WO2024111172A1 - Molded article quality variance estimation device, molded article quality variance estimation method, and injection molding system - Google Patents

Abstract

According to the present invention, quality variance of molded articles when using a material having inconsistent physical properties is estimated.　This molded article quality variance estimation device comprises: a physical property calculation model estimation unit that estimates physical property calculation models corresponding to a second material whose physical properties are unknown on the basis of process data when a molded article was manufactured using a first material whose physical properties are known and a learned regression model generated on the basis of a physical property calculation model coefficient corresponding to the first material, as well as process data when a molded article was manufactured using the second material; a flow analysis unit that performs a flow analysis on the basis of the estimated physical property calculation models corresponding to the second material, and estimates the quality variance of molded articles using the second material. The physical property calculation model estimation unit estimates two or more physical property calculation models, each corresponding to two or more differing physical properties. The flow analysis unit performs the flow analysis on the basis of the estimated two or more physical property calculation models.

Description

Apparatus and method for estimating quality variation of molded product, and injection molding system

The present invention relates to a molded product quality variation estimation device, a molded product quality variation estimation method, and an injection molding system. The present invention claims priority to Japanese Patent Application No. 2022-187609 filed on November 24, 2022, and for designated countries where incorporation by reference to literature is permitted, the contents of that application are incorporated by reference into this application.

In the field of injection molding, in response to societal demands for resource circulation, there is a push to switch from virgin synthetic resins and other materials to recycled materials, which have a lower environmental impact. However, because recycled materials are made from a variety of plastic products with different thermal histories, even if they are shipped from recycled material suppliers with the same product number, their physical properties are not necessarily the same or stable when comparing products from different shipping lots.

Instability in the physical properties of a material can affect the quality of a molded product. In particular, it is known that the viscosity and pressure-temperature dependence of specific volume (hereafter referred to as PVT characteristics) of the material's physical properties can cause dimensional variation in molded products.

When a new material with different physical properties such as viscosity or PVT characteristics is used for injection molding, flow analysis is performed using CAE (Computer Aided Engineering) to confirm moldability and assist in adjusting molding conditions. It is common to use flow analysis in the design and material selection of plastic parts. However, flow analysis requires the creation of a physical property calculation model based on the physical property data of the material, and creating a physical property calculation model for each lot of recycled material would be a huge amount of work, making it unrealistic.

Regarding flow analysis, for example, Patent Document 1 describes a "flow analysis device that performs flow analysis of molten material in the injection molding process, characterized in that it has a material property data calculation means that calculates viscosity data as material property data based on data obtained from an injection molding machine, and a flow analysis means that performs flow analysis based on an analytical shape model, the material property data, and molding condition data."

JP 2003-145577 A

In the case of the flow analysis device described in Patent Document 1, flow analysis is performed using only viscosity as the physical property of the material, so the analysis precision is lower than when flow analysis is performed using multiple physical properties. Therefore, for example, it is not possible to properly estimate the quality variation of molded products when multiple materials with the same product number but different lots are used.

The present invention was made in consideration of the above points, and aims to improve the accuracy of flow analysis when using materials with unstable physical properties, and to make it possible to estimate the quality variation of molded products.

The present application includes multiple means for solving at least some of the above problems, examples of which are as follows:

In order to solve the above problem, a molded product quality variation estimation device according to one embodiment of the present invention is a molded product quality variation estimation device that estimates the quality variation of a molded product produced by injection molding, and includes a physical property calculation model estimation unit that estimates a physical property calculation model corresponding to a second material based on process data when a molded product is manufactured using a first material whose physical properties are known, a learned regression model generated based on physical property calculation model coefficients corresponding to the first material, and process data when a molded product is manufactured using a second material whose physical properties are unknown, and a flow analysis unit that estimates the quality variation of the molded product when a molded product is manufactured using the second material by performing flow analysis based on the estimated physical property calculation model corresponding to the second material, where the physical property calculation model estimation unit estimates two or more types of physical property calculation models corresponding to two or more different types of physical properties based on two or more types of learned regression models corresponding to two or more different types of physical properties, and the flow analysis unit performs the flow analysis based on the estimated two or more types of physical property calculation models.

The present invention makes it possible to improve the accuracy of flow analysis when using materials with unstable physical properties, and to estimate the quality variation of molded products.

　Problems, configurations, and advantages other than those mentioned above will become clear from the description of the embodiments below.

FIG. 1 shows examples of the physical properties of multiple recycled materials having the same product number but different lots, where FIG. 1(A) is a graph showing the PVT characteristics and FIG. 1(B) is a graph showing the viscosity. FIG. 2 is a diagram showing an example of an injection molding system according to an embodiment of the present invention. FIG. 3 is a diagram showing an example of the functional block configuration of the molded product quality variation estimation device. FIG. 4 is a diagram illustrating an example of the sensor information. FIG. 5 is a diagram illustrating an example of a feature amount DB (database). FIG. 6 is a diagram illustrating an example of the quality DB. 7A and 7B are diagrams showing examples of selected learning data sets, where FIG. 7A is a diagram showing a learning data set for PVT characteristics, and FIG. 7B is a diagram showing a learning data set for viscosity. FIG. 8 is a diagram illustrating an example of the configuration of the PVT characteristic calculation model estimating unit and the viscosity calculation model estimating unit. FIG. 9 is a diagram showing an example of the configuration of a general computer. FIG. 10 is a cross-sectional view showing an example of the configuration of an injection molding machine. Figure 11 shows an example of a mold, where Figure 11(A) is a top view of the product part of the mold, Figure 11(B) is a side view of the product part of the mold, Figure 11(C) is a top view of the runner part of the mold, and Figure 11(D) is a side view of the runner part of the mold. FIG. 12 is a flowchart illustrating an example of the molding process data acquisition process. 13A and 13B are diagrams for explaining the effectiveness of an estimated physical property calculation model, in which FIG. 13A is a diagram showing an example of an estimated PVT characteristic calculation model, and FIG. 13B is a diagram showing an example of an estimated viscosity calculation model. Figure 14 is a diagram showing an example of a UI (User Interface) screen.

Below, an embodiment of the present invention will be described with reference to the drawings. The embodiment is an example for explaining the present invention, and appropriate omissions and simplifications have been made to clarify the explanation. The present invention can be implemented in various other forms. Unless otherwise specified, each component may be singular or plural. The position, size, shape, range, etc. of each component shown in the drawings may not represent the actual position, size, shape, range, etc. in order to facilitate understanding of the invention. For this reason, the present invention is not necessarily limited to the position, size, shape, range, etc. disclosed in the drawings. As examples of various information, descriptions may be given using expressions such as "table," "list," and "queue," but various information may be expressed using data structures other than these. For example, various information such as "XX table," "XX list," and "XX queue" may be expressed as "XX information." When explaining identification information, expressions such as "identification information," "identifier," "name," "ID," and "number" are used, but these are interchangeable. In all drawings for explaining the embodiment, the same components are generally given the same reference numerals, and repeated explanations are omitted. In the following embodiments, the components (including element steps, etc.) are not necessarily essential, unless otherwise specified or considered to be clearly essential in principle. Furthermore, when it is said that "consists of A," "is made of A," "has A," or "includes A," other elements are not excluded, unless otherwise specified to mean only that element. Similarly, in the following embodiments, when referring to the shape, positional relationship, etc. of components, etc., it includes things that are substantially similar or similar to that shape, etc., unless otherwise specified or considered to be clearly not essential in principle.

<Physical properties of recycled materials>
Fig. 1 shows an example of the physical properties of multiple recycled materials that are used as materials for molded products. The multiple recycled materials are recycled materials A, B, and C that have the same product number but different shipping lots (e.g., different delivery times).

FIG. 1A shows the PVT characteristics of recycled materials A, B, and C, with the horizontal axis representing the temperature of the material [°C] and the vertical axis representing the specific volume of the material [ ^cm3 /g]. The pressure of the material is a constant value of 50 [MPa]. The dot plots shown in FIG. 1A are actual measurements obtained using a dedicated measuring device. The line plots show the results of fitting a known calculation model called the 2-domain Tait PVT model shown in the following formula (1) to the actual measurements.

Figure (B) shows the viscosity of recycled materials A, B, and C, with the horizontal axis representing the shear rate of the material [1/sec] and the vertical axis representing the viscosity of the material [Pa·s]. The temperature of the material is kept constant at 200°C. The dot plots shown in Figure (B) are actual measurements obtained using a dedicated measuring device. The line plots show the results of fitting a known calculation model called the Cross-WLF model, shown in the following equation (2), to the actual measurements.

As is clear from Figure 1, in the case of recycled materials, even if the product number is the same, if the lot is different, the physical properties (PVT characteristics and viscosity in the figure) are not stable.

<Configuration example of injection molding system 10 according to an embodiment of the present invention>
2 shows an example of the configuration of an injection molding system 10 according to an embodiment of the present invention. The injection molding system 10 estimates a PVT characteristic calculation model and a viscosity calculation model corresponding to a material with unstable physical properties, i.e., a material with unknown physical properties, and performs flow analysis using the estimated PVT characteristic calculation model and viscosity calculation model, thereby quantifying the variation in quality of a molded product when a material with unstable physical properties is used.

The injection molding system 10 includes a manufacturing control device 20, factory equipment 30, and a molded product quality variation estimation device 40.

The manufacturing control device 20 instructs the injection molding machine 31 included in the factory equipment 30 to manufacture products (molded products) by injection molding. The manufacturing control device 20 includes a manufacturing condition determination unit 21 and a production instruction unit 22.

The manufacturing condition determination unit 21 determines manufacturing conditions based on standard injection molding conditions (hereinafter referred to as standard molding conditions) previously set by the user, and outputs them to the production instruction unit 22. The standard molding conditions include parameters corresponding to each step of the injection molding process (described in detail later).

For example, the reference molding conditions include parameters such as the metering position, suck back, back pressure, back pressure speed, and rotation speed as parameters for the metering and plasticizing process. The reference molding conditions also include parameters for the injection process and the pressure holding process such as material pressure, material temperature (temperature of the heater that heats the material), mold temperature (temperature and flow rate of the coolant that cools the mold), pressure holding time, and shear rate. The reference molding conditions also include parameters for the injection process and the pressure holding process such as the screw position (VP switching position) that switches between injection and pressure, and the clamping force of the mold 509. The reference molding conditions also include the cooling time after pressure holding as a parameter for the cooling process.

The manufacturing conditions include, in addition to the standard molding conditions, information specifying the injection molding machine 31, information specifying the mold, the capacity of the mold, the configuration of the runner part of the mold, information specifying the material (product number, lot number), the number of molded products to be produced, the production time, the required quality, etc.

The production instruction unit 22 instructs the injection molding machine 31 to manufacture a product (molded product) by outputting control information based on the manufacturing conditions. The manufacturing conditions are also supplied to the molded product quality variation estimation device 40 via the injection molding machine 31. The manufacturing conditions supplied to the molded product quality variation estimation device 40 are referenced when estimating the PVT characteristic calculation model and the viscosity calculation model for each material product number and lot number.

The factory equipment 30 includes an injection molding machine 31 and a quality inspection device 33.

The injection molding machine 31 executes the injection molding process according to the control information from the production instruction unit 22, and outputs the manufactured molded product to the quality inspection device 33. The injection molding machine 31 has an in-mold sensor 32. The in-mold sensor 32 is composed of a plurality of various sensors. The in-mold sensor 32 includes, for example, at least a temperature sensor for measuring the material temperature, a pressure sensor for measuring the material pressure, and a temperature sensor for measuring the mold temperature. The in-mold sensor 32 measures physical quantities (temperature, pressure, etc.) in the injection molding process as in-mold sensor values, and outputs the time series data of the measurement results as sensor information to the molded product quality variation estimation device 40. The in-mold sensor 32 may measure, for example, the shear rate of the material, the physical properties of the material, the opening amount of the mold (mold opening amount), etc. The physical properties of the material refer to, for example, the density of the material, the viscosity of the material, the fiber length distribution of the material (when the material contains reinforcing fibers), etc. Of these, the physical quantity most correlated with the fluidity of a material is the viscosity of the material, but other quantities correlated with fluidity calculated from pressure, temperature, speed, etc. can also be used as physical quantities correlated with the fluidity of a material.

The quality inspection device 33 inspects the quality of the molded product. For example, the quality inspection device 33 measures the weight and dimensions as the quality of the molded product. Inspection items by the quality inspection device 33 may include, for example, shape characteristics (thickness, sink marks, burrs, warping, etc.), surface characteristics such as appearance defects (welds, silver, burns, whitening, scratches, bubbles, peeling, flow marks, jetting, color, gloss, etc.), mechanical and optical characteristics (tensile strength, impact resistance, transmittance, etc.), etc. Also, instead of or in addition to the quality inspection device 33, an inspector (manually) may inspect the quality of the molded product. The quality inspection device 33 links the manufacturing conditions at the time of manufacturing the molded product with the quality inspection results of the molded product, and outputs them to the molded product quality variation estimation device 40.

The molded product quality variation estimation device 40 includes a property calculation model generation unit 41, a flow analysis unit 42, an input unit 43, a display unit 44, and a communication unit 45.

The physical property calculation model generation unit 41 generates a physical property calculation model (specifically, a PVT characteristic calculation model and a viscosity calculation model) corresponding to a material with unknown physical properties based on the manufacturing conditions from the manufacturing control device 20, the sensor information input from the production factory equipment 30, and the quality inspection results. Note that the physical property calculation model generation unit 41 may also generate physical property calculation models for other physical properties (e.g., degree of crystallinity, crystallization rate, etc.).

Generally, a computational model refers to a mathematical model that mathematically abstracts a certain phenomenon. A PVT characteristic computation model refers to a mathematical model that represents the PVT characteristics of a material. A viscosity computation model refers to a mathematical model that represents the viscosity of a material. Note that in this embodiment, the PVT characteristic computation model and the viscosity computation model refer to computational models in which the parameters (coefficients) of the mathematical model are determined for each specified unit of material (e.g., for each lot).

The flow analysis unit 42 uses the PVT characteristic calculation model and viscosity calculation model corresponding to the material with unknown physical properties estimated by the physical property calculation model generation unit 41 to perform flow analysis on the material with unknown physical properties, and estimates and outputs the variation in quality of the molded product when a material with unknown physical properties is used.

The input unit 43 accepts various inputs from the user. The display unit 44 displays the UI screen 1000 (FIG. 14). The communication unit 45 connects the factory equipment 30 via a network (not shown) such as the Internet, and communicates various information.

Next, FIG. 3 shows an example of the functional block configuration of the physical property calculation model generation unit 41 and the flow analysis unit 42.

The physical property calculation model generation unit 41 includes a sensor information storage unit 51, a feature extraction unit 52, a feature DB 53, a quality DB 54, a PVT characteristic calculation model coefficient DB 55, a viscosity calculation model coefficient DB 56, a linking processing unit 57, a learning DB 58, a learning data selection unit 59, a PVT characteristic calculation model estimation unit 60, and a viscosity calculation model estimation unit 61.

The sensor information storage unit 51 temporarily stores the sensor information. The feature extraction unit 52 reads the sensor information temporarily stored by the sensor information storage unit 51 and extracts the feature of the physical quantity contained in the sensor information.

Figure 4 shows an example of sensor information. The top part of the figure shows the time series change in material pressure in the injection molding process, and the bottom part shows the time series change in material temperature in the injection molding process. Note that lots A, B, and C in the figure refer to materials A, B, and C that have the same product number but different lots. The same applies to the subsequent figures.

For example, the feature extraction unit 52 calculates the integral value of the material pressure during the injection process (from the start of injection to the time when the material pressure reaches its peak value) in the injection molding process as the feature value. The feature extraction unit 52 also detects the peak value of the material temperature during the injection process as the feature value. Note that, as the feature value, for example, the peak value of the physical quantity contained in the sensor information may be detected, or the integral value of the physical quantity from the injection process to the cooling process or the maximum differential value of the physical quantity may be calculated. The physical quantity may also be used as the feature value directly. The feature extraction unit 52 records the extracted feature value in the feature DB 53. The feature values extracted by the feature extraction unit 52 and the quality inspection results of the molded product by the quality inspection device 33 correspond to the process data of the present invention.

Figure 5 shows an example of feature DB 53. In feature DB 53, the features of the sensor information, the combination of the injection molding machine 31 and the mold, and a specified material unit (lot) are linked and recorded. A data group that links the specified material unit, the combination of the injection molding machine 31 and the mold, and the extracted features recorded in feature DB 53 is called a feature data set.

Here, when the combination of the injection molding machine 31 and the mold is fixed and the reference molding conditions in the injection molding process are fixed, and only a specific material unit of the material (e.g., only the lot) is changed, the feature values extracted from the physical quantities obtained from the in-mold sensor 32 are strongly affected by the change in material information (e.g., fluidity and physical property values) between the material units. For this reason, it becomes possible to record the material information specific to each material unit as the difference between the feature values.

The feature quantities extracted from the physical quantities obtained from the in-mold sensor 32 are affected by differences between the injection molding machine 31 and the mold, so the feature quantity data set stores material information specific to each material unit for each combination of the injection molding machine 31 and the mold as differences between the feature quantities. In this embodiment, the feature quantity data set is based on the premise that all combinations of the injection molding machine 31 and the mold are fixed.

Returning to Figure 3, the quality inspection results input from the quality inspection device 33 are recorded in the quality DB 54.

FIG. 6 shows an example of quality DB 54. Quality DB 54 records quality inspection results (weight and dimensions in this figure) input from quality inspection device 33, the combination of injection molding machine 31 and mold, and a specified material unit (lot) in association with each other.

Returning to Figure 3, the PVT characteristic calculation model coefficient DB55 stores each coefficient of the PVT characteristic calculation model constructed based on the PVT characteristics of a material of a specific material unit whose physical properties are known, in association with information on the specific material unit. For the PVT characteristic calculation model, calculation models such as the 2-domain-tait model, the Spencer-Glimore model, the Modified-Cell model, and the Simha-Somcynsky model can be used.

In the viscosity calculation model coefficient DB56, each coefficient of a viscosity calculation model constructed based on the viscosity of a material of a predetermined material unit whose physical properties are known is stored in association with information of the predetermined material unit. In this embodiment, the viscosity calculation model may be, for example, a calculation model such as the Cross-WLF model.

The linking processing unit 57 associates the feature data set linked to the material information acquired from the feature DB 53, the molded product quality and manufacturing conditions acquired from the quality DB 54, the coefficients of the PVT characteristic calculation model acquired from the PVT characteristic calculation model coefficient DB 55, and the coefficients of the viscosity calculation model acquired from the viscosity calculation model coefficient DB 56, using the material information as a linking key.The linking processing unit 57 then records the associated feature data set, molded product quality, the coefficients of the PVT characteristic calculation model, the coefficients of the viscosity calculation model, and the materials used in manufacturing in the learning DB 58 as a learning data set.

The learning data selection unit 59 refers to the learning DB 58 and selects one or more learning data sets corresponding to a material (material with unknown physical properties) designated by the user as the analysis target and a material (material with known physical properties) having the same product number but a different lot. The more learning data sets selected here, the more the learning accuracy of the regression model described below can be improved. The learning data selection unit 59 then extracts values that are highly correlated with the PVT characteristics (described in detail below) from the selected learning data sets as PVT characteristic learning data sets, and stores them in the learning DB 601 (FIG. 8) of the PVT characteristic calculation model estimation unit 60. The learning data selection unit 59 also extracts values that are highly correlated with the viscosity (described in detail below) from the selected learning data sets as viscosity learning data sets, and stores them in the learning DB 611 (FIG. 8) of the viscosity calculation model estimation unit 61. The learning data selection unit 59 corresponds to the selection unit of the present invention.

FIG. 7 shows an example of a learning data set extracted by the learning data selection unit 59, where (A) shows an example of a PVT characteristic learning data set, and (B) shows an example of a viscosity learning data set.

As shown in the figure (A), the explanatory variables used in training the PVT characteristic calculation model are the characteristic value of material pressure, which is highly correlated with PVT characteristics, the feature value of material temperature, and weight and dimensions as molded product quality. As shown in the figure (B), the explanatory variables used in training the viscosity calculation model are the characteristic value of material pressure, which is highly correlated with viscosity.

Returning to Figure 3, the PVT characteristic calculation model estimation unit 60 estimates a PVT characteristic calculation model corresponding to a material with unknown physical properties. The viscosity calculation model estimation unit 61 estimates a viscosity calculation model corresponding to a material with unknown physical properties.

Next, FIG. 8 shows an example configuration of the PVT characteristic calculation model estimation unit 60 and the viscosity calculation model estimation unit 61.

The PVT characteristic calculation model estimation unit 60 includes a learning DB 601, an estimation DB 602, a regression model learning unit 603, a learned regression model DB 604, a physical property calculation model estimation unit 605, and a PVT characteristic calculation model DB 606.

The learning DB 601 stores a PVT characteristic learning data set used to generate a PVT characteristic calculation model selected by the learning data selection unit 59. The estimation DB 602 stores an estimation data set (feature data set, molded product quality) corresponding to a material whose physical properties are unknown.

The regression model learning unit 603 reads the PVT characteristic learning dataset from the learning DB 601, and specifies the feature dataset and molded product quality included in the PVT characteristic learning dataset as explanatory variables of the regression model. The regression model learning unit 603 also specifies the PVT characteristic calculation model coefficients as objective variables of the regression model, machine-learns a regression model that predicts the objective variable from the explanatory variables, generates a learned regression model, and stores it in the learned regression model DB 604.

Generally, a regression model refers to a model that predicts a response variable y from an explanatory variable x (y = f(x)). The parameters in the regression model are determined by training data. In this embodiment, "regression model" refers to regression models in general, and "trained regression model" refers to a regression model whose parameters are determined by a training dataset.

In this embodiment, the regression model may be, for example, linear regression, ridge regression, support vector machine, neural network, random forest regression, or a regression model that combines these. Furthermore, when a regression model that can have multiple objective variables, such as a neural network, is used, multiple coefficients may be selected as objective variables from among the coefficients of the PVT characteristic calculation model.

The physical property calculation model estimation unit 605 estimates a PVT characteristic calculation model for a material whose PVT characteristics are unknown. Specifically, the physical property calculation model estimation unit 605 inputs the estimation data set read from the estimation DB 602 into a learned regression model read from the learned regression model DB 604 (for example, a learned regression model corresponding to a material whose product number is the same as that of a material of a specific material unit whose PVT characteristics are unknown, but whose specific material unit (for example, lot) is different) and obtains coefficients of the PVT characteristic calculation model output from the learned regression model, thereby estimating a PVT characteristic calculation model for a material whose PVT characteristics are unknown. In addition, the physical property calculation model estimation unit 605 stores the estimated PVT characteristic calculation model (hereinafter referred to as the estimated PVT characteristic calculation model) in the PVT characteristic calculation model DB 606.

The PVT characteristic calculation model DB 606 stores the known PVT characteristic calculation models stored in the PVT characteristic calculation model coefficient DB 55 and the estimated PVT characteristic calculation models estimated by the physical property calculation model estimation unit 605 as data linked to the corresponding material units and information indicating whether the PVT characteristic calculation model is known or estimated.

Similarly, the viscosity calculation model estimation unit 61 includes a learning DB 611, an estimation DB 612, a regression model learning unit 613, a learned regression model DB 614, a physical property calculation model estimation unit 615, and a viscosity calculation model DB 616.

The learning DB 611 stores a viscosity learning data set used to generate a viscosity calculation model selected by the learning data selection unit 59. The estimation DB 612 stores an estimation data set (feature data set, molded product quality) corresponding to a material whose physical properties are unknown.

The regression model learning unit 613 reads out the viscosity learning dataset from the learning DB 611, and specifies the feature dataset included in the viscosity learning dataset as the explanatory variables of the regression model. The regression model learning unit 613 also specifies the viscosity calculation model coefficients as the objective variables of the regression model, machine-learns a regression model that predicts the objective variables from the explanatory variables, generates a learned regression model, and stores it in the learned regression model DB 614.

The physical property calculation model estimation unit 615 estimates a viscosity calculation model for a material whose viscosity is unknown. Specifically, the physical property calculation model estimation unit 615 inputs the estimation data set read from the estimation DB 612 into a learned regression model read from the learned regression model DB 614 (for example, a learned regression model corresponding to a material whose product number is the same as that of a material whose viscosity is unknown and whose predetermined material unit (for example, lot) is different) and obtains coefficients of the viscosity calculation model output from the learned regression model, thereby estimating a viscosity calculation model for a material whose viscosity is unknown and whose predetermined material unit. In addition, the physical property calculation model estimation unit 615 stores the estimated viscosity calculation model (hereinafter referred to as the estimated viscosity calculation model) in the viscosity calculation model DB 616.

The viscosity calculation model DB 616 stores the known viscosity calculation models stored in the viscosity calculation model coefficient DB 56 and the estimated viscosity calculation models estimated by the physical property calculation model estimation unit 615 as data linked to the corresponding material units and information indicating whether the viscosity calculation model is known or estimated.

Returning to FIG. 3, the flow analysis unit 42 includes an analysis property calculation model DB 71, an analysis property model reading unit 72, an analysis unit 73, and an analysis result output unit 74.

The analytical property calculation model DB71 stores the estimated PVT characteristic calculation model and the estimated viscosity calculation model generated by the property calculation model generation unit 41. The analytical property model readout unit 72 reads out the coefficients of the estimated PVT characteristic calculation model and the estimated viscosity calculation model from the analytical property calculation model DB71 and outputs them to the analysis unit 73. The analysis unit 73 performs flow analysis on a material with unknown physical properties using the coefficients of the estimated PVT characteristic calculation model and the coefficients of the estimated viscosity calculation model according to the analysis shape and analysis conditions specified by the user, and estimates, for example, the quality (weight, dimensions, etc.) of the molded product as the analysis result and outputs it to the analysis result output unit 74.

Here, the analysis shape specified by the user refers to the shape of the mold. The analysis shape (mold shape) may be different from the mold shape when the regression model was trained by machine learning. Furthermore, the analysis conditions specified by the user may be different from the reference molding conditions when the regression model was trained by machine learning.

The analysis result output unit 74 compiles the analysis results (quality of molded products) by the analysis unit 73 for each lot, and calculates, for example, the difference between the maximum and minimum values of the values representing quality (weight, dimensions, etc.) as the quality variation. Note that instead of or in addition to the difference between the maximum and minimum values of the values representing quality, for example, the variance, standard deviation, etc. of the values representing quality may be calculated. The analysis result output unit 74 also determines whether the calculated quality variation meets the management conditions preset by the user. Furthermore, the analysis result output unit 74 notifies the user by displaying the calculated quality variation and the determination result of whether the management conditions are met on the UI screen 1000 (Figure 14).

To determine whether the control conditions are met, for example, the control range specified by the user may be compared with the quality variation for each lot, and notification may be given as to whether the quality variation falls within the control range. Also, it may be possible to fix the material used and notify whether the quality of the molded product falls within the control range when the shape conditions and standard molding conditions are changed. Also, it may be possible to fix the shape conditions and standard molding conditions and notify whether the quality of the molded product falls within the control range when the material is changed.

Next, FIG. 9 shows an example of the configuration of a general computer 100, such as a personal computer or a server computer, that realizes the molded product quality variation estimation device 40.

The computer 100 includes a processor 101 such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit), memory 102 such as a DRAM (Dynamic Random Access Memory), storage 103 such as a HDD (Hard Disk Drive) or SSD (Solid State Drive), input devices 104 such as a keyboard, mouse, touch panel, etc., output devices 105 such as a display, etc., and a communication module 106 such as a NIC (Network Interface Card).

The computer 100 executes a program using the processor 101, and performs processing defined by the program using storage resources (memory 102 and storage 103), a communication module 106, etc. Therefore, the subject of processing performed by executing a program may be the processor 101. Similarly, the subject of processing performed by executing a program may be a controller, device, system, computer, or node having a processor. The subject of processing performed by executing a program may be a calculation unit, and may include a dedicated circuit that performs specific processing. Here, a dedicated circuit is, for example, an FPGA (Field Programmable Gate Array), an ASIC (Application Specific Integrated Circuit), or a CPLD (Complex Programmable Logic Device).

The program may be installed on the computer 100 from a program source. The program source may be, for example, a program distribution server or a storage medium readable by the computer 100. When the program source is a program distribution server, the program distribution server may include a processor and storage resources for storing the program to be distributed, and the processor of the program distribution server may distribute the program to be distributed to other computers. In addition, in an embodiment, two or more programs may be realized as one program, and one program may be realized as two or more programs.

The functional blocks of the sensor information storage unit 51, feature extraction unit 52, linking processing unit 57, learning data selection unit 59, PVT characteristic calculation model estimation unit 60 (excluding various internal DBs), and viscosity calculation model estimation unit 61 (excluding various internal DBs) in the physical property calculation model generation unit 41 are realized by the processor 101 provided in the computer 100. Similarly, the functional blocks of the analysis physical property model reading unit 72, analysis unit 73, and analysis result output unit 74 in the flow analysis unit 42 are realized by the processor 101 provided in the computer 100.

These functional blocks are realized by the processor 101 executing a specific program loaded into memory. However, some or all of these functional blocks may be realized as hardware using integrated circuits or the like. Furthermore, these functional blocks may be realized by one computer 100, or by multiple computers 100. When realized by multiple computers 100, the multiple computers 100 only need to be connected via a network, and may be distributed and located in remote locations.

The feature quantity DB 53, quality DB 54, PVT characteristic calculation model coefficient DB 55, viscosity calculation model coefficient DB 56, learning DB 58, various DBs within the PVT characteristic calculation model estimation unit 60, and various DBs within the viscosity calculation model estimation unit 61 in the physical property calculation model generation unit 41 of the molded product quality variation estimation device 40 are realized by the memory 102 and storage 103 provided in the computer 100. Similarly, the flow analysis unit 42 and the analysis physical property calculation model DB 71 are realized by the memory 102 and storage 103 provided in the computer 100.

The input unit 43 is realized by an input device 104 provided in the computer. The display unit 44 is realized by an output device 105 provided in the computer. The communication unit 45 is realized by a communication module 106 provided in the computer.

In addition, if the molded product quality variation estimation device 40 exists, for example, on a cloud server, the user will connect to the molded product quality variation estimation device 40 via a network using a terminal device such as a PC that the user uses, and operate the molded product quality variation estimation device 40.

<Injection molding process by injection molding machine 31>
Next, a description will be given of an injection molding process using the injection molding machine 31. Fig. 10 is a cross-sectional view showing an example of the configuration of the injection molding machine 31.

As mentioned above, the injection molding process is roughly divided into the following steps: weighing/plasticization, injection, pressure holding, cooling, and removal.

In the metering and plasticization process, the injection molding machine 31 uses the plasticization motor 501 as a driving force to retract the screw 502, and supplies the material, resin pellets 504, from the hopper 503 into the cylinder 505. The material is then plasticized into a uniform molten state by heating the cylinder 505 with the heater 506 and by rotating the screw 502. Depending on the settings of the back pressure and rotation speed of the screw 502, the density of the molten material and the degree of breakage of the reinforcing fibers change, and these changes can affect the quality of the molded product.

In the injection process and the pressure holding process, the injection molding machine 31 advances the screw 502 using the injection motor 507 as the driving force, and injects the molten material into the mold 509 through the nozzle 508. At this time, the pressure measured by the load cell 510 is controlled so as to approach the pressure holding value included in the injection molding conditions. The molten material injected into the mold 509 is simultaneously subjected to cooling from the wall surface of the mold 509 and shear heat caused by the flow. In other words, the molten material flows inside the mold 509 while being subjected to cooling and heating. If the clamping force that keeps the mold 509 closed is insufficient, tiny gaps will occur in the mold 509 after the molten material solidifies, which can affect the quality of the molded product.

In the cooling process, the injection molding machine 31 cools the mold 509 below the solidification temperature to solidify the molten material. Residual stresses that occur during the cooling process can affect the quality of the molded product. Residual stresses occur due to anisotropy of the material properties caused by the flow inside the mold 509, density distribution due to holding pressure, and uneven molding shrinkage rates.

In the removal process, the injection molding machine 31 opens the mold 509 by driving the mold clamping mechanism 512 using the motor 511 as the driving force. Then, the ejector mechanism 514 is driven by the ejection motor 513 as the driving force, and the molded product, which is the solidified molten material, is removed from the mold 509. At this time, if the ejector mechanism 514 does not exert a sufficient ejection force evenly on the molded product, residual stress will remain in the molded product, which may affect the quality of the molded product.

In the injection molding machine 31, the pressure value of the load cell 510 is controlled to approach the pressure value included in the reference molding conditions. The temperature of the cylinder 505 is controlled by multiple heaters 506. However, since there are machine differences in the shape of the screw 502, the shape of the cylinder 505, and the shape of the nozzle 508 in the injection molding machine 31, different pressure losses may occur in each injection molding machine 31. As a result, the pressure of the material at the material inlet of the mold 509 becomes a lower value than the pressure specified as the reference molding conditions. In addition, the temperature of the material at the material inlet of the mold 509 may differ from the temperature specified as the reference molding conditions due to the arrangement of the heater 506 and the shear heat of the material in the nozzle 508. Such differences in pressure and temperature may affect the quality of the molded product.

In this embodiment, the weight and dimensions of the molded product are evaluated as a quality inspection of the molded product. It is believed that the weight of the molded product varies depending on the specific volume of the material relative to the volume of the mold 509, provided that there are no defects in the shape of the molded product, such as short shots or overpacking. Therefore, the weight of the molded product is a physical quantity that is highly correlated with the PVT characteristics. It is believed that the dimensions of the molded product vary depending on the shrinkage rate of the material, provided that there are no defects in the shape of the molded product, such as short shots or overpacking. Therefore, the dimensions of the molded product are also a physical quantity that is highly correlated with the PVT characteristics.

The shape characteristics of a molded product (weight, dimensions, etc.) are strongly correlated with the pressure history, temperature history, and clamping force during the injection, pressure holding, and cooling processes.

Next, FIG. 11 shows an example of a mold 509, where (A) is a top view of a product portion 5091 of the mold 509, (B) is a side view of the product portion 5091, (C) is a top view of a runner portion 5092 of the mold 509, and (D) is a side view of the runner portion 5092.

In the mold 509, the molten material flows into the product section 5091 from five gates 5093 in the runner section 5092. In the product section 5091, a material temperature sensor 321 that measures the material temperature in the mold 509, a material pressure sensor 322 that measures the material pressure, and a mold temperature sensor 323 that measures the temperature of the mold 509 are arranged. In the runner section 5092, the material temperature sensor 321 and the material pressure sensor 322 are arranged. The material temperature sensor 321, the material pressure sensor 322, and the mold temperature sensor 323 correspond to the in-mold sensor 32 (Figure 1).

In addition, it is preferable that the arrangement of each sensor (material temperature sensor 321, material pressure sensor 322, and mold temperature sensor 323) includes at least the sprue section 5094 or runner section 5092 from the material inlet in the mold 509 to the product section (cavity) 5091. In this specification, "preferable" merely means that some advantageous effect can be expected, and does not mean that the configuration is essential.

Furthermore, the placement of each sensor may be in a location where a characteristic flow can be observed, such as directly below the gate in product section 5091, at the resin junction (weld), or at the end of the flow (all not shown). In this case, a physical quantity that correlates with the material's inherent PVT characteristics can be determined with higher accuracy from the physical quantities obtained from multiple sensors with different placements. Also, by using the integral value up to the peak obtained by the pressure sensor directly below the gate, which correlates with the material's inherent viscosity, highly accurate predictions are possible.

For example, during molding under standard molding conditions, the pressure and temperature of the molten material during molding vary depending on the position within the mold 509. Therefore, by using multiple sensors with different positions, it is possible to measure the physical quantities of the material under different pressure and temperature conditions under a single molding condition. This makes it possible to determine the physical quantities that correlate with the material's inherent PVT characteristics with greater precision.

The appropriate placement of each sensor varies depending on the mold structure and the physical quantity being measured. For physical quantities other than the mold opening amount, it is preferable to place the sensors in the sprue section 5094, if possible, regardless of the mold structure.

If the gate section 5093 is a side gate, jump gate, submarine gate, or banana gate, the sensors are placed in the runner section 5092 directly below the sprue section 5094, the runner section 5092 immediately before the gate section 5093, etc.

If the gate portion 5093 is a pin gate, a three-plate structure is required for the placement of the sensors, but the sensors are placed in the runner portion 5092 directly below the sprue portion 5094. Also, if the gate portion 5093 is a pin gate, a dummy runner portion 5092 that is not connected to the product portion 5091 may be provided for measurement purposes, and the sensors may be placed on that runner portion. Providing a runner portion 5092 dedicated to measurement improves the freedom of mold design.

If the gate section 5093 is a film gate or a fan gate, each sensor is installed in the runner section 5092 before it flows into the gate section 5093.

<Processing for Estimating Quality Variation of Molded Product by Injection Molding System 10>
Next, FIG. 12 is a flowchart illustrating an example of a molded product quality variation estimation process performed by the injection molding system 10.

As a premise for the molded product quality variation estimation process, it is assumed that the learning DB 58 of the physical property calculation model generation unit 41 already stores a learning data set corresponding to a material (material with known physical properties) that has the same product number as the material to be analyzed (material with unknown physical properties) that can be specified by the user but is from a different lot. In addition, it is assumed that the estimation DBs 602, 612 of the physical property calculation model generation unit 41 already store an estimation data set corresponding to the material to be analyzed (material with unknown physical properties) that can be specified by the user. However, if the estimation data set corresponding to the material to be analyzed (material with unknown physical properties) by the user is not stored in the estimation DBs 602, 612, it is sufficient to execute an injection molding process using the material in question in the injection molding machine 31, generate an estimation data set corresponding to the material, and store it in the estimation DBs 602, 612.

The molded product quality variation estimation process is started, for example, when the user specifies the product number and lot number of the material to be analyzed, the analysis shape (mold product number), analysis properties, analysis conditions, and the quality control range of the molded product on the UI screen 1000 (Figure 14), and operates the "Execute analysis" button 1006.

First, the learning data selection unit 59 of the physical property calculation model generation unit 41 refers to the learning DB 58 and selects one or more learning data sets corresponding to materials (materials with known physical properties) that have the same product number but different lots as the material (materials with unknown physical properties) specified by the user as the analysis target. The learning data selection unit 59 then extracts PVT characteristic learning data sets to be used for generating a PVT characteristic calculation model from the selected learning data sets, and stores them in the learning DB 601 of the PVT characteristic calculation model estimation unit 60. The learning data selection unit 59 also extracts a viscosity learning data set to be used for generating a viscosity calculation model from the selected learning data sets, and stores them in the learning DB 611 of the viscosity calculation model estimation unit 61 (step S1).

Next, the regression model learning unit 603 generates a learned regression model that outputs PVT characteristic calculation model coefficients and stores it in the learned regression model DB 604. At the same time, the regression model learning unit 613 generates a learned regression model that outputs viscosity calculation model coefficients and stores it in the learned regression model DB 614 (step S2).

Specifically, the regression model learning unit 603 reads out the PVT characteristic learning dataset from the learning DB 601, and specifies the feature dataset and molded product quality included in the PVT characteristic learning dataset as explanatory variables of the regression model. The regression model learning unit 603 also specifies the PVT characteristic calculation model coefficients as objective variables of the regression model, and generates a learned regression model by learning a regression model that predicts the objective variable from the explanatory variables. At the same time, the regression model learning unit 613 reads out the viscosity learning dataset from the learning DB 611, and specifies the feature dataset included in the viscosity learning dataset as explanatory variables of the regression model. The regression model learning unit 613 also specifies the viscosity calculation model coefficients as objective variables of the regression model, and generates a learned regression model by learning a regression model that predicts the objective variable from the explanatory variables.

Next, the physical property calculation model estimation unit 605 generates an estimated PVT characteristic calculation model corresponding to the material whose PVT characteristics are unknown, and stores it in the PVT characteristic calculation model DB 606. At the same time, the physical property calculation model estimation unit 615 generates an estimated viscosity calculation model for the material whose viscosity is unknown, and stores it in the viscosity calculation model DB 616 (step S3).

Specifically, the physical property calculation model estimation unit 605 reads out an estimation data set corresponding to the material to be analyzed from the estimation DB 602, inputs it to the learned regression model for the PVT characteristic calculation model generated in step S2, and obtains coefficients of the PVT characteristic calculation model, thereby generating an estimated PVT characteristic calculation model corresponding to the material to be analyzed. At the same time, the physical property calculation model estimation unit 605 inputs an estimation data set corresponding to the material to be analyzed from the estimation DB 612 to the learned regression model for the viscosity calculation model generated in step S2, and obtains coefficients of the viscosity calculation model, thereby estimating an estimated viscosity calculation model corresponding to the material to be analyzed.

The estimated PVT characteristic calculation model and the estimated viscosity calculation model generated in step S3 are also stored in the analysis property calculation model DB71 of the flow analysis unit 42.

Next, the analysis property model reading unit 72 of the flow analysis unit 42 reads out the coefficients of the estimated PVT characteristic calculation model and the coefficients of the estimated viscosity calculation model corresponding to the material to be analyzed from the analysis property calculation model DB 71, and outputs them to the analysis unit 73. Then, the analysis unit 73 executes a flow analysis of the material to be analyzed using the coefficients of the estimated PVT characteristic calculation model and the coefficients of the estimated viscosity calculation model according to the analysis shape and analysis conditions specified by the user, and estimates, for example, the quality (weight, dimensions, etc.) of the molded product as the analysis result and outputs it to the analysis result output unit 74 (step S4).

Then, the analysis result output unit 74 compiles the analysis results (molded product quality) by the analysis unit 73 for each lot, and calculates, for example, the difference between the maximum and minimum values representing the quality (weight, dimensions, etc.) as the quality variation. The analysis result output unit 74 also determines whether the calculated quality variation passes the management conditions preset by the user. The analysis result output unit 74 then displays the quality variation and the pass/fail judgment results for the management conditions on the UI screen 1000 (Figure 14) (step S5). This is an example of the molded product quality variation estimation process by the injection molding system 10.

The molded product quality variation estimation process can generate an estimated PVT property calculation model and an estimated viscosity calculation model that correspond to the material to be analyzed. The molded product quality variation estimation process can also estimate the PVT characteristics and viscosity that correspond to the material to be analyzed based on the generated estimated PVT property calculation model and estimated viscosity calculation model. The molded product quality variation estimation process can then estimate the quality and its variation of the molded product when the material to be analyzed is used, based on the estimated PVT characteristics and viscosity that correspond to the material to be analyzed. This allows the user to obtain the quality variation without mass-producing the product (molded product) using the material to be analyzed, and therefore allows the user to appropriately select a material suitable for mass production of the product.

<Validity of the estimated PVT property calculation model and the estimated viscosity calculation model>
Next, FIG. 13 is a diagram for explaining the validity of the estimated PVT property calculation model and the estimated viscosity calculation model generated by the molded product quality variation estimation process.

In this validity evaluation, materials A, B, C, and D with the same product number but different lots were used, and the physical properties of materials A to D were measured using existing methods. After that, a trained regression model was created using the training datasets corresponding to materials A, B, and C. Then, an estimated PVT physical property calculation model and an estimated viscosity calculation model corresponding to material D were generated by inputting the estimation training dataset corresponding to material D into the trained regression model based on materials A, B, and C. Furthermore, the estimated PVT physical property calculation model and estimated viscosity calculation model were compared and evaluated with the actual measured values of the physical properties of material D.

Figure (A) shows a comparison between the generated estimated PVT characteristic calculation model (line plot) corresponding to material D and the actual measured values (dot plot) of the PVT characteristics of material D when the pressure level is fixed at 50 MPa, 100 MPa, or 150 MPa and the temperature is changed from 30°C to 250°C in 5°C increments. The horizontal axis of Figure (A) is temperature, and the vertical axis is specific volume. In Figure (A), the estimated PVT characteristic calculation model is almost identical to the actual measured values, so it can be seen that the generated estimated PVT physical property calculation model is highly valid.

Figure (B) shows a comparison between the generated estimated viscosity calculation model (line plot) corresponding to material D and the actual measured viscosity values (dot plot) of material D when the temperature is fixed at 200°C and the shear rate is changed in the range of 6 to 12,000 [1/sec]. Figure (B) is a double logarithmic graph, with the horizontal axis representing shear rate and the vertical axis representing viscosity. In Figure (B), the estimated viscosity calculation model is almost identical to the actual measured values, so it can be seen that the generated estimated viscosity calculation model is highly valid.

<Display example of UI screen 1000>
14 shows a display example of a UI screen 1000. The UI screen 1000 is provided with an input field 1001 for specifying a material to be analyzed, an input field 1002 for specifying a mold as an analysis shape, an input field 1003 for selecting and specifying analysis conditions, an input field 1004 for selecting and specifying an analysis physical property model, an input field 1005 for specifying a control width, an "Execute Analysis" button 1006 for instructing execution of a fluid analysis under various specified conditions, and a display field 1007 for displaying the analysis results.

In input field 1001, for example, one product number and multiple lot numbers for the material to be analyzed are input. In input field 1002, for example, the product number of the mold is input. Note that the mold product number input here may be different from the mold product number corresponding to the learning data set used to train the regression model. In input field 1003, the analysis conditions can be selected and specified by checking the items displayed (material temperature, mold temperature, injection speed, etc.). In input field 1004, the analysis property model can be selected and specified by checking the displayed physical properties (viscosity, PVT characteristics). In input field 1005, the type of quality (weight, dimensions) and its lower and upper limits can be input as the control range.

Display field 1007 displays the control range specified by the user, as well as a quantified version of the quality variation that may occur when materials D, E, and F, which have the same product number but different lots and are specified in input field 1001, are used. In addition, display field 1007 displays the pass/fail result of the quality variation relative to the control range.

Ultra-visible display 1000 allows the user to understand the quality variation that may occur when using the material that the user has specified as the analysis target, and visually check whether the variation falls within the control range, allowing the user to select an appropriate material.

The present invention is not limited to the above-described embodiments, and various modifications are possible. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to having all of the configurations described. Furthermore, it is possible to replace part of the configuration of one embodiment with or add to the configuration of another embodiment.

Furthermore, the above-mentioned configurations, functions, processing units, processing means, etc. may be realized in part or in whole in hardware, for example by designing them as integrated circuits. The above-mentioned configurations, functions, etc. may also be realized in software by a processor interpreting and executing a program that realizes each function. Information such as the program, table, file, etc. that realizes each function can be stored in a memory, a recording device such as a hard disk or SSD, or a recording medium such as an IC card, SD card, DVD, etc. Furthermore, the control lines and information lines shown are those that are considered necessary for the explanation, and do not necessarily show all control lines and information lines in the product. In reality, it can be considered that almost all configurations are connected to each other.

10: injection molding system, 20: manufacturing control device, 21: manufacturing condition determination unit, 22: production instruction unit, 31: injection molding machine, 32: in-mold sensor, 321: material temperature sensor, 322: material pressure sensor, 323: mold temperature sensor, 33: quality inspection device, 40: estimation device, 41: physical property calculation model generation unit, 42: flow analysis unit, 43: input unit, 44: display unit, 45: communication unit, 51: sensor information storage unit, 52: feature extraction unit, 53: feature DB, 54: quality DB, 55: PVT characteristic calculation model coefficient DB, 56: viscosity calculation model coefficient DB, 57: linking processing unit, 58: learning DB, 59: learning data selection unit, 60: PVT characteristic calculation model estimation unit, 601...Learning DB, 602...Estimation DB, 603...Regression model learning section, 604...Learned regression model DB, 605...Physical property calculation model estimation section, 606...PVT characteristic calculation model DB, 61...Viscosity calculation model estimation section, 611...Learning DB, 612...Estimation DB, 613...Regression model learning section, 614...Learned regression model DB, 615...Physical property calculation model estimation section, 606...Viscosity calculation model DB, 71...Analysis physical property calculation model DB, 72...Analysis physical property calculation model reading section, 73...Analysis section, 74...Analysis result output section, 100...Computer, 509...Mold, 5091...Product section, 5092...Runner section, 5093...Gate section, 5094...Spur section, 1000...UI screen

Claims (15)

1. A molded product quality variation estimation device that estimates quality variation of a molded product produced by injection molding, comprising:
   a physical property calculation model estimating unit that estimates a physical property calculation model corresponding to a second material based on process data when a molded product is manufactured using a first material whose physical properties are known, a learned regression model generated based on a physical property calculation model coefficient corresponding to the first material, and process data when a molded product is manufactured using a second material whose physical properties are unknown;
   a flow analysis unit that estimates a quality variation of a molded product when the molded product is manufactured using the second material by performing a flow analysis based on the physical property calculation model corresponding to the estimated second material,
   the physical property calculation model estimation unit estimates two or more types of the physical property calculation models corresponding to two or more different types of physical properties, based on two or more types of the trained regression models corresponding to two or more different types of physical properties, respectively;
   The flow analysis unit is a molded product quality variation estimation device that performs the flow analysis based on the estimated two or more types of physical property calculation models.
2. The molded product quality variation estimation device according to claim 1,
   the first material and the second material are recycled materials,
   The first material and the second material have the same product number but are from different lots.
3. The molded product quality variation estimation device according to claim 1,
   a regression model learning unit that generates the learned regression model by learning a regression model in which the process data when a molded product is manufactured using the first material is used as an explanatory variable and the physical property calculation model coefficients are used as objective variables.
4. The molded product quality variation estimation device according to claim 3,
   A molded product quality variation estimation device comprising: a selection unit that selects the explanatory variables to be used in learning the regression model.
5. The molded product quality variation estimation device according to claim 4,
   The selection unit selects, as the explanatory variable, the process data that is highly correlated with the physical property corresponding to the physical property calculation model coefficient that is set as the objective variable.
6. The molded product quality variation estimation device according to claim 1,
   The two or more different physical properties include at least one of PVT characteristics, viscosity, crystallinity, and crystallization rate.
7. The molded product quality variation estimation device according to claim 5,
   The process data includes feature values of in-mold sensor values and the quality of the molded product.
8. The molded product quality variation estimation device according to claim 7,
   The molded product quality variation estimation device, wherein the in-mold sensor value includes at least one of pressure and temperature.
9. The molded product quality variation estimation device according to claim 8,
   The feature value includes at least one of an integral value of the in-mold sensor value during the injection process, a peak value of the in-mold sensor value, an integral value of the in-mold sensor value from the injection process to the cooling process, and a maximum differential value of the in-mold sensor value.
10. The molded product quality variation estimation device according to claim 9,
    The quality of the molded product includes at least one of weight and dimensions.
11. The molded product quality variation estimation device according to claim 10,
    The selection unit is
    An integral value of the pressure as the in-mold sensor value during the injection process is selected as the explanatory variable of the regression model having a viscosity calculation model coefficient as a response variable;
    A molded product quality variation estimation device that selects a feature based on pressure as the in-mold sensor value, a feature based on temperature as the in-mold sensor value, and weight and dimensions as the quality as the explanatory variables of the regression model with PVT characteristic calculation model coefficients as the objective variable.
12. The molded product quality variation estimation device according to claim 1,
    The flow analysis unit is a molded product quality variation estimation device that estimates the minimum and maximum values of at least one of the weight and dimensions of the molded product as the quality variation of the molded product when the molded product is manufactured using the second material.
13. The molded product quality variation estimation device according to claim 12,
    The flow analysis unit compares the estimated quality variation with a control range, and if the quality variation falls outside the control range, notifies a user to that effect.
14. A method for estimating quality variation of a molded product by a molded product quality variation estimation device that estimates quality variation of a molded product produced by injection molding, comprising:
    a physical property calculation model estimation step of estimating a physical property calculation model corresponding to the second material based on process data when a molded product is manufactured using a first material whose physical properties are known, a learned regression model generated based on a physical property calculation model coefficient for the first material, and process data when a molded product is manufactured using a second material whose physical properties are unknown;
    A flow analysis step of estimating a quality variation of a molded product when the molded product is manufactured using the second material by performing a flow analysis based on the physical property calculation model corresponding to the estimated second material,
    the physical property calculation model estimation step estimates two or more types of the physical property calculation models corresponding to two or more different types of physical properties, based on two or more types of the trained regression models corresponding to two or more different types of physical properties, respectively;
    The flow analysis step is a method for estimating quality variation of a molded product, the flow analysis being performed based on the estimated two or more types of physical property calculation models.
15. Factory facilities including injection molding machines and quality inspection equipment;
    a molded product quality variation estimation device that estimates quality variation of a molded product manufactured by the injection molding machine,
    The injection molding machine outputs an in-mold sensor value during injection molding,
    the quality inspection device outputs a quality inspection result of a molded product manufactured by the injection molding machine;
    The molded product quality variation estimation device includes:
    a physical property calculation model estimating unit that estimates a physical property calculation model corresponding to a second material based on process data including the feature amount based on the in-mold sensor value and the quality inspection result when a molded product is manufactured using a first material having known physical properties, a learned regression model generated based on a physical property calculation model coefficient for the first material, and process data when a molded product is manufactured using a second material having unknown physical properties; and
    a flow analysis unit that estimates a quality variation of a molded product when the molded product is manufactured using the second material by performing a flow analysis based on the physical property calculation model corresponding to the estimated second material,
    the physical property calculation model estimation unit estimates two or more types of physical property calculation models corresponding to two or more different types of physical properties, based on two or more types of trained regression models corresponding to two or more different types of physical properties, respectively;
    The flow analysis unit is an injection molding system that performs the flow analysis based on the two or more types of estimated physical property calculation models.

Images