JP2021012475A - Apparatus and method for information processing and program - Google Patents

Abstract

To provide an apparatus for information processing adapted to stabilize the quality of a manufactured object.SOLUTION: An apparatus for information processing includes a control apparatus (1) comprising a molding-condition calculating section (12) that derives a to-input variable to be inputted to a manufacturing apparatus in order to control the manufacturing apparatus, and an acquiring section (11) that acquires a not-to-input variable that is a variable other than the to-input variable the molding-condition calculating section (12) derives. The molding-condition calculating section (12) derives a to-input variable by making reference to the not-to-input variable the acquiring section (11) acquired.SELECTED DRAWING: Figure 2

Description

The present invention relates to an information processing device, an information processing method and a program.

As a conventional technique, there is a manufacturing apparatus that manufactures a product under certain molding conditions by using data of molding conditions that are considered to be optimal.

Japanese Unexamined Patent Publication No. 2014-4828 (published on January 16, 2014)

However, the conventional manufacturing apparatus has a problem that the quality of the manufactured product varies and is not stable when the temperature and humidity change.

For example, as shown in FIG. 12, the daily temperature and humidity change. When a product is manufactured under certain molding conditions under such an environment, as shown in FIG. 13, the dimensions of the product change with changes in temperature and humidity.

One aspect of the present invention is intended to stabilize the quality of a product.

In order to solve the above problems, the information processing device according to one aspect of the present invention has a derivation unit for deriving an input target variable input to the manufacturing device in order to control the manufacturing device, and a derivation unit derived by the derivation unit. It includes an acquisition unit for acquiring a non-input target variable which is a variable other than the input target variable to be input, and the derivation unit derives the input target variable with reference to the non-input target variable acquired by the acquisition unit. ..

According to the above configuration, since the input target variable input to the manufacturing apparatus is derived with reference to the non-input target variable, the quality of the product manufactured by the manufacturing apparatus can be stabilized.

In the information processing apparatus according to one aspect of the present invention, the non-input target variable may include at least one of a value related to the production environment at the time of manufacturing the manufacturing object and an actually measured value of the manufacturing conditions of the manufacturing apparatus. Good.

In the information processing apparatus according to one aspect of the present invention, the input target variable may include a specified value that defines the manufacturing conditions by the manufacturing apparatus.

The information processing apparatus according to one aspect of the present invention is a prediction model in which the derivation unit inputs at least one of the input target variable and the non-input target variable and outputs a predicted value of quality information regarding the quality of the manufacturing target. The input target variable may be derived with reference to.

According to the above configuration, since the prediction model that outputs the predicted value of the quality information is referred to, the input target variable with high accuracy can be derived. Therefore, the quality of the product can be further stabilized.

The information processing apparatus according to one aspect of the present invention is learned by the prediction model so that the difference between the measured value of the quality information related to the actual quality and the predicted value of the quality information derived from the prediction model becomes smaller. It may be a model.

The information processing apparatus according to one aspect of the present invention uses the data acquisition unit for acquiring teacher data in which the input target variable, the non-input target variable, and the quality information are associated with each other, and the prediction model. A learning unit for learning using the teacher data acquired by the data acquisition unit may be further provided.

According to the above configuration, since the prediction model is trained using the teacher data in which the input target variable, the non-input target variable, and the quality information are associated with each other, the prediction model can be made highly accurate.

In the information processing apparatus according to one aspect of the present invention, the derivation unit repeatedly executes the grid search and the local gradient method using the output value of the prediction model, and responds to the non-input target variable acquired by the acquisition unit. You may derive the input target variable.

According to the above configuration, since the grid search as a rough search and the local gradient method as a detailed search are repeatedly executed, it is possible to derive an input target variable with high accuracy. Therefore, the quality of the product can be further stabilized.

In the information processing apparatus according to one aspect of the present invention, the derivation unit may derive an input target variable for controlling a manufacturing apparatus that manufactures a manufacturing object by injection molding.

According to the above configuration, injection molding can be performed with high accuracy.

The information processing apparatus according to one aspect of the present invention may further include a display data generation unit that generates display data for displaying the derived input target variable.

According to the above configuration, since the display data for displaying the input target variable is generated, it is possible to display the input target variable with high accuracy.

The information processing device according to one aspect of the present invention includes an input target variable input to the manufacturing device to control the manufacturing device, a non-input target variable which is a variable other than the input target variable, and a manufacturing target. A data acquisition unit that acquires teacher data in which quality information related to quality is associated with each other, and at least one of the input target variable and the non-input target variable is input and a predicted value of quality information related to the quality of the manufacturing object is output. A learning unit for learning the prediction model to be performed using the teacher data acquired by the data acquisition unit may be provided.

According to the above configuration, since the prediction model is trained using the teacher data in which the input target variable, the non-input target variable, and the quality information are associated with each other, the prediction model can be made highly accurate.

The information processing method according to one aspect of the present invention includes an acquisition step of acquiring a non-input target variable which is a variable other than the input target variable input to the manufacturing device in order to control the manufacturing device, and the input target variable. The input target variable may be derived by referring to the non-input target variable acquired in the acquisition step in the derivation step including the derivation step of deriving.

According to the above configuration, since the input target variable input to the manufacturing apparatus is derived with reference to the non-input target variable, the quality of the product manufactured by the manufacturing apparatus can be stabilized.

The information processing method according to one aspect of the present invention includes an input target variable input to the manufacturing device to control the manufacturing device, a non-input target variable which is a variable other than the input target variable, and a manufacturing target. A data acquisition process for acquiring teacher data in which quality information related to quality is associated with each other, and at least one of the input target variable and the non-input target variable is input and a predicted value of quality information related to the quality of the manufacturing object is output. A learning step of training the prediction model to be performed using the teacher data acquired in the data acquisition step may be included.

According to the above configuration, since the prediction model is trained using the teacher data in which the input target variable, the non-input target variable, and the quality information are associated with each other, the prediction model can be made highly accurate.

Further, the information processing device according to each aspect of the present invention may be realized by a computer. In this case, the information processing device is operated by operating the computer as each part (software element) included in the information processing device. A control program realized by a computer and a computer-readable recording medium on which the control program is recorded also fall within the scope of the present invention.

According to one aspect of the present invention, the quality of the product can be stabilized.

It is a schematic diagram which shows the production environment in which the control device which concerns on one Embodiment of this invention is used. It is a block diagram which shows the structure of the control device which concerns on one Embodiment of this invention. It is a flowchart which shows the learning process of the control device which concerns on one Embodiment of this invention. It is a figure which shows the example of the teacher data which concerns on one Embodiment of this invention. It is a figure which shows the regression line and the regression curve output by the product quality prediction model which concerns on one Embodiment of this invention. It is a flowchart which shows the molding condition search process of the control apparatus which concerns on one Embodiment of this invention. It is a figure which shows an example of the molding condition which concerns on one Embodiment of this invention. It is a figure which shows an example of the judgment criteria of the pass / fail judgment of the predicted value of the appearance quality of the product quality prediction model which concerns on one Embodiment of this invention. It is a graph which shows the correspondence relationship between the quality quantitative value of appearance quality and the evaluation function which concerns on one Embodiment of this invention. It is a figure which shows an example of the judgment criteria of the pass / fail judgment of the prediction value of the dimension data of the product quality prediction model which concerns on one Embodiment of this invention. It is a graph which shows the correspondence relationship between the dimensional data of the product quality prediction model which concerns on one Embodiment of this invention, and an evaluation function. It is a graph which shows the change of the temperature and humidity of the day which concerns on the prior art. It is a graph which shows the change of the dimension of the product which concerns on the prior art.

Hereinafter, one embodiment of the present invention will be described in detail.

(Overview of production environment)
FIG. 1 is a schematic diagram showing a production environment in which the control device (information processing device) 1 according to the present embodiment is used. The control device 1 is, for example, a PC (Personal Computer), a server, or the like. The manufacturing apparatus 2 is an apparatus for manufacturing a product, for example, an injection molding apparatus. Hereinafter, in the embodiment of the present invention, an injection molding apparatus will be described as an example of the manufacturing apparatus 2.

The control device 1 generates a product quality prediction model learned from the quality results obtained from the actual molding conditions, disturbance conditions, and semi-disturbance conditions for the manufacturing device 2.

The molding condition is a condition of a production environment in which injection molding is performed by the manufacturing apparatus 2, and includes, for example, an injection speed, a holding pressure, a holding time, and the like. The molding conditions are conditions that can be controlled by the manufacturing apparatus 2. The disturbance condition is a condition related to disturbance, which is an external action that attempts to disturb the state of the control system, and is a condition that is not subject to control by the manufacturing apparatus 2 among the factors that affect the quality of the molded product. For example, including temperature, humidity and the like.

Semi-disturbance conditions are factors that affect the quality results of molded products, and are conditions that do not always follow the settings even if they are set, for example, conditions that do not immediately follow even if the settings are changed, and quality prediction. This is a factor for which the measured value should be used. For semi-disturbance conditions, measured values that differ from the set values affect the quality results. For example, semi-disturbance conditions include mold temperature movement, mold temperature fixation, and the like. If there is a factor for which the condition of a certain factor is fixed and the optimum condition is to be calculated, it can be calculated by treating it as a semi-disturbance. In addition, the best quality is the best quality obtained by adjusting the production conditions among the disturbance conditions at that time. Optimal conditions are production conditions that give the best quality.

The control device 1 acquires the current disturbance conditions and semi-disturbance conditions, and calculates the optimum molding conditions for the manufacturing device 2 by using the product quality prediction model. The manufacturing apparatus 2 manually or automatically acquires molding conditions from the control device 1, and performs injection molding according to the molding conditions to mold an object to be molded. As a result, a product having the best quality can be manufactured and the quality can be stabilized. The molded product is inspected and then transported to the customer.

(Configuration of control device 1)
FIG. 2 is a block diagram showing the configuration of the control device 1. As shown in FIG. 2, the control device 1 includes an acquisition unit 11, a molding condition calculation unit (out-licensing unit) 12, a molding condition display unit (display data generation unit) 13, a quality score calculation unit (out-licensing unit) 14, and data. It includes an acquisition unit 15, a learning unit 16, a first product quality prediction model (prediction model) MA, and a second product quality model (prediction model) MB.

The acquisition unit 11 acquires a non-input target variable which is a variable other than the input target variable derived by the molding condition calculation unit 12. The molding condition calculation unit 12 derives an input target variable input to the manufacturing apparatus 2 in order to control the manufacturing apparatus 2. Specifically, the molding condition calculation unit 12 derives the input target variable with reference to the non-input target variable acquired by the acquisition unit 11.

Here, the input target variable includes a variable indicating the molding condition (a specified value that defines the manufacturing condition by the manufacturing apparatus 2). The non-input target variable includes at least one of a variable indicating a disturbance condition (a value related to the production environment at the time of manufacturing the manufacturing object) and a variable indicating a semi-disturbing condition (actual measurement value of the manufacturing condition of the manufacturing apparatus 2). ..

The molding condition display unit 13 generates display data for displaying the input target variable derived by the molding condition calculation unit 12, and outputs the display data to a display device (not shown). The display device displays the input target variable derived by the molding condition calculation unit 12 using the input display data.

At least one of the input target variable and the non-input target variable is input to the first product quality prediction model MA and the second product quality prediction model MB, and the predicted value of the quality information regarding the quality of the molded object is output. Specifically, the first product quality prediction model MA and the second product quality prediction model MB are derived from the actually measured values of quality information related to the actual quality and the first product quality prediction model MA and the second product quality prediction model MB. This model is trained so that the difference from the predicted value of quality information becomes smaller.

The first product quality prediction model MA and the second product quality prediction model MB are, for example, trained models using different recursive algorithms. For example, the algorithm used for the first product quality prediction model MA is MLP (MultiLayer Perceptron), and the algorithm used for the second product quality prediction model MB is Elastic Net.

The quality score calculation unit 14 acquires the predicted value of the quality information from the first product quality prediction model MA and the second product quality prediction model MB (referring to the prediction model), calculates the quality score, and calculates the quality score. Output to the molding condition calculation unit 12. The molding condition calculation unit 12 acquires a quality score from the quality score calculation unit 14 and refers to it to derive an input target variable.

The data acquisition unit 15 acquires teacher data in which the input target variable, the non-input target variable, and the quality information are associated with each other. The learning unit 16 trains the first product quality prediction model MA and the second product quality prediction model MB using the teacher data acquired by the data acquisition unit 15.

(Learning process of control device 1)
FIG. 3 is a flowchart showing the learning process of the control device 1. FIG. 4 is a diagram showing an example of teacher data used in the learning process of the control device 1. FIG. 5 is a diagram showing a regression line output by the first product quality prediction model MA and a regression curve output by the second product quality prediction model MB. Hereinafter, the learning process of the control device 1 will be described with reference to FIGS. 3 to 5.

(Step S301)
The control device 1 determines each item to be included in the molding condition, and determines the swing width of the value for each item. Here, in the teacher data shown in FIG. 4, the items included in the molding conditions are injection speed, holding pressure, holding time, cooling time, nozzle temperature, mold temperature movable, and mold temperature fixing. Since the molding condition is a condition that can be adjusted so that the quality result described later becomes a desired value, the molding condition is also referred to as an explanatory variable, and the quality result described later is also referred to as an objective variable. ..

Next, the control device 1 determines the set value to be set in the manufacturing device 2 within the range of the swing width for each item. In this way, the control device 1 determines a plurality of sets of set values for the plurality of items. In FIG. 4, each of the data sets is assigned a data set No.

As shown in FIG. 4, as an example, in the data set No. 1, each set value of the molding conditions has an injection speed of 30 (mm / sec), a holding pressure of 15 (Mpa), and a holding time of 5 (sec). The cooling time is 20 (sec), the nozzle temperature is 210 (° C.), the mold temperature is movable at 50 (° C.), and the mold temperature is fixed at 40 (° C.).

In the process of determining the set value of each item, it is preferable that each set value in the entire plurality of data sets is distributed as evenly as possible.

(Step S302)
The control device 1 transmits each set value included in the molding conditions determined in step S301 to the manufacturing device 2. Then, the manufacturing apparatus 2 acquires each set value from the control device 1, sets the set value as a parameter of its own apparatus, and performs an injection molding process for each data set.

(Step S303: Data acquisition process)
The control device 1 acquires the disturbance conditions and semi-disturbance conditions during the injection molding process using each data set, and the measured values of the quality results after molding, and includes them in the corresponding data sets. Here, in the example shown in FIG. 4, the disturbance conditions include air temperature and humidity, and the semi-turbulence conditions include mold temperature movable and mold temperature fixing. In addition, the measured values of the quality results after molding include the left dimension and the right dimension. In addition, the disturbance condition and the semi-disturbance condition are parameters that can change according to the passage of time in one day and the change of seasons.

In the example shown in FIG. 4, the measured values of the disturbance conditions during the injection molding process using only the molding conditions of the data set No. 1 are a temperature of 22.8 (° C.) and a humidity of 31.6 (%), which are semi-disturbance conditions. As the actually measured values of, the mold temperature is movable 45.6 (° C.) and the mold temperature is fixed 46.8 (° C.), and these values are included in the data set No1. Further, as actual measurement values of the quality result after molding, the left dimension 10.098 (mm) and the right dimension 10.082 (mm) are included in the data set No1.

(Step S304)
The control device 1 converts each data set including the molding condition, the disturbance condition, the semi-disturbance condition, and the quality result into data as teacher data.

(Step S305: Learning process)
The control device 1 performs machine learning using the teacher data converted into data in step S304. Specifically, in the control device 1, the data acquisition unit 15 acquires teacher data. The learning unit 16 trains the first product quality prediction model MA and the second product quality prediction model MB using the teacher data.

There may be three or more product quality prediction models that use the recursive algorithm. As an example, when six regression algorithms are used, for each of the six product quality prediction models, for example, the regression algorithms MLP, ElasticNet, MRA (Multiple Regression Analysis), SVR (Support Vector Regression)- RBF (Radial Basis Function), SVR-Poly, and SVR-Rinear are assigned.

Each product quality prediction model outputs learning result information indicating the learning result. When the regression algorithm is used as described above, the learning result information includes coefficient information for specifying a regression curve which is a relational expression between each parameter.

In the example shown in FIG. 5, when the disturbance condition and the quality result are plotted in two-dimensional coordinates, the relational expression is specified by a straight line graph for the first product quality prediction model MA, and the second product quality prediction model MB The relational expression is specified by the graph of the curve.

(Step S306)
The control device 1 selects a product quality prediction model having a high degree of correlation for each quality item. For example, a square error is used to evaluate the correlation. For example, in the example shown in FIG. 5, it can be said that the second product quality prediction model MB has a higher degree of correlation than the first product quality prediction model MA.

(Molding condition search process of control device 1)
The control device 1 performs a molding condition search process using the first product quality prediction model MA and the second product quality prediction model MB learned as described above.

FIG. 6 is a flowchart showing a molding condition search process of the control device 1. FIG. 7 is a diagram showing an example of molding conditions. Hereinafter, the molding condition search process of the control device 1 will be described with reference to FIGS. 6 and 7.

(Step S601: Acquisition step)
In the control device 1, the acquisition unit 11 acquires the disturbance condition, the semi-disturbance condition, and the quality target value, and outputs the acquisition to the molding condition calculation unit 12. The molding condition calculation unit 12 acquires the disturbance condition, the semi-turbulence condition, and the quality target value from the acquisition unit 11. The quality target value is the target value of the quality score.

(Step S602: Search step (derivation step))
In step S602, the molding condition calculation unit 12 searches for molding conditions with reference to the disturbance condition, the semi-turbulence condition, and the quality target value acquired by the molding condition calculation unit 12. Specifically, the molding condition calculation unit 12 determines a local optimization algorithm (L-BFGS (Broyden Fletcher Goldfarb Shanno) -B, TNC (Truncated Newton CG), COBYLA (L-BFGS (Broyden Fletcher Goldfarb Shanno) -B) for determining hyperparameters of an AI (Artificial Intelligence) model. Constrained Optimization BY Linear Approximation), SLSQP (Sequential Least SQuares Programming optimization algorithm), etc.) are used to derive the optimum molding conditions.

For example, when COBYLA is used, first, the molding condition calculation unit 12 searches for molding conditions up to the vicinity of the optimum solution by a grid search as a rough search. After that, the molding condition calculation unit 12 searches for molding conditions until an optimum solution is obtained by using a method such as a gradient method as a detailed search.

That is, the molding condition calculation unit 12 repeatedly executes the grid search and the local gradient method using the output values of the first product quality prediction model MA and the second product quality prediction model MB, and the disturbance acquired by the acquisition unit 11 Derivation of molding conditions (input target variables) according to conditions and semi-disturbance conditions (non-input target variables). In particular, the molding condition calculation unit 12 derives molding conditions for controlling the manufacturing apparatus 2 that molds (manufactures) the molding target (manufacturing target) by injection molding.

As an example, as shown in FIG. 6, step S602 includes substeps of S6021 to S6025. In the following description, an example in which the grid search is performed in substeps S6021 to S6024 and the gradient method is used in substep S6025 is given, but this is not limited to this embodiment.

(Substep S6021)
In sub-step S6021, the molding condition calculation unit 12 sets a certain molding condition, and sets the certain molding condition and the disturbance condition and the semi-disturbance condition acquired in step S601 to the first product quality prediction model MA and the second product quality. Enter in the prediction model MB. In this case, the molding condition calculation unit 12 inputs the molding condition, the disturbance condition and the semi-disturbance condition into the first product quality prediction model MA and the second product quality prediction model MB, and the quality prediction result is sent to the quality score calculation unit 14. Output.

(Substep S6022)
In sub-step S6022, the quality score calculation unit 14 acquires the quality prediction result corresponding to the molding conditions set in sub-step S6022 from each of the first product quality prediction model MA and the second product quality prediction model MB.

(Substep S6023)
In sub-step S6023, the quality score calculation unit 14 calculates the quality score from the quality prediction result acquired in step S6022, and outputs the quality score to the molding condition calculation unit 12. The quality score is, for example, a combination of appearance quality and dimensional data. The details of the quality score will be described later. The appearance quality is the quality when the molded object is visually observed. The dimensional data is data such as the length, width, and height of the object to be molded.

(Substep S6024)
In sub-step S6024, the molding condition calculation unit 12 determines whether or not to continue the search. When the molding condition calculation unit 12 determines that the search is to be continued (YES in substep S6024), the process returns to substep S6021, a molding condition having a value different from the previously set molding condition is set, and subsequent steps in substep S6021 Continue processing. When the molding condition calculation unit 12 determines that the search is not continued (NO in sub-step S6024), the process proceeds to sub-step S6025.

As an example, the molding condition calculation unit 12 determines that the search is continued when the number of searches is less than the preset number of times or the quality score in the search so far is larger than the quality target value.

(Substep S6025)
In the sub-step S6025, the molding condition calculation unit 12 identifies the molding condition corresponding to the minimum quality score by referring to the quality score obtained by the search using the sub-steps S6021 to S6024.

As an example, when a search using a grid search is performed using substeps S6021 to S6024, in this step, a more detailed search is performed by the gradient method with reference to the information obtained by the grid search. , It can be configured to specify the molding conditions corresponding to the minimum quality score.

(Step S603: Display process)
The molding condition calculation unit 12 outputs the molding condition specified in step S602 to the molding condition display unit 13. The molding condition display unit 13 acquires molding conditions from the molding condition calculation unit 12, generates display data for displaying the molding conditions, and causes the display device to display the display data. For example, the display device displays the molding conditions as shown in FIG. 7.

The control device 1 displays the molding conditions on the display device and transmits the molding conditions to the manufacturing device 2. Further, the control device 1 may transmit the molding conditions to the manufacturing device 2 without displaying the molding conditions on the display device.

<Calculation method of quality score>
Hereinafter, the method of calculating the quality score of the first product quality prediction model MA in the quality score calculation unit 14 will be described with reference to FIGS. 8 to 11. This calculation method can also be applied to the second product quality prediction model MB.

FIG. 8 is a diagram showing an example of a determination criterion for pass / fail determination of the predicted value of the appearance quality by the first product quality prediction model MA. The determination standard shown in FIG. 8 is a determination standard stored in a storage unit (not shown) included in the control device 1, and is referred to by the quality score calculation unit 14 for determining the pass / fail of the predicted value of the appearance quality. It is a figure which shows an example of a judgment standard. As shown in FIG. 8, the appearance quality is defined by a predetermined quality quantitative value. The quality score calculation unit 14 determines that the appearance quality of the first product quality prediction model MA is acceptable when the quality quantitative value of the appearance quality is less than 5, and when the quality quantitative value of the appearance quality is larger than 5, the first 1 It is determined that the appearance quality of the product quality prediction model MA is unacceptable.

FIG. 9 is a graph showing the correspondence between the quality quantitative value of the appearance quality and the evaluation function. The graph shown in FIG. 9 is a graph that explicitly shows the correspondence between the quality quantitative value of the appearance quality stored in the storage unit included in the control device 1 and the evaluation function, and the quality score calculation unit 14 indicates the appearance quality. It is a graph which shows an example of the judgment criteria which is referred to for calculating the evaluation function of. As shown in FIG. 9, the evaluation function is set so that the quantitative quality value and the evaluation function have a proportional correlation with a positive slope in the pass range of the appearance quality. Further, as shown in FIG. 9, at the boundary between the pass range and the fail range of the appearance quality, the evaluation function in the fail range has a very large value by adding an offset of, for example, +100 to the evaluation function. Is set to. Further, the quality score calculation unit 14 is set so that the evaluation function has a linear correlation with a positive slope between the quality quantitative value and the evaluation function in the fail range of the appearance quality.

The quality score calculation unit 14 calculates the appearance quality evaluation function from the predicted value of the appearance quality included in the first product quality prediction model MA by referring to the graph shown in FIG. Here, as described above, the quality score calculation unit 14 sets the evaluation function in the fail range to be significantly larger than the evaluation function in the pass range, so that the quality quantitative value is rejected in the molding condition calculation unit 12. It is possible to prevent the molding conditions of the first product quality prediction model MA, which has become, from being calculated.

FIG. 10 is a diagram showing an example of a determination criterion for pass / fail determination of the predicted value of the dimensional data by the first product quality prediction model MA. The judgment standard shown in FIG. 10 is a judgment standard stored in a storage unit included in the control device 1, and is an example of a judgment standard referred to by the quality score calculation unit 14 for determining the pass / fail of the predicted value of the dimensional data. It is a figure which shows. As shown in FIG. 10, as an example, when the dimensional data is 9 mm or more and 13 mm or less, the quality score calculation unit 14 determines that the dimensional data of the first product quality prediction model MA is acceptable, and the dimensional data is 9 mm or more. If it is smaller or larger than 13 mm, it is determined that the dimensional data of the first product quality prediction model MA is rejected.

FIG. 11 is a graph showing the correspondence between the dimensional data by the first product quality prediction model MA and the evaluation function. As shown in FIG. 11, the evaluation function is set so that the evaluation function at 11 mm is the minimum value and the dimensional data is centered at 11 mm and the evaluation function becomes larger as the distance from the center value increases in the pass range of the dimensional data. ing. Specifically, in the range where the dimensional data is 11 mm or more and 13 mm or less, the quality quantitative value and the evaluation function are set to have a proportional correlation with a positive slope. Further, in the range where the dimensional data is 9 mm or more and 11 mm or less, the quality quantitative value and the evaluation function are set to have a proportional correlation with a negative slope.

Further, as shown in FIG. 11, at the boundary between the pass range and the fail range of the dimensional data, for example, by adding an offset of +100 to the evaluation function, the evaluation function in the fail range has a very large value. It is set.

Further, as shown in FIG. 11, the evaluation function is set to increase as the distance from the center value increases in the reject range of the dimensional data. Specifically, in the range where the dimensional data is larger than 13 mm, the quality quantitative value and the evaluation function are set to have a proportional correlation with a positive slope. Further, in the range where the dimensional data is smaller than 9 mm, the quality quantitative value and the evaluation function are set to have a proportional correlation with a negative slope.

The quality score calculation unit 14 calculates the evaluation function of the dimensional data from the predicted value of the dimensional data included in the first product quality prediction model MA by referring to the graph shown in FIG. Here, as described above, the quality score calculation unit 14 sets the evaluation function in the fail range to be significantly larger than the evaluation function in the pass range, so that the dimensional data is rejected in the molding condition calculation unit 12. It is possible to prevent the molding conditions of the first product quality prediction model MA from being calculated.

The quality score calculation unit 14 calculates the total value of the appearance quality evaluation function and the dimensional data evaluation function as the quality score of the first product quality prediction model MA. In the present embodiment, the quality score calculation unit 14 has described a method of calculating the quality score of the first product quality prediction model MA with reference to the appearance quality and the dimensional data, but the present embodiment is limited to this. However, for example, the quality score calculation unit 14 calculates at least one evaluation function such as appearance quality, dimensional data, roughness, and hardness, and calculates the quality score of the first product quality prediction model MA. May be good.

[Example of realization by software]
The control block of the control device 1 (particularly, the acquisition unit 11, the molding condition calculation unit 12, the molding condition display unit 13, the quality score calculation unit 14, the data acquisition unit 15, and the learning unit 16) is connected to an integrated circuit (IC chip) or the like. It may be realized by the formed logic circuit (hardware) or by software.

In the latter case, the control device 1 includes a computer that executes instructions of a program that is software that realizes each function. The computer includes, for example, one or more processors and a computer-readable recording medium that stores the program. Then, in the computer, the processor reads the program from the recording medium and executes it, thereby achieving the object of the present invention. As the processor, for example, a CPU (Central Processing Unit) can be used. As the recording medium, a "non-temporary tangible medium", for example, a ROM (Read Only Memory) or the like, a tape, a disk, a card, a semiconductor memory, a programmable logic circuit, or the like can be used. Further, a RAM (Random Access Memory) for expanding the above program may be further provided. Further, the program may be supplied to the computer via an arbitrary transmission medium (communication network, broadcast wave, etc.) capable of transmitting the program. It should be noted that one aspect of the present invention can also be realized in the form of a data signal embedded in a carrier wave, in which the above program is embodied by electronic transmission.

The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the claims, and the embodiments obtained by appropriately combining the technical means disclosed in the different embodiments. Is also included in the technical scope of the present invention.

1 Control device 2 Manufacturing device 11 Acquisition section 12 Molding condition calculation section (leading section)
13 Molding condition display unit 14 Quality score calculation unit (deriving unit)
15 Data acquisition department 16 Learning department MA 1st product quality prediction model MB 2nd product quality prediction model

Claims (14)

A derivation unit that derives input target variables to be input to the manufacturing equipment to control the manufacturing equipment,
An acquisition unit that acquires a non-input target variable that is a variable other than the input target variable derived by the derivation unit,
With
The information processing device is characterized in that the derivation unit derives the input target variable with reference to the non-input target variable acquired by the acquisition unit. The information processing apparatus according to claim 1, wherein the non-input target variable includes at least one of a value relating to a production environment at the time of manufacturing the manufacturing object and a measured value of the manufacturing conditions of the manufacturing apparatus. The information processing apparatus according to claim 1 or 2, wherein the input target variable includes a specified value that defines manufacturing conditions by the manufacturing apparatus. The derivation unit derives the input target variable by referring to a prediction model in which at least one of the input target variable and the non-input target variable is input and outputs a predicted value of quality information regarding the quality of the manufacturing target. , The information processing apparatus according to any one of claims 1 to 3. The information according to claim 4, wherein the prediction model is a model trained so that the difference between the measured value of the quality information regarding the actual quality and the predicted value of the quality information derived from the prediction model becomes smaller. Processing equipment. A data acquisition unit that acquires teacher data in which the input target variable, the non-input target variable, and the quality information are associated with each other.
A learning unit that trains the prediction model using the teacher data acquired by the data acquisition unit, and
The information processing apparatus according to claim 4 or 5, further comprising. 4. The derivation unit repeatedly executes a grid search and a local gradient method using the output value of the prediction model to derive an input target variable corresponding to the non-input target variable acquired by the acquisition unit. The information processing apparatus according to any one of Items to 6. The information processing device according to any one of claims 1 to 7, wherein the derivation unit derives an input target variable for controlling a manufacturing device that manufactures a manufacturing object by injection molding. The information processing apparatus according to any one of claims 1 to 8, further comprising a display data generation unit that generates display data for displaying the derived input target variable. Teacher data in which input target variables input to the manufacturing device to control the manufacturing device, non-input target variables that are variables other than the input target variables, and quality information regarding the quality of the manufacturing device are associated with each other. Data acquisition unit to acquire
Learning to learn a prediction model in which at least one of the input target variable and the non-input target variable is input and outputs a predicted value of quality information regarding the quality of the manufacturing object using the teacher data acquired by the data acquisition unit. Department and
An information processing device characterized by being equipped with. An acquisition process for acquiring a non-input target variable, which is a variable other than the input target variable input to the manufacturing device in order to control the manufacturing device,
The derivation process for deriving the input target variable and
Including
An information processing method characterized in that, in the derivation step, the input target variable is derived with reference to the non-input target variable acquired in the acquisition step. Teacher data in which input target variables input to the manufacturing device to control the manufacturing device, non-input target variables that are variables other than the input target variables, and quality information regarding the quality of the manufacturing device are associated with each other. Data acquisition process to acquire
Learning to learn a prediction model in which at least one of the input target variable and the non-input target variable is input and outputs a predicted value of quality information regarding the quality of the manufacturing object using the teacher data acquired in the data acquisition process. Process and
An information processing method characterized by containing. The program for operating a computer as the information processing device according to claim 1, wherein the computer functions as the derivation unit and the acquisition unit. The program for operating a computer as the information processing device according to claim 10, wherein the computer functions as the data acquisition unit and the learning unit.

Images