Information processing methods, information processing devices, manufacturing systems, methods for manufacturing goods, programs, recording media

Abstract

Even in fields where it is difficult to determine the target variable with high accuracy solely through simulation, there has been a need for technology that can efficiently acquire data for machine learning using simulation and generate trained models with high predictive accuracy. [Solution] Actual process data measured when the manufacturing equipment manufactures a product under first manufacturing conditions, actual quality data measuring the quality of the product, and first virtual process data generated by simulating the manufacturing process in which the manufacturing equipment manufactures the product are acquired. A process data prediction unit and a quality prediction unit are generated by machine learning. Second virtual process data is acquired by simulating the manufacturing process in which the manufacturing equipment manufactures a product under second manufacturing conditions. The second virtual process data is input to the process data prediction unit to output virtual process data for educational purposes. The second virtual process data is input to the quality prediction unit to output virtual quality data for educational purposes. A product quality prediction model is generated by machine learning.

Description

Technical Field

[0001] The present invention relates to an information processing method, an information processing apparatus, etc. for predicting the quality of a product using process data in a manufacturing process.

Background Art

[0002] In the field of product manufacturing, quality control is important to prevent defective products from being shipped. The method of inspecting the manufactured products to eliminate defective products is effective, but there are concerns such as that it requires a great deal of load if all products are to be inspected, and that it takes time to feedback the inspection results to the manufacturing process and there is a possibility that a large number of defective products will occur. Therefore, techniques for predicting the quality of a product using process data in a manufacturing process have been studied.

[0003] On the other hand, in recent years, learned models generated using machine learning have been used for predicting events in various fields, and their use has also been studied in the field of product manufacturing. In order to generate a learned model with high prediction accuracy, it is necessary to prepare a large amount of learning data and let a computer perform machine learning.

[0004] In order to generate a learned model for predicting the quality of a product using process data in a manufacturing process, it is conceivable to prototype and experiment with products under various manufacturing conditions to collect learning data. For that, it is necessary to collect process data for each of various manufacturing conditions, measure the quality of products for each manufacturing condition, and prepare a teacher data group with process data as explanatory variables and quality as target variables. However, prototyping products and collecting learning data under various manufacturing conditions requires a great deal of labor, time, and cost, and thus may be difficult to actually implement.

[0005] Patent Document 1 describes a method for creating multiple trained models using machine learning with experimental and simulation data, evaluating the prediction accuracy of the multiple trained models using validation experimental and simulation data, and selecting a model for prediction. Rather than preparing all the training data by actually conducting experiments, the method aims to efficiently generate multiple trained models by creating training data using simulations in conjunction with experiments.

[0006] Generally, simulation data contains errors, so Patent Document 1 creates a modified simulation dataset by correcting the simulation data. Then, an integrated dataset, which combines the modified simulation dataset and the experimental dataset, is used as training data for machine learning to create multiple trained models that can serve as candidates for predictive models. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2021-22276 [Overview of the Initiative] [Problems that the invention aims to solve]

[0008] Patent Document 1 describes storing sets of values ​​for multiple explanatory variables and the values ​​of feature variables (dependent variables) that associate with the values ​​of the explanatory variables. Then, for the feature variables calculated by simulation, the values ​​of the feature variables are modified based on their correspondence with experimental data, and a modified simulation dataset is created. Patent Document 1 describes that when modifying the values ​​of feature variables calculated by simulation, the focus should be on whether they lie between the maximum and minimum values ​​of the feature variables in multiple experimental data, or whether the values ​​of the explanatory variables match within an acceptable range.

[0009] For example, in the field of injection molding using molds, the surface quality and shape accuracy of the manufactured product are important aspects of quality. However, current simulation technology makes it difficult to quantitatively calculate certain types of quality. For instance, the causal relationship between the phenomenon known as "shrinkage," which reduces surface quality, and specific process conditions has not been fully elucidated, making it difficult to calculate "shrinkage" as a feature (dependent variable) in simulations. Furthermore, "warpage" and "shrinkage," which reduce the shape accuracy of the product, occur due to the combined action of various process conditions in injection molding. Therefore, while simulations can grasp the relative trends of "warpage" and "shrinkage," it is difficult to calculate their absolute values ​​with high accuracy. Consequently, even if simulation results are modified using methods such as those described in Patent Document 1 for types of quality that are difficult to quantitatively calculate, it has been difficult to create appropriate modified simulation datasets for machine learning.

[0010] Therefore, even in fields where it is difficult to accurately determine the feature variables as target variables through simulation alone, there was a need for a technology that could efficiently acquire data for machine learning using simulation and generate trained models with high predictive accuracy. [Means for solving the problem]

[0011] A first aspect of the present invention involves an information processing unit acquiring actual process data measured when a manufacturing apparatus manufactures a product under first manufacturing conditions, actual quality data measuring the quality of the product manufactured under the first manufacturing conditions, and first virtual process data generated by simulating the manufacturing process in which the manufacturing apparatus manufactures the product under the first manufacturing conditions. The information processing unit then generates a process data prediction unit using machine learning with the first virtual process data as explanatory variables and the actual process data as the target variable as training data. The information processing unit then generates a quality prediction unit using machine learning with the first virtual process data as explanatory variables and the actual quality data as the target variable as training data. This information processing method is characterized by generating a first manufacturing condition, simulating a manufacturing process in which the manufacturing apparatus manufactures a product under second manufacturing conditions different from the first manufacturing conditions to acquire second virtual process data, inputting the second virtual process data into the process data prediction unit to output virtual process data for educational purposes, inputting the second virtual process data into the quality prediction unit to output virtual quality data for educational purposes, and generating a product quality prediction model by machine learning using training data in which the measured process data and the virtual process data for educational purposes are used as explanatory variables and the measured quality data and the virtual quality data for educational purposes are used as target variables.

[0012] Furthermore, in a second aspect of the present invention, the information processing unit acquires actual process data measured when the manufacturing apparatus manufactures a product under first manufacturing conditions, actual quality data measuring the quality of the product manufactured under the first manufacturing conditions, and first virtual process data generated by simulating the manufacturing process in which the manufacturing apparatus manufactures the product under the first manufacturing conditions. The information processing unit then generates a process data prediction unit by machine learning using the first virtual process data as explanatory variables and the actual process data as the target variable as training data. The quality prediction unit then generates a quality prediction unit by machine learning using the first virtual process data as explanatory variables and the actual quality data as the target variable as training data. This information processing device is characterized by generating a measurement unit, simulating a manufacturing process in which the manufacturing apparatus manufactures a product under second manufacturing conditions different from the first manufacturing conditions to acquire second virtual process data, inputting the second virtual process data into the process data prediction unit to output virtual process data for educational purposes, inputting the second virtual process data into the quality prediction unit to output virtual quality data for educational purposes, and generating a product quality prediction model by machine learning using training data in which the measured process data and the virtual process data for educational purposes are used as explanatory variables and the measured quality data and the virtual quality data for educational purposes are used as objective variables. [Effects of the Invention]

[0013] According to the present invention, even in fields where it is difficult to accurately determine feature quantities as target variables solely through simulation, it is possible to efficiently acquire data for machine learning by utilizing simulation and generate a trained model with high predictive accuracy. [Brief explanation of the drawing]

[0014] [Figure 1] A diagram showing the configuration of an information processing system equipped with an information processing device according to the embodiment. [Figure 2] A schematic block diagram illustrating the configuration of an information processing device and its peripheral equipment. [Figure 3] A schematic block diagram illustrating the configuration of manufacturing equipment and quality evaluation equipment connected to an information processing device. [Figure 4] Schematic functional block diagram for explaining the functions of an information processing apparatus. [Figure 5] Schematic functional block diagram for explaining the configuration of an actual measurement data learning unit. [Figure 6] Schematic functional block diagram for explaining the configuration of a quality result learning unit. [Figure 7] Schematic functional block diagram for explaining the configuration of a product quality prediction unit. [Figure 8] Flowchart for explaining the information processing procedure for generating a process data prediction unit and a quality prediction unit. [Figure 9] Flowchart for explaining the information processing procedure for generating a product quality prediction unit. [Figure 10] Flowchart for explaining the information processing procedure for predicting the quality of a product being manufactured by an information processing apparatus when a manufacturing device manufactures the product. [Figure 11] (a) Perspective view showing the outer shape of a resin molded product as a product. (b) Schematic diagram for explaining the arrangement of components of an injection molding apparatus. [Figure 12] Diagram for explaining an example of process data.

Mode for Carrying Out the Invention

[0015] Referring to the drawings, an information processing apparatus, an information processing method, etc. according to an embodiment of the present invention will be described. The following embodiments are examples, and for example, those skilled in the art can appropriately change and implement the detailed configuration without departing from the gist of the present invention.

[0016] In the drawings referred to in the following description of the embodiments and examples, unless otherwise specified, elements denoted by the same reference numerals have the same functions. In the drawings, when a plurality of the same elements are arranged, the assignment of reference numerals and their descriptions may be omitted.

[0017] In addition, for the sake of convenience in illustration and description, the drawings may be presented schematically. Therefore, the shapes, sizes, arrangements, etc. of the elements shown in the drawings may not necessarily be exactly the same as those of the actual objects.

[0018] [Embodiment] (Information Processing System) FIG. 1 shows the configuration of an information processing system including an information processing apparatus according to an embodiment. The information processing system includes a manufacturing device 100 that manufactures products, a quality evaluation device 200 that evaluates the quality of products, and an information processing apparatus 300. The manufacturing device 100 is a device that manufactures products. The types of products to be manufactured and the types of manufacturing devices are not particularly limited. The quality evaluation device 200 is a device that evaluates the quality of products manufactured by the manufacturing device 100, and is a device that can measure evaluation items related to quality by measurement means such as sensors and convert them into numerical data. Specific evaluation items and measurement means are not particularly limited. The information processing apparatus 300 is an information processing apparatus (computer) that handles information related to the manufacture of products. Note that the illustrated system can be said to be an information processing system in which the information processing apparatus 300 handles information related to the manufacture of products, and at the same time, a manufacturing system in which the manufacturing device 100 manufactures products.

[0019] The information processing apparatus 300 is communicably connected to the manufacturing device 100 and the quality evaluation device 200 via wired or wireless communication means. The information processing apparatus 300 can receive information (data) related to process parameters, for example, when manufacturing products, from the manufacturing device 100. In addition, the information processing apparatus 300 can receive information (data) related to the quality evaluation results of products manufactured by the manufacturing device 100, for example, from the quality evaluation device 200.

[0020] FIG. 2 is a schematic block diagram illustrating the configuration of the information processing apparatus 300 and its peripheral devices. The information processing apparatus 300 is a computer for performing information processing related to the manufacture of articles (for example, predicting the quality of manufactured products).

[0021] The information processing device 300 includes a CPU 301 (Central Processing Unit) as a processor, a memory 302 and an HDD 303 as storage units, and an I / O port 305 as an input / output interface. These elements are connected to each other via a system bus 304. The elements shown are functional concepts and do not necessarily have to be physically configured as shown. For example, the specific forms of distribution and integration of each block are not limited to the examples shown, and all or part of them can be functionally or physically distributed and integrated in any unit depending on the usage situation, etc.

[0022] Memory 302 includes ROM (Read Only Memory) and RAM (Random Access Memory), while HDD 303 includes an HDD (Hard Disk Drive). The ROM in Memory 302 stores basic programs such as the BIOS. RAM is a storage device that temporarily stores various data, such as the results of calculations performed by the CPU 301. HDD 303 is a computer-readable, non-temporary storage medium that stores information processing programs executed by the CPU 301, the results of calculations performed by the CPU 301, and various data acquired from external sources. HDD 303 may contain programs that cause the CPU 301 to perform information processing related to product quality prediction. Based on the programs recorded (stored) in HDD 303, the CPU 301 performs information processing related to the creation of quality prediction models, product quality prediction, and product manufacturing management. These information processing programs may be stored in HDD 303 along with other control programs, or they may be loaded into RAM from an external source via a network. Alternatively, they may be loaded into RAM via external memory readable by the computer on which the programs are stored.

[0023] The information processing device 300 may be connected to an input device 306 and a display 307 as a user interface, or to an external memory 308 as an external storage device, via an I / O port 305.

[0024] The input device 306 can consist of, for example, a terminal such as a handheld terminal, or a device such as a keyboard, jog dial, mouse, pointing device, or voice input device (or a control terminal equipped with these). The display 307 can be any device capable of displaying information related to information processing on a display screen, and for example, a liquid crystal display or an organic EL display can be used. The external memory 308 can be, for example, a rewritable non-volatile memory or an external storage device such as an external HDD.

[0025] The information processing device 300 is connected to the manufacturing equipment 100 and the quality evaluation equipment 200 via I / O port 305 in a communicative manner. Figure 3 is a schematic block diagram illustrating a partial configuration of the manufacturing equipment 100 and the quality evaluation equipment 200 connected to the information processing device 300. Each functional element shown is a functional concept and does not necessarily have to be physically configured as shown. For example, the specific forms of distribution and integration of each functional block are not limited to the illustrated example, and all or part of them can be functionally or physically distributed and integrated in any unit depending on the usage situation, etc. These functional blocks can be configured, for example, by the CPU reading and executing an information processing program stored in a memory device or a non-temporary recording medium. Alternatively, some or all of the functional blocks may be configured by hardware such as an ASIC.

[0026] First, let's describe the configuration of the manufacturing equipment 100. The manufacturing equipment 100 includes a product manufacturing control unit 101, a process data acquisition unit 102, a process data output unit 103, a quality information receiving unit 104, and a defective product sorting unit 105.

[0027] The product manufacturing control unit 101 is a computer that controls the operation of the manufacturing equipment 100, and when manufacturing a product, it controls the operation of the manufacturing mechanism (not shown in the illustration, such as an injection molding mechanism, heating mechanism, painting mechanism, cutting mechanism, polishing mechanism, assembly robot, etc., with no particular limitations).

[0028] The process data acquisition unit 102 has the function of measuring and acquiring various process data (such as waveforms and numerical values ​​of temperature, pressure, processing speed, light intensity, sound, etc., without any particular restrictions) when manufacturing a product, for example, using various sensors. In addition to measurement data acquired from various sensors, the process data may also include instruction data used to control the manufacturing mechanism. Preferably, the process data is acquired in association with the individual product being manufactured, but for example, the average value of a manufacturing lot corresponding to a certain number of products may be acquired in association with the manufacturing lot. The process data output unit 103 has the function of outputting the process data acquired by the process data acquisition unit 102 to the information processing device 300 in a timely manner.

[0029] As described later, when the manufacturing machine 100 manufactures a product, the information processing device 300 predicts (determines) whether the quality of the product corresponding to the process data is poor, based on the process data output by the process data output unit 103. If the information processing device 300 predicts (determines) that the product quality is poor, it transmits quality information (for example, a defective signal indicating that the product is defective) to the manufacturing machine 100. When the quality information receiving unit 104 of the manufacturing machine 100 receives the defective signal, the defective product sorting unit 105 sorts out and discharges the products that have been determined to be defective.

[0030] Next, the configuration of the quality evaluation device 200 will be described. The quality evaluation device 200 includes a quality evaluation control unit 201 and a quality evaluation data output unit 202. The quality evaluation control unit 201 is a computer that controls the quality evaluation device 200 and controls evaluation mechanisms (not shown) (e.g., various sensors, sensor signal processing unit, camera, image processing unit, quality determination unit, etc.) when evaluating the quality of products manufactured by the manufacturing equipment 100. The quality evaluation control unit 201 generates quality evaluation result data linked to the product, but for quality items that are difficult to measure with sensors, such as the quality of appearance, it may be configured to accept evaluation input from the user and generate quality evaluation result data. The quality evaluation data output unit 202 has the function of outputting the quality evaluation data generated by the quality evaluation control unit 201 to the information processing device 300 in a timely manner.

[0031] Next, referring to the functional block diagram in Figure 4, the functions of the information processing device 300 will be described. The information processing unit of the information processing device 300 includes a measured data learning unit 310, a quality result learning unit 320, and a product quality prediction unit 330. These will be described in order below.

[0032] The measured data learning unit 310 has the function of machine learning the relationship between virtual data obtained from simulation and measured data acquired in the actual manufacturing process, and generating a trained model. The measured data learning unit 310 simulates the manufacturing process when manufacturing a product to generate virtual process data, and uses the virtual process data (explanatory variables) and measured process data based on actual experiments (dependent variables) as teaching materials to perform machine learning and generate a trained model.

[0033] Figure 5 is a schematic functional block diagram illustrating the configuration of the measured data learning unit 310. The measured data learning unit 310 includes an experimental condition determination unit 311, a measured process data acquisition unit 312, a measured quality data acquisition unit 313, a simulation execution unit 314, a virtual process data acquisition unit 315, a process data learning unit 316, and a measured quality learning unit 317. Each functional element shown is a functional concept and does not necessarily have to be physically configured as shown. For example, the specific forms of distribution and integration of each functional block are not limited to the examples shown, and all or part of them can be functionally or physically distributed and integrated in any unit depending on the usage situation, etc. These functional blocks can be configured, for example, by the CPU reading and executing an information processing program stored in a memory device or a non-temporary recording medium. Alternatively, some or all of the functional blocks may be configured by hardware such as an ASIC.

[0034] The experimental condition determination unit 311 has the function of determining the experimental conditions when operating the manufacturing equipment 100 to acquire actual measurement data. The experimental conditions should be such that when the conditions are changed, the process data also changes simultaneously. To select the experimental conditions, the user may use domain knowledge, as well as simulations to output virtual process data, allowing the user to verify how the virtual process data changes in response to changes in the conditions. The experimental condition determination unit 311 may then accept the user's selection.

[0035] The measured process data acquisition unit 312 has the function of acquiring measured process data when the manufacturing equipment 100 is operated under the conditions determined by the experimental condition determination unit 311, and storing it in association with the conditions determined by the experimental condition determination unit 311. Process data can include, for example, waveform data such as pressure and temperature measured by sensors. The measured process data acquisition unit 312 can extract numerical information such as the integral value, slope, and peak value of the waveform by performing signal processing on the waveform data. It can also extract characteristic information of the waveform shape and frequency information obtained by Fourier transform. The measured process data acquisition unit 312 acquires and stores characteristic information that has a high correlation with the quality result as measured process data, either from the waveform data itself or from the information obtained by signal processing the waveform data.

[0036] The measured quality data acquisition unit 313 has the function of acquiring quality data from the quality evaluation equipment 200 as a result of actually measuring the quality of the product manufactured by operating the manufacturing equipment 100 under the conditions determined by the experimental condition determination unit 311. Quality may include, for example, dimensional errors, shape errors (e.g., warp), weight errors, etc., of the manufactured product. For quality items that are difficult to measure with sensors, etc., the measured quality data acquisition unit 313 may be configured to accept evaluation results entered by the user. The measured quality data acquisition unit 313 acquires and stores measured quality evaluation data that links the product and the quality evaluation result.

[0037] The simulation unit 314 performs a computer simulation to simulate the manufacturing process when the manufacturing equipment 100 is operated based on the conditions determined by the experimental condition determination unit 311. For example, it calculates and outputs waveform data such as pressure and temperature that would be measured by sensors if the manufacturing equipment 100 were actually operated, through computer simulation. In this way, the simulation unit 314 simulates the manufacturing process (operating state) of the manufacturing equipment 100, but it does not simulate the product evaluation results by the quality evaluation equipment 200. Computer simulation can be performed by having the CPU 301 or another computer execute a suitable simulation program, including commercially available simulation software.

[0038] The virtual process data acquisition unit 315 has the function of acquiring virtual waveform data, etc., output by the simulation execution unit 314, linking it with the conditions determined by the experimental condition determination unit 311, and storing it as virtual process data. Virtual process data may include, for example, waveform data such as pressure and temperature output by the simulation execution unit 314. The virtual process data acquisition unit 315 can extract numerical information such as the integral value, slope, and peak value of the waveform by performing signal processing on the waveform data. It can also extract shape information of the waveform itself and frequency information obtained by Fourier transform. The virtual process data acquisition unit 315 acquires and stores characteristic information that has a high correlation with the quality results as virtual process data from the waveform data of the simulation results itself, or from the information obtained by signal processing the waveform data, as virtual process data. The virtual process data acquired by the virtual process data acquisition unit 315 is of the same type as the measured process data acquired by the measured process data acquisition unit 312, so that it can be compared with it.

[0039] The process data learning unit 316 uses virtual process data (explanatory variables) acquired by the virtual process data acquisition unit 315 for the conditions determined by the experimental condition determination unit 311, and actual process data (target variable) acquired by the actual process data acquisition unit 312 under the same manufacturing conditions, as teaching materials to perform machine learning and generate a process data prediction unit 322 (Figure 6) as a trained model. The process data prediction unit is a trained model that, when input with virtual process data (explanatory variables) generated using simulation, predicts and outputs the actual process data (target variable) that should be measured by the actual machine. The output of the process data prediction unit 322 is used as teaching material when the product quality learning unit 324 performs machine learning, so the data output by the process data prediction unit 322 can be called virtual process data for teaching materials.

[0040] In the machine learning performed by the process data learning unit 316, various feature information extracted from the virtual process data waveform obtained by the virtual process data acquisition unit 315 can be selected as explanatory variables. Furthermore, feature information extracted from the measured process data waveform obtained by the measured process data acquisition unit 312 can be selected as the objective variable. The feature information selected is that which has a high correlation with the quality results.

[0041] For the learning algorithm, general regression analysis methods can be used. For example, linear regression methods such as LinearReggression and ElasticNet, or decision trees such as RandomForest, XGBoost, and LightGBM can be used. In addition, if the shape feature information of the waveform itself is used, a convolutional neural network may be used. Furthermore, hyperparameter tuning may be performed to improve prediction accuracy, and methods such as grid search and Bayesian optimization can be used for this purpose.

[0042] The process data learning unit 316 performs the above information processing, thereby generating a trained model of the process data prediction unit 322 (Figure 6, described later) which predicts actual process data from virtual process data.

[0043] The measured quality learning unit 317 uses the virtual process data (explanatory variables) acquired by the virtual process data acquisition unit 315 and the measured quality data (target variable) acquired by the measured quality data acquisition unit 313 as learning materials to perform machine learning and generate a quality prediction unit as a trained model. The quality prediction unit is a trained model that, when given virtual process data (explanatory variables) generated using simulation as input, predicts and outputs the quality (target variable) as a result of evaluating a product manufactured under those process conditions using the quality evaluation equipment 200.

[0044] In the machine learning performed by the measured quality learning unit 317, various feature information extracted from the virtual process data waveform obtained by the virtual process data acquisition unit 315 can be selected as explanatory variables. Furthermore, the measured quality data of the product obtained from the measured quality data acquisition unit 313 can be used as the target variable. The explanatory and target variables must correspond to the same experimental conditions. The machine learning algorithm and other related methods may be the same as those used in the process data learning unit 316 described above.

[0045] The measured quality learning unit 317 performs the above information processing, thereby generating a trained model for the quality prediction unit 323 (Figure 6, described later) that predicts measured quality data from virtual process data. The output of the quality prediction unit 323 is used as teaching material when the product quality learning unit 324 performs machine learning, so the data output by the quality prediction unit 323 can be called virtual quality data for teaching materials.

[0046] Next, the quality result learning unit 320 shown in Figure 4 will be described. The quality result learning unit 320 has the function of machine learning the relationship between virtual process data obtained from simulation and actual measured quality data for products that have actually been manufactured, and generating a product quality prediction model 332 (Figure 7) as a trained model.

[0047] Figure 6 is a schematic functional block diagram illustrating the configuration of the quality result learning unit 320. Each functional element shown is conceptual and does not necessarily need to be physically configured as shown. For example, the specific forms of distribution and integration of each functional block are not limited to the example shown, and all or part of them can be functionally or physically distributed and integrated in any unit depending on the usage situation. These functional blocks can be configured, for example, by the CPU reading and executing an information processing program stored in a memory device or non-temporary recording medium. Alternatively, some or all of the functional blocks may be configured by hardware such as an ASIC.

[0048] The quality result learning unit 320 performs simulations under manufacturing conditions different from those used in the experiment and generates virtual process data through the simulation. The virtual process data from the simulation is input to the process data prediction unit 322, which outputs predicted values ​​for the process data. The virtual process data from the simulation is also input to the quality prediction unit 323, which outputs predicted values ​​for quality. The product quality learning unit 324 uses actual measured process data based on the actual experiment and the predicted values ​​for process data predicted by the process data prediction unit 322 as explanatory variables, and actual measured quality data and the predicted values ​​for quality predicted by the quality prediction unit 323 as dependent variables to perform machine learning and generate a trained product quality prediction model 332 (Figure 7).

[0049] The quality result learning unit 320 includes, as functional blocks, an experimental condition determination unit 311, an actual measurement process data acquisition unit 312, an actual measurement quality data acquisition unit 313, a simulation condition determination unit 321, a simulation execution unit 314, a virtual process data acquisition unit 315, a process data prediction unit 322, and a quality prediction unit 323. Of these, the experimental condition determination unit 311, the actual measurement process data acquisition unit 312, the actual measurement quality data acquisition unit 313, the simulation execution unit 314, and the virtual process data acquisition unit 315 are the same as the components of the actual measurement data learning unit 310 described with reference to Figure 5, so their explanation is omitted here.

[0050] However, in the actual data learning unit 310 shown in Figure 5, the simulation execution unit 314 performs the simulation based on the manufacturing conditions determined by the experimental condition determination unit 311, whereas in the quality result learning unit 320 shown in Figure 6, the simulation execution unit 314 performs the simulation based on the manufacturing conditions determined by the simulation condition determination unit 321.

[0051] The process data prediction unit 322 is a trained model generated as a result of machine learning performed by the process data learning unit 316 (Figure 5) of the actual measurement data learning unit 310. When virtual process data is input, it predicts and outputs the actual measurement process data.

[0052] The quality prediction unit 323 is a trained model generated as a result of machine learning performed by the measured quality learning unit 317 (Figure 5) of the measured data learning unit 310. When virtual process data is input, it predicts and outputs the quality evaluation result if the product were manufactured using that virtual process.

[0053] The simulation condition determination unit 321 can set arbitrary manufacturing conditions that may occur in the actual manufacturing equipment 100 and cause the simulation execution unit 314 to perform a simulation. For example, the simulation condition determination unit 321 can set multiple manufacturing conditions to comprehensively cover the range of variation of actual manufacturing conditions and cause the simulation execution unit 314 to perform a simulation. If there are multiple conditions, the manufacturing conditions may be set using a grid search or the like to cover all combinations of those conditions.

[0054] The simulation condition determination unit 321 sets multiple manufacturing conditions, the simulation execution unit 314 performs simulations for these conditions, and the virtual process data acquisition unit 315 acquires virtual process data. The process data prediction unit 322 predicts the process data that would be measured under each manufacturing condition based on the virtual process data and outputs it as a predicted value of the process data. In addition, the quality prediction unit 323 predicts the evaluation result of the quality of the products manufactured under each manufacturing condition based on the virtual process data and outputs it as a predicted value of the quality.

[0055] The product quality learning unit 324 performs machine learning and generates a product quality prediction model 332 (Figure 7) as a trained model. The product quality prediction model 332 is a trained model that, when actual measured process data (explanatory variables) is input during the manufacturing of a product, predicts and outputs the quality of the manufactured product (dependent variable).

[0056] The training data used for machine learning includes, as explanatory variables, predicted values ​​of process data output by the process data prediction unit 322 (virtual process data for educational purposes) and actual process data acquired by the actual process data acquisition unit 312. The training data also includes, as objective variables, predicted values ​​of quality output by the quality prediction unit 323 (virtual quality data for educational purposes) and actual quality data acquired by the actual quality data acquisition unit 313.

[0057] Next, the product quality prediction unit 330 shown in Figure 4 will be described. The product quality prediction unit 330 has the function of acquiring measured process data when the manufacturing equipment 100 actually manufactures a product, predicting the quality of the product based on that data, and outputting quality prediction information (for example, a prediction that the quality is poor).

[0058] Figure 7 is a schematic functional block diagram illustrating the configuration of the product quality prediction unit 330. The product quality prediction unit 330 includes a manufacturing process data acquisition unit 331, a product quality prediction model 332, and a quality information output unit 333.

[0059] The manufacturing process data acquisition unit 331 has the function of acquiring actual process data acquired by the process data acquisition unit 102 (Figure 3) of the manufacturing equipment 100 and output from the process data output unit 103 when the manufacturing equipment 100 is actually manufacturing a product, via the I / O port 305 (Figure 2).

[0060] The product quality prediction model 332 is a trained model generated as a result of machine learning performed by the product quality learning unit 324 (Figure 6) of the quality result learning unit 320. When actual process data is input, it predicts and outputs the quality of the product manufactured in that manufacturing process. The product quality prediction model 332 can determine whether the predicted quality meets the quality standards.

[0061] The quality information output unit 333 has the function of outputting information related to product quality (for example, the result of determining whether the predicted quality value meets the quality standard) to the quality information receiving unit 104 of the manufacturing equipment 100 via the I / O port 305 (Figure 2). It is desirable that the product quality prediction unit 330 generates and outputs information related to product quality within the manufacturing cycle of the manufacturing equipment 100.

[0062] The configuration and functions of the information processing device (information processing system) according to the embodiment have been described above. Next, the information processing method according to the embodiment will be described.

[0063] (Information processing methods) Figure 8 is a flowchart illustrating the procedure by which the actual data learning unit 310 (Figures 4 and 5) of the information processing device 300 generates the trained models, namely the process data prediction unit 322 (Figure 6) and the quality prediction unit 323 (Figure 6). The actual data learning unit 310 generates the trained models by machine learning the relationship between virtual data obtained from simulations and actual data acquired in the manufacturing process.

[0064] First, in step S10, the experimental conditions for the experiment, which involves operating the manufacturing equipment 100 to acquire actual measurement data, are determined by the experimental condition determination unit 311 (Figure 5). The experimental conditions selected are such that when the conditions are changed, the process data also changes simultaneously. The user can select experimental conditions by outputting virtual process data using domain knowledge and simulations, and by checking how the virtual process data changes in response to changes in the conditions. The experimental condition determination unit 311 receives and stores information on the experimental conditions (manufacturing conditions to be implemented in the experiment) input by the user.

[0065] In the subsequent step S11, the manufacturing equipment 100 manufactures the product according to the manufacturing conditions determined by the experimental condition determination unit 311 in step S10. At that time, the measured process data acquisition unit 312 of the information processing device 300 acquires the measured process data output from the process data output unit 103 of the manufacturing equipment 100.

[0066] In step S12, the simulation execution unit 314 performs a computer simulation to simulate the operation of the manufacturing equipment 100 based on the manufacturing conditions determined by the experimental condition determination unit 311 in step S10. Here, the simulation execution unit 314 simulates the operating state of the manufacturing equipment 100, but does not simulate the product evaluation results by the quality evaluation equipment 200.

[0067] In step S13, the virtual process data acquisition unit 315 links the virtual waveform data output by the simulation execution unit 314 in step S12 with the conditions determined by the experimental condition determination unit 311 and acquires it as virtual process data.

[0068] In step S14, the process data learning unit 316 uses the virtual process data (explanatory variables) acquired by the virtual process data acquisition unit 315 and the actual process data (target variable) acquired by the actual process data acquisition unit 312 as learning materials to perform machine learning and generate the process data prediction unit 322 as a trained model.

[0069] In step S15, the measured quality data acquisition unit 313 acquires measured quality evaluation data. The measured quality evaluation data is quality data obtained by the quality evaluation equipment 200 from products actually manufactured under the conditions determined by the experimental condition determination unit 311.

[0070] In step S16, the virtual process data (explanatory variables) acquired by the virtual process data acquisition unit 315 and the actual quality data (dependent variable) acquired by the actual quality data acquisition unit 313 are used as teaching materials by the actual quality learning unit 317 to perform machine learning and generate the quality prediction unit 323 as a trained model.

[0071] Through the above procedure, the trained models, process data prediction unit 322 (Figure 6) and quality prediction unit 323 (Figure 6), are generated.

[0072] Next, Figure 9 is a flowchart illustrating the information processing procedure by which the quality result learning unit 320 (Figures 4 and 6) of the information processing device 300 generates a trained model, the product quality prediction unit 330 (Figures 4 and 7). The quality result learning unit 320 generates a trained model by machine learning the relationship between virtual process data obtained from simulations and actual measured quality data for products that have actually been manufactured.

[0073] First, in step S20, the simulation condition determination unit 321 sets arbitrary manufacturing conditions that may occur in the actual manufacturing equipment 100 as simulation conditions.

[0074] In the following step S21, the simulation execution unit 314 performs a simulation according to the manufacturing conditions determined by the simulation condition determination unit 321 in step S20.

[0075] In step S22, the virtual process data acquisition unit 315 links the virtual waveform data output by the simulation execution unit 314 with the conditions determined by the simulation condition determination unit 321 and acquires it as virtual process data.

[0076] In step S23, the virtual process data acquired by the virtual process data acquisition unit 315 in step S22 is input as explanatory variables to the process data prediction unit 322, which outputs predicted values ​​of the process data that would be measured if an experiment were conducted. The process data prediction unit 322 outputs the predicted values ​​of the process data to the product quality learning unit 324.

[0077] In step S24, the virtual process data acquired by the virtual process data acquisition unit 315 in step S22 is input as explanatory variables to the quality prediction unit 323, which predicts the evaluation results if the actual product were evaluated by the quality evaluation equipment 200, and outputs a quality prediction. The quality prediction unit 323 outputs the quality prediction to the product quality learning unit 324. Note that steps S23 and S24 described above may be performed in the reverse order.

[0078] In step S25, the product quality learning unit 324 performs machine learning and generates a product quality prediction model 332 (Figure 7) as a trained model. The learning materials for machine learning include, as explanatory variables, predicted values ​​of process data output by the process data prediction unit 322 and actual process data acquired by the actual process data acquisition unit 312. The learning materials for machine learning also include, as objective variables, the quality prediction output by the quality prediction unit 323 and actual quality data acquired by the actual quality data acquisition unit 313. Following the above procedure, the product quality prediction unit 330 generates a trained product quality prediction model 332 (Figure 7).

[0079] Next, Figure 10 is a flowchart illustrating the procedure for the information processing device 300 to predict the quality of the products to be manufactured when the manufacturing equipment 100 manufactures the products.

[0080] First, in step S30, the manufacturing process data acquisition unit 331 (Figure 7) of the product quality prediction unit 330 acquires process data measured during product manufacturing from the process data output unit 103 (Figure 3) of the manufacturing equipment 100.

[0081] In the subsequent step S31, the product quality prediction unit 330 inputs the process data acquired by the manufacturing process data acquisition unit 331 in step S30 into the product quality prediction model 332 and predicts the product quality. The product quality prediction model 332 outputs the quality prediction result to the quality information output unit 333.

[0082] In step S32, the quality information output unit 333 determines whether the prediction result from the product quality prediction model 332 meets a predetermined quality standard. If it is determined that the predetermined quality standard is met (step S32: YES), the information processing for the product under manufacture is terminated. If it is determined that the predetermined quality standard is not met (step S32: NO), the process proceeds to step S33, where the quality information output unit 333 transmits a defect signal to the quality information receiving unit 104 (Figure 3) of the manufacturing equipment 100 indicating that the quality of the product under manufacture is predicted to be poor, and terminates the information processing for the product under manufacture.

[0083] Furthermore, the processing procedure may be configured such that the quality information output unit 333 transmits the prediction result to the quality information receiving unit 104 of the manufacturing equipment 100, regardless of whether the prediction result from the product quality prediction model 332 meets the predetermined quality standards.

[0084] Products predicted to be of poor quality can be handled in various ways, such as being sorted out and not shipped by the defective product sorting unit 105 of the manufacturing equipment 100, or having their quality measured by the quality evaluation equipment 200 to make a final determination of whether they are good or bad. Furthermore, if, for example, multiple products are predicted to be of poor quality in succession, the information processing device 300 may be configured to automatically transmit this information to the manager of the manufacturing equipment 100.

[0085] According to this embodiment, when generating a product quality prediction model that predicts product quality based on manufacturing process data using machine learning, training data for machine learning can be efficiently prepared by using a combination of experiments and simulations.

[0086] In other words, virtual process data obtained through simulation for a given manufacturing condition is acquired as explanatory variables, and actual process data measured by manufacturing equipment for the same manufacturing condition is acquired as the target variable. These are then used to create a training dataset. Then, using machine learning with this training dataset, a process data prediction unit (trained model) that can predict the actually measured process data with high accuracy is generated.

[0087] Furthermore, virtual process data obtained through simulation for certain manufacturing conditions is acquired as explanatory variables, and actual quality data measured for products actually manufactured under the same manufacturing conditions is acquired as the target variable. These are then used to create a training dataset. Finally, machine learning using this training dataset is used to generate a quality prediction unit (a trained model) that can predict the quality evaluation results of products with high accuracy.

[0088] Furthermore, a simulation is performed using manufacturing conditions different from those actually used in the experiment, and the simulation results are input into the process data prediction unit (trained model) and the quality prediction unit (trained model). Then, a training dataset is created with the predicted process data values ​​obtained from the process data prediction unit (trained model) as explanatory variables and the predicted quality evaluation results obtained from the quality prediction unit (trained model) as the dependent variable. Then, a product quality prediction model (trained model) capable of predicting product quality with high accuracy based on process data is generated through machine learning using this training dataset. It is desirable to perform machine learning in conjunction with a training dataset in which the dependent variable is actual process data measured by manufacturing equipment in an actual experiment, and the dependent variable is quality data measured on actually manufactured products.

[0089] Thus, according to this embodiment, it is possible to easily and abundantly prepare training data that appropriately reflects the relationship between process data obtained through simulation for various manufacturing conditions and quality data measured on actually manufactured products. In other words, a product quality prediction model with high prediction accuracy can be generated by machine learning using appropriate and abundant training data. Therefore, the product quality prediction model according to this embodiment can exhibit higher prediction accuracy compared to a product quality prediction model generated by machine learning using only a small amount of training data consisting only of process data obtained through actual experiments and quality evaluation results of actual products. In other words, since it is possible to prepare a large amount of appropriate training data without conducting large-scale experiments to obtain a large amount of actual measurement data, it is possible to generate a product quality prediction model with high prediction accuracy with less effort.

[0090] [Examples] This example shows how the information processing device according to the above embodiment is applied to a manufacturing system for producing resin molded products using an injection molding apparatus.

[0091] Figure 11(a) is a perspective view showing the external shape of the resin molded product 51. The material of the resin molded product 51 is, for example, polystyrene. The external dimensions of the resin molded product 51 are, as an example, 210 mm in length, 30 mm in width, and 1.5 mm in thickness.

[0092] The resin molded product 51 is manufactured by injecting molten resin from the injection gate into a cavity defined by a fixed mold and a movable mold of an injection molding machine, allowing it to cool and solidify, then moving the movable mold to open the mold and release the solidified molded product. The movable mold is positioned on the side of the three rib-shaped resin molded product 51, and the fixed mold is positioned on the opposite side.

[0093] Figure 11(b) is a schematic diagram of the cavity CAV 51 for molding the resin molded product 51, as seen from the fixed mold side, to illustrate the arrangement of the components of the injection molding apparatus. An injection gate for injecting molten resin into the cavity CAV 51 is located at gate position 61.

[0094] Pressure sensors are positioned at pressure sensor positions 64 and 65 to measure the pressure at which molten resin is injected, or the pressure inside the cavity during holding pressure. The pressure sensor at pressure sensor position 64 measures the pressure waveform (time-dependent profile) near the injection gate, while the pressure sensor at pressure sensor position 65 measures the pressure waveform at a position far from the injection gate.

[0095] Temperature sensors are positioned at temperature sensor positions 66 and 67 to measure the temperature of the resin material or the mold. The temperature sensor at temperature sensor position 66 measures the temperature waveform (time profile) near the injection gate, and the temperature sensor at temperature sensor position 67 measures the temperature waveform at a position far from the injection gate. The measured values ​​from the pressure sensor and temperature sensor are output as actual process data from the injection molding machine to the information processing device.

[0096] Marks engraved on the molding surface of the mold are placed at dimensional reference positions 62 and 63. The distance between dimensional reference positions 62 and 63 is set to, for example, 170 mm. The marks placed at the dimensional reference positions are transferred to the resin molded product, and by measuring the position and shape of the marks transferred to the resin molded product with the quality evaluation equipment 200, actual quality data on the product shape can be obtained. The measured values ​​for the position and shape of the marks are output from the quality evaluation equipment 200 to the information processing device as actual quality data.

[0097] In the information processing that generates the process data prediction unit 322 (Figure 6) and the quality prediction unit 323 (Figure 6), as explained with reference to Figure 8, the experimental conditions were determined in step S10 as follows. Manufacturing conditions that correlate with dimensional accuracy as product quality were selected, such as holding pressure amount, holding pressure time, and resin temperature, and the range of variation for each condition (parameter) was determined. The range of variation was set in accordance with the range that can be changed in actual production or the range that can fluctuate in actual production. In actual manufacturing, ranges in which molding defects such as sink marks, shorts, and burrs are known to occur are not set as manufacturing conditions, so these ranges were excluded when setting the range of variation for the experimental conditions.

[0098] In this experiment, the holding pressure was set to 25 MPa, 30 MPa, and 35 MPa, and the injection resin temperature was set to 200°C, 215°C, and 230°C. Nine different experimental conditions were established by combining these settings. The injection time for the molten resin was set to 1.7 seconds, and the holding pressure time during cooling was set to 4 seconds.

[0099] Figure 12 shows an example of process data acquired in step S11 of the information processing flow shown in Figure 8. Figure 12 shows the waveform (time-series profile) of the pressure near the injection gate measured by a pressure sensor located at pressure sensor position 64, with the horizontal axis of the graph representing time (seconds) and the vertical axis representing pressure (bars).

[0100] As actual process data, in addition to the pressure waveform 71 itself, feature quantities such as peak values, integral values, and slope values ​​extracted by signal processing of the pressure waveform 71 can be used. For example, the pressure waveform 71 may be integrated only for the holding period 72 corresponding to the holding process and used as actual process data. Similarly, for temperature, feature quantities may be extracted using only the measurement data corresponding to a specific process in product manufacturing and used as actual process data.

[0101] In step S15 of the information processing flow shown in Figure 8, the position and shape of the marks transferred to the resin molded product manufactured in the experiment were measured using the quality evaluation equipment 200, and actual quality data regarding the product shape was obtained.

[0102] In steps S14 and S16 of the information processing flow shown in Figure 8, machine learning was performed using ElasticNet, a linear regression algorithm, to generate a trained model. Note that if the number of training data points used for machine learning is 50 or more, a decision tree algorithm may be used instead of the linear regression algorithm.

[0103] Furthermore, in step S20 of the information processing flow shown in Figure 9, more conditions were set as simulation conditions than the experimental conditions set in step S10. In step S10, three values ​​each were selected for the pressure of the holding process and the temperature of the injected resin, and these were combined to set nine different experimental conditions. However, in step S20, the pressure and temperature were set more finely within the same range (for example, 100 values ​​each), and all combinations of these (for example, 10,000 combinations) were set as simulation conditions.

[0104] In this manner, the information processing device according to the embodiment was configured, and product quality (shape accuracy) was predicted when actually manufacturing resin molded products using an injection molding machine. For comparison, a product quality prediction model was generated by machine learning using only experimental data without simulation, and the prediction accuracy was compared. The prediction accuracy was evaluated using the coefficient of determination, which represents how well the explanatory variables explain the dependent variable. The coefficient of determination of the product quality prediction model of the comparative example was 0.82, while the coefficient of determination of the information processing device of the embodiment was 0.90, confirming that the embodiment had significantly better prediction accuracy than the comparative example.

[0105] [Other embodiments] It should be noted that the present invention is not limited to the embodiments and examples described above, and many modifications are possible within the technical concept of the present invention. For example, all or part of the different embodiments and examples described above may be combined and implemented.

[0106] For example, the types of manufacturing equipment and the types of products to be manufactured are not limited to the examples described above. The information processing apparatus and information processing method according to the present invention may be applied, for example, to a manufacturing system that processes the surface of a metal material using a cutting machine to manufacture a metal product. In that case, for example, experimental conditions (step S10) and simulation conditions (step S20) may be the rotational speed of the cutting tool, the feed rate, the depth of cut, etc., and actual process data may be the vibration of the tool measured by a vibration sensor, the sound during cutting measured by an acoustic wave sensor, or the load torque of the cutting tool measured by a torque sensor. In addition, actual quality data may be the surface quality of the machined surface (e.g., surface roughness) measured by a shape measuring device.

[0107] Furthermore, the information processing apparatus and information processing method according to the present invention can be applied, for example, to a manufacturing system that uses a robot to assemble parts and manufacture products. In this case, for example, experimental conditions (step S10) and simulation conditions (step S20) can be used, such as the gripping force of the robot hand gripping the part, the rotational speed and movement speed of the robot arm, and measured values ​​obtained from the tachometer of the motors driving each joint and the torque sensors of each joint can be used as measured process data. In addition, as measured quality data, results of measuring the shape of the product assembled by the robot using a shape measuring device can be used.

[0108] Manufacturing apparatus and manufacturing systems equipped with the information processing apparatus according to the present invention are also included in embodiments of the present invention. Methods for manufacturing articles that produce products by applying the information processing apparatus or information processing method according to the present invention are also included in embodiments of the present invention. Programs that cause a computer to execute the information processing method according to the present invention, and recording media that can be read by a computer on which the program is recorded are also included in embodiments of the present invention. Flexible disks, optical disks, magneto-optical disks, magnetic tapes, USB memory, SSDs, etc., can be used as recording media.

[0109] The present invention can also be realized by supplying a program that implements one or more of the functions of the embodiments to a system or device via a network or storage medium, and by having one or more processors in the computer of that system or device read and execute the program. It can also be realized by a circuit (e.g., an ASIC) that implements one or more functions.

[0110] This specification discloses at least the following: [Matter 1] The information processing unit, Actual process data measured when the manufacturing equipment produces a product under the first manufacturing conditions, Actual quality data obtained by measuring the quality of the product manufactured under the first manufacturing conditions, First virtual process data generated by simulating the manufacturing process in which the manufacturing apparatus manufactures the product under the first manufacturing conditions is obtained, A process data prediction unit is generated by machine learning using training data in which the first virtual process data is the explanatory variable and the measured process data is the target variable. A quality prediction unit is generated by machine learning using training data in which the first virtual process data is the explanatory variable and the measured quality data is the target variable. A manufacturing process in which the manufacturing apparatus manufactures a product under second manufacturing conditions different from the first manufacturing conditions is simulated to obtain second virtual process data. The second virtual process data is input to the process data prediction unit to output virtual process data for educational materials. The second virtual process data is input to the quality prediction unit to output virtual quality data for educational materials. Using the aforementioned measured process data and the aforementioned virtual process data for educational purposes as explanatory variables, and the aforementioned measured quality data and the aforementioned virtual quality data for educational purposes as training data, a product quality prediction model is generated by machine learning. An information processing method characterized by the following: [Matter 2] The aforementioned information processing unit, The second measured process data obtained when the manufacturing apparatus manufactures a product is acquired. The second measured process data is input into the product quality prediction model, and quality prediction information for the product manufactured by the manufacturing equipment when the second measured process data was acquired is output. The information processing method described in item 1, characterized by the features described above. [Matter 3] The information processing unit transmits the product quality prediction information output by the product quality prediction model to the manufacturing apparatus. The information processing method described in item 2, characterized by the features described above. [Matter 4] The information processing unit transmits the product quality prediction information output by the product quality prediction model to the user of the manufacturing apparatus. The information processing method described in item 2 or 3, characterized by the features described herein. [Matter 5] The information processing unit, Actual process data measured when the manufacturing equipment produces a product under the first manufacturing conditions, Actual quality data obtained by measuring the quality of the product manufactured under the first manufacturing conditions, First virtual process data generated by simulating the manufacturing process in which the manufacturing apparatus manufactures the product under the first manufacturing conditions is obtained, A process data prediction unit is generated by machine learning using training data in which the first virtual process data is the explanatory variable and the measured process data is the target variable. A quality prediction unit is generated by machine learning using training data in which the first virtual process data is the explanatory variable and the measured quality data is the target variable. A manufacturing process in which the manufacturing apparatus manufactures a product under second manufacturing conditions different from the first manufacturing conditions is simulated to obtain second virtual process data. The second virtual process data is input to the process data prediction unit to output virtual process data for educational materials. The second virtual process data is input to the quality prediction unit to output virtual quality data for educational materials. Using the aforementioned measured process data and the aforementioned virtual process data for educational purposes as explanatory variables, and the aforementioned measured quality data and the aforementioned virtual quality data for educational purposes as training data, a product quality prediction model is generated by machine learning. An information processing device characterized by the following: [Matter 6] The aforementioned information processing unit, The second measured process data obtained when the manufacturing apparatus manufactures a product is acquired. The second measured process data is input into the product quality prediction model, and quality prediction information for the product manufactured by the manufacturing equipment when the second measured process data was acquired is output. The information processing device according to item 5, characterized by the features described herein. [Matter 7] The information processing unit transmits the product quality prediction information output by the product quality prediction model to the manufacturing apparatus. The information processing device according to item 6, characterized in that it is a processing device. [Matter 8] The information processing unit transmits the product quality prediction information output by the product quality prediction model to the user of the manufacturing apparatus. An information processing device according to item 6 or 7, characterized by the features described therein. [Matter 9] The manufacturing apparatus comprises an information processing device described in any one of items 5 to 8, A manufacturing system characterized by the following features. [Matter 10] A manufacturing system comprising the information processing device described in item 7 and the manufacturing apparatus, The manufacturing apparatus determines how to handle the product based on the quality prediction information transmitted from the information processing device. A manufacturing system characterized by the following features. [Matter 11] The manufacturing apparatus selects the products that are predicted to be of poor quality based on the quality prediction information. A manufacturing system as described in item 10, characterized by the features described herein. [Matter 12] The product is manufactured using the manufacturing system described in item 9 or 10. A method for manufacturing an article characterized by the following: [Matter 13] A program that causes a computer to execute the information processing method described in any one of items 1 through 4. [Matter 14] A computer-readable recording medium containing the program described in item 13. [Explanation of Symbols]

[0111] 51···Resin molded product / 61···Gate position / 62···Dimensional reference position / 63···Dimensional reference position / 64···Pressure sensor position / 65···Pressure sensor position / 66···Temperature sensor position / 67···Temperature sensor position / 71···Pressure waveform / 72···Holding pressure period / 100···Manufacturing equipment / 101···Product manufacturing control unit / 102···Process data acquisition unit / 103···Process data output unit / 104···Quality information receiving unit / 105···Defective product sorting unit / 200···Quality evaluation equipment / 201···Quality evaluation control unit / 202···Quality evaluation data output unit / 300···Information processing device / 301···CPU / 302···Memory / 303···HDD / 304···System bus / 305···I / O port / 306· ··Input Device / 307···Display / 308···External Memory / 310···Measured Data Learning Unit / 311···Experimental Condition Determination Unit / 312···Measured Process Data Acquisition Unit / 313···Measured Quality Data Acquisition Unit / 314···Simulation Execution Unit / 315···Virtual Process Data Acquisition Unit / 316···Process Data Learning Unit / 317···Measured Quality Learning Unit / 320···Quality Result Learning Unit / 321···Simulation Condition Determination Unit / 322···Process Data Prediction Unit / 323···Quality Prediction Unit / 324···Product Quality Learning Unit / 330···Product Quality Prediction Unit / 331···Manufacturing Process Data Acquisition Unit / 332···Product Quality Prediction Model / 333···Quality Information Output Unit / CAV51···Cavity

Claims

1. The information processing unit, Actual process data measured when the manufacturing equipment produces a product under the first manufacturing conditions, Actual quality data obtained by measuring the quality of the product manufactured under the first manufacturing conditions, First virtual process data generated by simulating the manufacturing process in which the manufacturing apparatus manufactures the product under the first manufacturing conditions is obtained, A process data prediction unit is generated by machine learning using training data in which the first virtual process data is the explanatory variable and the measured process data is the target variable. A quality prediction unit is generated by machine learning using training data in which the first virtual process data is the explanatory variable and the measured quality data is the target variable. A manufacturing process in which the manufacturing apparatus manufactures a product under second manufacturing conditions different from the first manufacturing conditions is simulated to obtain second virtual process data. The second virtual process data is input to the process data prediction unit to output virtual process data for educational materials. The second virtual process data is input to the quality prediction unit to output virtual quality data for educational materials. Using the aforementioned measured process data and the aforementioned virtual process data for educational purposes as explanatory variables, and the aforementioned measured quality data and the aforementioned virtual quality data for educational purposes as training data, a product quality prediction model is generated by machine learning. An information processing method characterized by the following:

2. The aforementioned information processing unit, The manufacturing apparatus acquires second measured process data measured when it manufactures a product. The second measured process data is input into the product quality prediction model, and quality prediction information for the product manufactured by the manufacturing equipment when the second measured process data was acquired is output. The information processing method according to feature 1.

3. The information processing unit transmits the product quality prediction information output by the product quality prediction model to the manufacturing apparatus. The information processing method according to feature 2.

4. The information processing unit transmits the product quality prediction information output by the product quality prediction model to the user of the manufacturing apparatus. The information processing method according to feature 2.

5. The information processing unit, Actual process data measured when the manufacturing equipment produces a product under the first manufacturing conditions, Actual quality data obtained by measuring the quality of the product manufactured under the first manufacturing conditions, First virtual process data generated by simulating the manufacturing process in which the manufacturing apparatus manufactures the product under the first manufacturing conditions is obtained, A process data prediction unit is generated by machine learning using training data in which the first virtual process data is the explanatory variable and the measured process data is the target variable. A quality prediction unit is generated by machine learning using training data in which the first virtual process data is the explanatory variable and the measured quality data is the target variable. A manufacturing process in which the manufacturing apparatus manufactures a product under second manufacturing conditions different from the first manufacturing conditions is simulated to obtain second virtual process data. The second virtual process data is input to the process data prediction unit to output virtual process data for educational materials. The second virtual process data is input to the quality prediction unit to output virtual quality data for educational materials. Using the aforementioned measured process data and the aforementioned virtual process data for educational purposes as explanatory variables, and the aforementioned measured quality data and the aforementioned virtual quality data for educational purposes as training data, a product quality prediction model is generated by machine learning. An information processing device characterized by the following:

6. The aforementioned information processing unit, The manufacturing apparatus acquires second measured process data measured when it manufactures a product. The second measured process data is input into the product quality prediction model, and quality prediction information for the product manufactured by the manufacturing equipment when the second measured process data was acquired is output. The information processing apparatus according to feature 5.

7. The information processing unit transmits the product quality prediction information output by the product quality prediction model to the manufacturing apparatus. The information processing apparatus according to feature 6.

8. The information processing unit transmits the product quality prediction information output by the product quality prediction model to the user of the manufacturing apparatus. The information processing apparatus according to feature 6.

9. A manufacturing apparatus comprising an information processing apparatus as described in any one of claims 5 to 8, A manufacturing system characterized by the following features.

10. A manufacturing system comprising the information processing device described in claim 7 and the manufacturing apparatus, The manufacturing apparatus determines how to handle the product based on the quality prediction information transmitted from the information processing device. A manufacturing system characterized by the following features.

11. The manufacturing apparatus selects the products that are predicted to be of poor quality based on the quality prediction information. The manufacturing system according to claim 10.

12. A product is manufactured using the manufacturing system described in claim 9. A method for manufacturing an article characterized by the following:

13. A program for causing a computer to execute the information processing method described in any one of claims 1 to 4.

14. A computer-readable recording medium having the program described in claim 13 recorded on it.

Images