculation unit that optimizes the manufacturing conditions based on the prediction results of the quality prediction unit, wherein the variable value prediction unit predicts the values ​​of some explanatory variables selected from a plurality of explanatory variables that cannot be manipulated or are difficult to manipulate in the quality prediction model based on the degree of their contribution to the prediction results.

2. The information processing device of claim 1, wherein the variable value prediction unit predicts values ​​of some of the explanatory variables in the quality prediction model from the manufacturing conditions using the variable value prediction model, and the optimization calculation unit optimizes the manufacturing conditions based on each prediction result obtained by inputting into the quality prediction model the predicted value predicted by the variable value prediction unit and an upper end value and a lower end value calculated from the predicted value based on the range of prediction error of the variable value prediction model.

3. The information processing device according to claim 1 or 2, wherein the resin molded product is a molded product containing a biodegradable resin.

4. The information processing device according to claim 3, wherein the biodegradable resin is a polyhydroxyalkanoate resin.

5. An information processing method executed by one or more information processing devices, comprising: a variable value prediction step of predicting the values ​​of explanatory variables that cannot be manipulated or are difficult to manipulate in a quality prediction model that predicts the quality of a resin molded product from the manufacturing conditions of the resin molded product; a quality prediction step of predicting the quality of a resin molded product manufactured under the manufacturing conditions from the values ​​of variables predicted in the variable value prediction step using the quality prediction model; and an optimization calculation step of optimizing the manufacturing conditions based on the prediction results of the quality prediction step, wherein in the variable value prediction step, values ​​of some explanatory variables selected from a plurality of explanatory variables that cannot be manipulated or are difficult to manipulate in the quality prediction model are predicted from the manufacturing conditions.

6. An information processing program for causing a computer to function as the information processing device according to claim 1, the information processing program causing a computer to function as the variable value prediction unit, the quality prediction unit, and the optimization calculation unit.

Images