Prediction device, prediction system, and prediction program

Abstract

This prediction device comprises: an acquisition unit that acquires first condition information related to an injection molding condition detected by a first sensor over a predetermined detection period and second condition information related to an injection molding condition detected by a second sensor different from the first sensor over the detection period when a resin material is injection-molded; a partition unit that partitions each of the acquired first condition information and second condition information into divided areas by a predetermined time width; an extraction unit that extracts a first feature quantity from the first condition information for each divided area and extracts a second feature quantity from the second condition information for each divided area; and a prediction unit that predicts characteristics of the molded article of the resin material manufactured by injection molding on the basis of the extracted first feature quantity and second feature quantity.

Description

Prediction device, prediction system, and prediction program

[0001] The present invention relates to a prediction device, a prediction system, and a prediction program.

[0002] Fiber-reinforced resins containing fibrous materials and resins have excellent strength and rigidity and are widely used in electrical and electronic applications, civil engineering and construction applications, automotive applications, aircraft applications, and the like. Resin molded articles are desired to have properties suited to their applications. The properties of resin molded articles are evaluated, for example, using destructive testing. Patent Document 1 describes the relationship between injection molding process conditions and molded article properties.

[0003] Japanese Patent Application Laid-Open No. 2022-141495

[0004] Thus, when molding a resin material by injection molding, it is desirable to be able to predict the properties of the molded product from information about the manufacturing process.

[0005] The present invention has been made in view of the above circumstances, and has an object to provide a prediction device, a prediction system, and a prediction program that are capable of predicting the characteristics of a molded product from information about the manufacturing process.

[0006] The above object of the present invention can be achieved by the following means.

[0007] (1) A prediction device including: an acquisition unit that acquires, when injection molding a resin material, first condition information related to injection molding conditions detected by a first sensor over a predetermined detection period, and second condition information related to injection molding conditions detected by a second sensor different from the first sensor over the detection period; a partitioning unit that partitions each of the acquired first condition information and second condition information into divided regions each having a predetermined time width; an extraction unit that extracts a first feature value from the first condition information for each divided region, and extracts a second feature value from the second condition information for each divided region; and a prediction unit that predicts characteristics of a molded product made from the resin material by injection molding, based on the extracted first feature value and second feature value.

[0008] (2) The resin material is injection molded by an injection molding device including a cylinder and a mold, the first condition information is information regarding changes in physical quantities within the cylinder over the detection period, and the second condition information is information regarding changes in physical quantities within the mold over the detection period.

[0009] (3) The prediction device according to (1), further comprising a determination unit that determines the first feature amount and the second feature amount, and the extraction unit extracts the determined first feature amount and the second feature amount.

[0010] (4) The prediction device described in (3) above, wherein the determination unit determines at least one of the first feature and the second feature using at least one of principal component analysis, kernel principal component analysis, partial least squares regression, a statistical method, a deterministic algorithm, and a deep learning model.

[0011] (5) The prediction device according to (1), wherein at least one of the first feature amount and the second feature amount is a parameter related to a length of time obtained for each of the divided regions.

[0012] (6) The prediction device described in (1) above, wherein the acquisition unit further acquires model condition information generated using a model-based development technique, the partitioning unit partitions the acquired model condition information into divided regions each having a predetermined time width, the extraction unit extracts model features for each divided region from the model condition information, and the prediction unit predicts the characteristics of a molded product made of the resin material based on the first feature, the second feature, and the model feature.

[0013] (7) The prediction device described in (1) above, wherein the acquisition unit further acquires third condition information detected by a third sensor different from the first sensor and the second sensor over the detection period during the injection molding, the partitioning unit partitions the third condition information into divided areas each having a predetermined time width, the extraction unit extracts a third feature value from the third condition information for each divided area, and the prediction unit predicts the characteristics based on the extracted first feature value, second feature value, and third feature value.

[0014] (8) The prediction device according to (1) above, wherein the prediction unit includes a machine learning model.

[0015] (9) The prediction device according to (1) above, wherein the resin material includes a fiber-reinforced resin material.

[0016] (10) The prediction device according to (1) above, wherein the properties of the molded product include at least one of flexural modulus and weather resistance.

[0017] (11) The prediction device according to (1) above, further comprising an output unit that outputs characteristic information relating to the predicted characteristics of the molded product.

[0018] (12) A prediction system comprising the prediction device according to (1) above, the first sensor, and the second sensor.

[0019] (13) A prediction program for causing a computer to execute a process including: acquiring, when injection molding a resin material, first condition information relating to injection molding conditions detected by a first sensor over a predetermined detection period, and second condition information relating to injection molding conditions detected by a second sensor different from the first sensor over the detection period; dividing the acquired first condition information and second condition information into divided regions each having a predetermined time width; extracting a first feature value for each divided region from the first condition information and a second feature value for each divided region from the second condition information; and predicting characteristics of a molded product made of the resin material by the injection molding based on the extracted first feature value and second feature value.

[0020] In the prediction device, prediction system, and prediction program according to the present invention, the first condition information and the second condition information detected by the first sensor and the second sensor, respectively, are divided into divided regions. Then, the first feature amount and the second feature amount are extracted for each divided region, and the characteristics of a molded product made of a resin material are predicted based on the extracted first feature amount and second feature amount. This makes it possible to predict the characteristics of the molded product from information about the manufacturing process.

[0021] Advantages and features provided by one or more embodiments of the present invention will be more fully understood from the following detailed description and the accompanying drawings, which are for illustrative purposes only and are not intended to define limitations of the present invention.

[0014] Figure 1 is a diagram showing the overall configuration of a prediction system according to one embodiment.

[0015] Figure 2 is a block diagram showing a schematic configuration of the prediction device shown in Figure 1.

[0016] Figure 3 is a diagram showing an example of the configuration of an injection molding machine to which the first sensor, second sensor, and third sensor shown in Figure 1 are attached.

[0017] Figure 4 is a block diagram showing the functional configuration of the prediction device shown in Figure 2.

[0018] Figure 5 is a diagram showing an example of first condition information acquired by the acquisition unit shown in Figure 4.

[0019] Figure 6 is a diagram showing an example of divided regions of the first condition information partitioned by the partition unit shown in Figure 4.

[0020] Figure 7 is a flowchart showing the procedure of a prediction process executed in the prediction device shown in Figure 2.

[0021] Figure 8 is a flowchart showing a machine learning method for a trained model included in the prediction unit shown in Figure 2.

[0022] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, the scope of the present invention is not limited to the disclosed embodiments. In the description of the drawings, the same elements are denoted by the same reference numerals, and duplicate explanations will be omitted. Furthermore, the dimensional proportions in the drawings are exaggerated for the convenience of explanation and may differ from the actual proportions.

[0023] [Embodiment] <Configuration of Prediction System> FIG. 1 is a diagram showing the overall configuration of a prediction system according to one embodiment. The prediction system includes, for example, a prediction device 100, a first sensor 200, a second sensor 300, and a third sensor 400. This prediction system predicts the properties of a molded product made of a resin material using injection molding conditions detected by the first sensor 200, the second sensor 300, and the third sensor 400 during injection molding. The resin material may be a single resin material or may contain various additives. The resin material may be, for example, a fiber-reinforced resin material containing a resin and a fibrous substance. The resin material may be, for example, a filler-added resin material containing a resin and an inorganic filler. The resin material may also contain an antioxidant, a plasticizer, a light stabilizer, a flame retardant, or the like. For parts that require thinness and strength, fiber-reinforced resin materials are preferably used because they can achieve high strength, elastic modulus, and the like. Furthermore, in the injection molding of thin parts, molding defects are common, making it even more important to obtain information about the manufacturing process.

[0024] Examples of resins included in the resin material include known thermosetting resins and thermoplastic resins. Examples of resins included in the resin material include polypropylene resin, polyolefin resin, epoxy resin, phenolic resin, unsaturated polyester resin, vinyl ester resin, polycarbonate resin, and polyester resin. Examples of polyolefin resins include maleic anhydride-modified polypropylene. Examples of resins included in the resin material include polyamide resin, liquid crystal polymer resin, polyethersulfone resin, polyetheretherketone resin, polyarylate resin, polyphenylene ether resin, and polyphenylene sulfide resin. Examples of resins included in the resin material include polyacetal resin, polysulfone resin, polyimide resin, polyetherimide resin, polystyrene resin, modified polystyrene resin, AS resin, and ABS resin. AS resin is a copolymer of acrylonitrile and styrene. ABS resin is a copolymer of acrylonitrile, butadiene, and styrene. Examples of resins contained in the resin material include modified ABS resin, MBS resin, modified MBS resin, polymethyl methacrylate resin, and modified polymethyl methacrylate resin. MBS resin is a copolymer of methyl methacrylate, butadiene, and styrene. The resin contained in the resin material may be one of these resins, or a mixture of two or more of these resins.

[0025] For example, fibrous materials contained in fiber-reinforced resin materials are added to resins for the purpose of improving the strength of the fiber-reinforced resin material. Examples of such fibrous materials include glass fiber, carbon fiber, aramid fiber, alumina fiber, silicon carbide fiber, boron fiber, and silicon carbide fiber. Examples of carbon fibers that can be used include polyacrylonitrile, pitch-based, cellulose-based, and hydrocarbon vapor-grown carbon fibers and graphite fibers. Examples of glass fibers that can be used include E-glass and S-glass. The fibrous material contained in the fiber-reinforced resin material may be one of these materials, or a mixture of two or more of them.

[0026] The filler contained in the filler-added resin material is added, for example, for the purpose of reducing dimensional changes of the filler-added resin material due to temperature and humidity. The filler is, for example, alumina, talc, silica, mica, kaolin, calcium carbonate, glass beads, etc. The filler-added resin material may contain a mixture of two or more fillers.

[0027] The prediction device 100 is a computer such as a personal computer, a smartphone, or a tablet terminal. The prediction device 100 is configured to be connectable to each of a first sensor 200, a second sensor 300, and a third sensor 400. The prediction device 100 transmits and receives various types of information to and from each of the first sensor 200, the second sensor 300, and the third sensor 400.

[0028] FIG. 2 is a block diagram showing a schematic configuration of the prediction device 100.

[0029] 2 , the prediction device 100 includes a CPU 110, a ROM 120, a RAM 130, a storage 140, a communication interface 150, a display unit 160, and an operation reception unit 170. CPU is an abbreviation for Central Processing Unit. ROM is an abbreviation for Read Only Memory. RAM is an abbreviation for Random Access Memory. Each component is connected to each other via a bus so that they can communicate with each other.

[0030] The CPU 110 controls the above components and performs various arithmetic processing in accordance with a program stored in the ROM 120 or the storage 140. The specific functions of the CPU 110 will be described later.

[0031] The ROM 120 stores various programs and various data.

[0032] The RAM 130 serves as a working area for temporarily storing programs or data.

[0033] The storage 140 stores various programs including an operating system or various data. For example, an application for predicting the characteristics of a molded product made of a resin material using a trained classifier is installed in the storage 140. The storage 140 may also store first condition information, second condition information, and third condition information acquired from the first sensor 200, the second sensor 300, and the third sensor 400. The first condition information, second condition information, and third condition information are information related to injection molding conditions for the resin material. Details of the first condition information, second condition information, and third condition information will be described later. The storage 140 may also store a trained model used as a classifier and training data used in machine learning.

[0034] The communication interface 150 is an interface for communicating with other devices. A wired or wireless communication interface conforming to various standards is used as the communication interface 150. The communication interface 150 is used, for example, to receive the first condition information, the second condition information, and the third condition information from the first sensor 200, the second sensor 300, and the third sensor 400, and to transmit characteristic information relating to the predicted characteristics of the molded product to a server or the like.

[0035] The display unit 160 includes a liquid crystal display, an organic EL display, etc., and displays various information. The display unit 160 may be configured with viewer software, a printer, etc.

[0036] The operation reception unit 170 includes a touch sensor, a pointing device such as a mouse, a keyboard, etc., and receives various operations from the user. Note that the display unit 160 and the operation reception unit 170 may form a touch panel by superimposing a touch sensor serving as the operation reception unit 170 on the display surface serving as the display unit 160.

[0037] The first sensor 200, the second sensor 300, and the third sensor 400 detect injection molding conditions during injection molding of a resin material using the injection molding device 500 over a predetermined detection period. The first sensor 200, the second sensor 300, and the third sensor 400 detect different injection molding conditions. The injection molding conditions detected by the first sensor 200, the second sensor 300, and the third sensor 400 are transmitted to the prediction device 100.

[0038] 3 shows an example of the configuration of first sensor 200, second sensor 300, and third sensor 400 attached to injection molding apparatus 500. Injection molding apparatus 500 has, for example, a molding machine 510, a nozzle 520, and a mold 530. The injection molding process includes, for example, an injection process of about 0.5 to 10 seconds, a pressure holding process of about 1 to 60 seconds, and a cooling process of about 10 to 50 seconds.

[0039] The molding machine 510 includes, for example, a hopper 511, a cylinder 512, a screw 513, and a heater 514. The resin material introduced into the hopper 511 enters the cylinder 512. Within the cylinder 512, the resin material moves toward the nozzle 520 due to the rotation of the screw 513. At this time, the resin material is heated by the heater 514 and melts. The melted resin material is injected from the nozzle 520 into a mold 530. The resin material solidifies within the mold 530, thereby producing a molded product. While FIG. 3 shows a screw-type molding machine, the molding machine 510 may be a molding machine of another type, such as a plunger-type molding machine.

[0040] The first sensor 200 and the third sensor 400 detect temporal changes in physical quantities in the molding machine 510, for example, over a predetermined detection period. The first sensor 200 detects temporal changes in, for example, the pressure of the screw 513, the moving speed of the screw 513, or the position of the screw 513. The third sensor 400 includes, for example, an AE sensor, and detects temporal changes in elastic waves in the cylinder 512. AE is an abbreviation for Acoustic Emission.

[0041] The second sensor 300 detects, for example, over a predetermined detection period, a change in a physical quantity within the mold 530 over time. The second sensor 300 detects, for example, a change in the pressure within the mold 530 or a change in the temperature within the mold 530 over time.

[0042] 4 is a block diagram showing the functional configuration of the prediction device 100. The prediction device 100 functions as an acquisition unit 111, a partition unit 112, a determination unit 113, an extraction unit 114, a prediction unit 115, and an output unit 116 by the CPU 110 reading a program stored in the storage 140 and executing processing.

[0043] The acquisition unit 111 acquires first condition information, second condition information, and third condition information. The first condition information, second condition information, and third condition information are information regarding injection molding conditions when a resin material is injection molded by the injection molding device 500. The first condition information is information detected by the first sensor 200 over a predetermined detection period. The second condition information is information detected by the second sensor 300 over the same detection period as the first sensor 200. The third condition information is information detected by the third sensor 400 over the same detection period as the first sensor 200 and the second sensor 300. The detection period is, for example, approximately 0.2 seconds to 100 seconds. The first condition information, second condition information, and third condition information include, for example, information regarding the measurement values ​​of the first sensor 200, the second sensor 300, and the third sensor 400 at each time.

[0044] 5 shows an example of the first condition information. The first condition information is, for example, a change over time in the measurement value of the first sensor 200. For example, the second condition information is a change over time in the measurement value of the second sensor 300, and the third condition information is a change over time in the measurement value of the third sensor 400.

[0045] For example, the prediction device 100 receives detection results from the first sensor 200, the second sensor 300, and the third sensor 400, and the acquisition unit 111 acquires the first condition information, the second condition information, and the third condition information. The acquisition unit 111 may acquire the first condition information, the second condition information, and the third condition information from the storage 140.

[0046] It is preferable that the acquisition unit 111 further acquires model condition information generated using a model-based development technique. The model condition information is a so-called MBD model, which is pseudo-condition information generated by virtual simulation. MBD is an abbreviation for Model Based Development. By the acquisition unit 111 further acquiring the model condition information, the amount of data of the condition information increases, making it possible to improve the prediction accuracy of the characteristics of the molded product. The acquisition unit 111 acquires the model condition information from, for example, the storage 140.

[0047] The MBD model has a function of generating model condition information from physical, chemical, or phenomenological MBD model parameters and control conditions of the injection molding machine. The MBD model parameters desirably include parameters related to fluid properties, parameters related to thermal properties, and physicochemical parameters related to the bonding strength of the fibers and resin. The fluid properties are, for example, the dynamic viscoelasticity of the flowing resin and fibers. The thermal properties are, for example, the specific heat of the resin, fibers, and mold. The thermal properties may also be the thermal resistance of the resin, fibers, and mold.

[0048] The control conditions of the injection molding machine used in the MBD model include, for example, control target values ​​and setting conditions. The control target values ​​are, for example, the time changes of the pressure of the screw 513, the movement speed of the screw 513, the position of the screw 513, and the temperature of the mold 530. The setting conditions are, for example, the gains and control time constants of the controller and mold temperature controller. Configuring the control conditions of the injection molding machine in this way allows for detailed determination of the influence of fluid and thermal properties on the model condition information, enabling more accurate prediction of molded product characteristics. Furthermore, using the MBD model has the advantage of clarifying the physicochemical or phenomenological basis for the temporal or spatial changes of each model condition information when multiple model condition information is used, thereby providing insights for improving molded product characteristics.

[0049] The MBD model is preferably configured to generate first model condition information, second model condition information, and third model condition information. The first model condition information is a predicted value of the injection molding condition detected by the first sensor 200. The second model condition information is a predicted value of the injection molding condition detected by the second sensor 300. The third model condition information is a predicted value of the injection molding condition that cannot be measured by the first sensor 200 or the second sensor 300. By configuring the MBD model in this manner, it is possible to confirm the consistency between the predicted value and the actual sensor value and improve the accuracy of the MBD model parameters. Furthermore, it is possible to generate model condition information for locations and times that cannot be predicted by the actual sensors. Therefore, it is possible to predict the characteristics of the molded product with higher accuracy.

[0050] For example, in the MBD model, the temperature inside the mold 530 is generated as model condition information. The MBD model parameters include the specific heat of the flowing resin and the mold, the thermal resistance of the resin and the mold, and thermal circuit constants. The thermal circuit constants include the location and time dependency of the resin and the mold. By solving the thermal circuit equation for these MBD model parameters, the temperature changes over time at specific locations at multiple locations can be generated as multiple model condition information. By configuring the MBD model in this way, model condition information can be generated more quickly with fewer parameters than with 3D simulation, while also improving the accuracy of predicting molded product characteristics. Here, 3D simulation refers to a technique in which a spatial shape is divided into a 3D mesh and the time evolution of spatial fields such as temperature or pressure is calculated using the finite element method. 3D simulation also includes simulation techniques that use discrete elements such as particle methods to solve the time evolution of fluid movement. 3D simulation has the disadvantage of requiring significant computational resources to accurately calculate time evolution.

[0051] The partitioning unit 112 partitions each of the first condition information, the second condition information, the third condition information, and the model condition information acquired by the acquisition unit 111 into divided regions each having a predetermined time width.

[0052] FIG. 6 shows an example of first condition information divided into sections each having a time span t. The dividing unit 112, for example, divides the first condition information into sections A1 to A14, each having a time span t. Similarly, the dividing unit 112 divides the second condition information, the third condition information, and the model condition information into sections each having a time span t. The time span t is, for example, 0.0002 seconds to 50 seconds, preferably 0.05 seconds to 10 seconds. The dividing unit 112 divides the first condition information, the second condition information, the third condition information, and the model condition information into sections each having a time span t, for example, 2 to 100, preferably 10 to 25. The dividing unit 112 may divide the first condition information, the second condition information, the third condition information, and the model condition information into sections each having a different time span. The dividing unit 112 may standardize the time axes of the first condition information, the second condition information, the third condition information, and the model condition information, and then divide them into sections each having a predetermined time span.

[0053] The determination unit 113 determines a first feature amount, a second feature amount, a third feature amount, and a model feature amount. The first feature amount is a feature amount extracted from each divided area of ​​the first condition information divided into predetermined time intervals. The second feature amount is a feature amount extracted from each divided area of ​​the second condition information divided into predetermined time intervals. The third feature amount is a feature amount extracted from each divided area of ​​the third condition information divided into predetermined time intervals. The model feature amount is a feature amount extracted from each divided area of ​​the model condition information divided into predetermined time intervals. The determination unit 113 may determine the first feature amount based on the first condition information, the second feature amount based on the second condition information, the third feature amount based on the third condition information, and the model feature amount based on the model condition information.

[0054] At least one of the first feature amount, the second feature amount, the third feature amount, and the model feature amount is, for example, a parameter related to the length of time. The parameter related to the length of time is, for example, the length of time from a first time to a second time. For example, the first time is the start time of the detection period, and the second time is the time when the measurement value of the first sensor 200, the second sensor 300, or the third sensor 400 becomes an extreme value. For example, the first time is the time when the measurement value of the first sensor 200, the second sensor 300, or the third sensor 400 becomes a first extreme value, and the second time is the time when the measurement value of the first sensor 200, the second sensor 300, or the third sensor 400 becomes a second extreme value different from the first extreme value. For example, the first time is the start time of the detection period, and the second time is the end time of the detection period. The extreme value is, for example, a maximum value or a minimum value. In the injection molding process, for example, the period from the start of injection until the pressure inside the mold reaches its maximum is an important characteristic for the properties of the molded product. By analyzing the period until the temperature inside the mold reaches its maximum value in addition to that, it becomes possible to predict the properties of the molded product with greater accuracy.

[0055] At least one of the first feature amount, the second feature amount, the third feature amount, and the model feature amount is, for example, a statistical amount, such as the average, maximum, minimum, mode, or median of the measurement values ​​of the first sensor 200, the second sensor 300, or the third sensor 400, or the simulation values ​​of the model condition information.

[0056] The determination unit 113 may determine the first feature, the second feature, the third feature, or the model feature using, for example, principal component analysis, kernel principal component analysis, PLS, a statistical method, a method using a deterministic algorithm, or a method using a deep learning model. PLS is an abbreviation for partial least squares regression, i.e., partial least squares regression. Examples of statistical methods include statistical distribution functions and cumulants. Examples of methods using a deterministic algorithm include noise separation and curve fitting. Examples of methods using a deep learning model include CVAE, GAN, LSTM, and BERT. CVAE is an abbreviation for Conditional Variational AutoEncoder. GAN is an abbreviation for Generative Adversarial Networks. LSTM is an abbreviation for Long Short Term Memory. BERT is an abbreviation for Bidirectional Encoder Representations from Transformers.

[0057] The determination unit 113 may use a feature extraction algorithm that focuses on vibration phenomena, such as a fast Fourier transform (FFT). The determination unit 113 extracts features that focus on the magnitude of vibration phenomena, thereby obtaining information about the stability of injection molding. In the prediction system according to this embodiment, the determination unit 113 preferably employs a dimension reduction algorithm that uses molding conditions as a condition vector or molded product characteristics as a target variable. This allows for the extraction of features suitable for predicting molded product characteristics that depend on injection molding conditions. Feature extraction algorithms that use molding conditions as a condition vector include conditional generative models using deep learning, such as CVAE and conditional GAN. Among conditional generative models, CVAE is particularly preferred. This is because CVAE can embed information about injection stability and disturbances, which cannot be fully expressed by injection molding conditions, into dimension-reduced features. Information about disturbances includes, for example, information about environmental fluctuations or equipment deterioration. Furthermore, feature extraction using PLS can also be suitably used in the determination unit 113. This is because PLS is a linear algorithm with high extrapolation performance. In this PLS, the characteristics of the molded product are used as the objective variables.

[0058] The extraction unit 114 extracts a first feature amount for each divided region from the first condition information. The extraction unit 114 extracts a second feature amount for each divided region from the second condition information. The extraction unit 114 extracts a third feature amount for each divided region from the third condition information. The extraction unit 114 extracts a model feature amount for each divided region from the model condition information. The first feature amount, second feature amount, third feature amount, and model feature amount are feature amounts determined by the determination unit 113. When the first feature amount is, for example, the maximum value of the measurement value of the first sensor 200, the extraction unit 114 extracts, for example, the maximum value of the measurement value of the first sensor 200 in each of the divided regions A1 to A14.

[0059] The prediction unit 115 predicts the properties of a molded product made of a resin material based on the first feature amount, second feature amount, third feature amount, and model feature amount extracted by the extraction unit 114. This molded product is manufactured by the injection molding apparatus 500. The prediction unit 115 includes, for example, a machine learning model. Specifically, the prediction unit 115 uses a trained classifier to input the first feature amount, second feature amount, third feature amount, and model feature amount extracted by the extraction unit 114 and predicts the properties of the molded product. The properties of the molded product include, for example, shape properties, surface properties, mechanical properties, optical properties, and weather resistance. The shape properties include, for example, weight and dimensions. The surface properties include, for example, appearance, color, and gloss. The mechanical properties include, for example, tensile strength, impact resistance, and flexural modulus. The optical properties include, for example, light transmittance. Preferably, the properties of the molded product predicted by the prediction unit 115 include weather resistance. Weather resistance is one of the important properties when using a molded product.

[0060] The output unit 116 outputs characteristic information relating to the characteristics of the molded product predicted by the prediction unit 115. The output unit 116 outputs the characteristic information to, for example, the display unit 160. The display unit 160 displays the predicted characteristics in association with, for example, the resin material of the molded product and the injection molding conditions.

[0061] <Overview of Processing by Prediction Device> The processing executed by the prediction device 100 will be described in detail below.

[0062] Fig. 7 is a flowchart showing the procedure of the prediction process executed in the prediction device 100. The process of the prediction device 100 shown in the flowchart in Fig. 7 is stored as a program in the storage 140 of the prediction device 100, and is executed by the CPU 110 controlling each unit.

[0063] (Step S101) The prediction device 100 first acquires, for example, first condition information, second condition information, third condition information, and model condition information. The prediction device 100 may acquire at least two of the first condition information, the second condition information, and the third condition information.

[0064] (Step S102) Next, the prediction device 100 divides each of the first condition information, the second condition information, the third condition information, and the model condition information acquired in the process of step S101 into divided regions each having a predetermined time width.

[0065] (Step S103) Next, the prediction device 100 determines a first feature amount, a second feature amount, a third feature amount, and a model feature amount based on the first condition information, the second condition information, the third condition information, and the model condition information acquired in the processing of step S101, respectively.

[0066] (Step S104) Next, the prediction device 100 extracts a first feature amount, a second feature amount, a third feature amount, and a model feature amount for each divided region partitioned in the process of step S102. The first feature amount, the second feature amount, the third feature amount, and the model feature amount are the feature amounts determined in the process of step S103.

[0067] (Step S105) The prediction device 100 inputs the first feature amount, the second feature amount, the third feature amount, and the model feature amount extracted in the process of step S104 into a classifier that has been trained in advance through machine learning, and predicts the properties of the molded product made from the resin material. For example, the classifier is trained in advance through machine learning using a large amount of training data prepared in advance through a learning method such as that described below. The input data for the classifier are features extracted from training condition information related to the injection molding conditions detected by the first sensor 200, the second sensor 300, and the third sensor 400 when each of the multiple resin materials is injection molded. The output data of the classifier are actual measurement values ​​of the properties of each of the molded products obtained by this injection molding.

[0068] The information input to the classifier is not limited to the first feature amount, the second feature amount, the third feature amount, and the model feature amount. For example, in addition to the first feature amount, the second feature amount, and the like, information on the resin material, the fibrous material, and the like may be input to the classifier and used as information for learning and prediction.

[0069] (Step S106) The prediction device 100 outputs characteristic information regarding the characteristics of the molded article predicted in the process of step S105. For example, the prediction device 100 displays the weather resistance of the molded article predicted in the process of step S105 on the display unit 160 together with information on the resin material and the fibrous material.

[0070] <Learning Process> Next, a machine learning method for a trained model used in a classifier will be described.

[0071] FIG. 8 is a flowchart showing a machine learning method for a trained model.

[0072] In the process of FIG. 8 , machine learning is performed using a large number (i sets) of data sets prepared in advance as training sample data. A learning device (not shown) functioning as a classifier may be, for example, a standalone high-performance computer using a CPU and GPU processor, or a cloud computer. Below, a learning method using a neural network configured by combining perceptrons such as deep learning in the learning device will be described, but this is not limiting, and various other methods may be applied. For example, random forests, decision trees, support vector machines (SVMs), logistic regression, k-nearest neighbors, topic models, etc. may be applied.

[0073] (Step S111) The learning device reads learning sample data, which is teacher data. If it is the first time, the first set of learning sample data is read, and if it is the i-th time, the i-th set of learning sample data is read.

[0074] (Step S112) The learning device inputs the input data from the read learning sample data to the neural network.

[0075] (Step S113) The learning device compares the prediction result of the neural network with the correct answer data.

[0076] (Step S114) The learning device adjusts the parameters based on the comparison results. For example, the learning device adjusts the parameters by performing a process based on back-propagation (back-propagation) so that the difference in the comparison results becomes smaller.

[0077] (Step S115) If the learning device has completed processing of all data from the first to i-th sets (YES), the process proceeds to step S116; if not (NO), the process returns to step S111, the next learning sample data is read, and the process from step S111 onwards is repeated.

[0078] (Step S116) The learning device determines whether or not to continue learning. If it decides to continue (YES), the process returns to step S111, and it executes the processes from the first set to the i-th set again in steps S111 to S115. If it decides not to continue (NO), the process proceeds to step S117.

[0079] (Step S117) The learning device stores the trained model constructed in the processing up to this point and ends the processing (END). Storage destinations include the internal memory of the prediction device 100. In the processing of FIG. 7 described above, the trained model generated in this manner is used to predict the characteristics of a molded product made of a resin material.

[0080] <Effects of the Prediction Device and Prediction System> The prediction system of this embodiment includes a first sensor 200, a second sensor 300, and a third sensor 400, and detects injection molding conditions during injection molding of a resin material over a predetermined detection period. The prediction device 100 acquires first condition information, second condition information, and third condition information related to the injection molding conditions, as well as model condition information, and divides each into divided regions of a predetermined time width. The prediction device 100 then extracts first feature values, second feature values, third feature values, and model feature values ​​for each divided region, and predicts the characteristics of a molded product made of a resin material based on the extracted first feature values, second feature values, third feature values, and model feature values. This makes it possible to predict the characteristics of a molded product from the manufacturing process. The effects of this are described below.

[0081] The properties of a molded product made of a resin material are measured, for example, by destructive testing. In other words, it has been difficult to understand the properties of a molded product until a destructive test is conducted on the molded product. This has required a lot of time and effort to understand the properties of a molded product made of a resin material. In contrast, the prediction system and prediction device 100 of this embodiment predict the properties of a molded product made of a resin material from the injection molding conditions detected by the first sensor 200, the second sensor 300, and the third sensor 400. This makes it possible to understand the properties of the molded product earlier and more easily.

[0082] Furthermore, the prediction device 100 can determine the first feature, the second feature, the third feature, or the model feature using principal component analysis, kernel principal component analysis, PLS, a statistical method, a deterministic algorithm, or a deep learning model. This allows for effective extraction of feature values ​​from each of the divided regions separated by a predetermined time width. This makes it possible to predict the characteristics of the molded product with higher accuracy.

[0083] Furthermore, the prediction device 100 can predict the weather resistance of a molded product made of a resin material, thereby enabling the user to easily understand the weather resistance, which is an important property of the molded product.

[0084] In addition, the prediction device 100 can predict the characteristics of a molded product by taking into account model feature quantities extracted using model-based development technology. This allows for an increased number of injection molding conditions to be simulated, making it possible to predict the characteristics of a molded product with higher accuracy.

[0085] As described above, the prediction device 100 and prediction system of this embodiment make it possible to predict the properties of a molded product made of a resin material.

[0086] The effects of the present invention will be explained using the following examples, although the technical scope of the present invention is not limited to the following examples.

[0087] (Creating a Trained Classifier) ​​First, training data was created and trained on the classifier.

[0088] Example 1 First, a fiber-reinforced resin was injection-molded using a plunger-type injection molding machine. Specifically, a glass fiber-reinforced polycarbonate resin (Iupilon™ GSH2040KR, manufactured by Mitsubishi Engineering) was used as the fiber-reinforced resin to produce JIS K7139 Type A1 test specimens. Test specimens were molded under different conditions by varying the plasticization temperature, mold temperature, injection speed, VP switching position, holding pressure, holding pressure time, and cooling time during molding. Both ends of the produced test specimens were cut off to adjust the length to 80 mm. Subsequently, a bending test was performed based on the method specified in JIS K7171 to determine the elastic modulus of each test specimen. Furthermore, a bending test was performed before and after storing the test specimens in an environment of 85°C and 85% RH for 200 hours to determine the elastic modulus. The ratio of the bending modulus before and after storage was defined as weather resistance. During injection molding, the first sensor 200 detected the plunger movement speed, and the second sensor 300 detected the plunger pressure. Specifically, the first condition information was information regarding the plunger movement speed over a predetermined detection period, and the second condition information was information regarding the pressure inside the cylinder over the predetermined detection period. The prediction device 100 divided the first condition information into 25 divided regions for each predetermined time interval and extracted a first feature value for each divided region. The prediction device 100 divided the second condition information into 25 divided regions for each predetermined time interval and extracted a second feature value for each divided region. The first feature value and the second feature value were statistical quantities. Specifically, the first feature value and the second feature value were the maximum values ​​measured by the first sensor 200 and the second sensor 300 in each divided region. The first feature value and the second feature value were input to a trained classifier to predict the flexural modulus and weather resistance of a molded product made of a resin material.

[0089] Example 2 The flexural modulus and weather resistance of a molded article made of a resin material were predicted in the same manner as in Example 1, except that the pressure inside the mold was detected by the second sensor 300 .

[0090] (Example 3) Furthermore, the flexural modulus and weather resistance of a molded product made of a resin material were predicted in the same manner as in Example 2, except that the third sensor 400 detected elastic waves inside the cylinder. That is, the third condition information was information about elastic waves inside the cylinder over a predetermined detection period. The prediction device 100 divided the third condition information into 25 divided regions for each predetermined time width and extracted a third feature value for each divided region. The third feature value was a statistical quantity. Specifically, the third feature value was the maximum value of the amplitude of the measurement value measured by the third sensor 400 in each divided region.

[0091] Example 4 Except for changing the third feature amount, the flexural modulus and weather resistance of a molded article made of a resin material were predicted in the same manner as in Example 3. In Example 4, the third feature amount was extracted from the average value of the measurement values ​​in each divided region by principal component analysis.

[0092] Example 5 The flexural modulus and weather resistance of a molded article made of a resin material were predicted in the same manner as in Example 4, except that the second and third feature quantities were changed. In Example 5, the second feature quantity was extracted using a deterministic algorithm from the average values ​​of the measured values ​​in each divided region. Curve fitting was used as the deterministic algorithm. In Example 5, the third feature quantity was extracted using a deep learning model from the average values ​​of the measured values ​​in each divided region. CVAE was used as the deep learning model.

[0093] Example 6 The flexural modulus and weather resistance of a molded article made of a resin material were predicted in the same manner as in Example 5, except that the first feature amount and the second feature amount were changed. In Example 6, the first feature amount and the second feature amount were parameters related to the length of time. Specifically, the first feature amount and the second feature amount were the length of time from the start of measurement to the time when the measured value reached its maximum value in each divided region.

[0094] Example 7 The flexural modulus and weather resistance of a molded product made of a resin material were predicted in the same manner as in Example 6, except that model features extracted from model condition information were used. The model condition information included information related to the plunger movement speed and information related to the pressure inside the mold. The prediction device 100 divided the model condition information into 25 divided regions for each predetermined time interval and extracted model features for each divided region. The model features were parameters related to the length of time. Specifically, the model features were the length of time from the start of measurement to the time when the simulation value reached its maximum value in each divided region.

[0095] (Comparative Example) The flexural modulus of a molded article made of a resin material was predicted in the same manner as in Example 1, except that the second condition information was not acquired and the first condition information was not partitioned into divided regions. In the comparative example, weather resistance was not predicted.

[0096] Table 1 below shows the errors in the predicted flexural modulus and weather resistance for each of Examples 1 to 7 and the Comparative Example. In Table 1, the errors are indicated by the coefficient of determination. A coefficient of determination of 0.9 or more is A, a coefficient of determination of 0.8 or more but less than 0.9 is B, a coefficient of determination of 0.7 or more but less than 0.8 is C, a coefficient of determination of 0.6 or more but less than 0.7 is D, a coefficient of determination of 0.5 or more but less than 0.6 is E, and a coefficient of determination of less than 0.5 is F.

[0097]

[0098] Comparing the comparative example with Examples 1 to 5, Examples 1 to 5, which used statistics for each divided region, showed improved prediction accuracy. Furthermore, among Examples 1 to 7, Examples 3 to 7, which used the third condition information, were able to reduce errors compared to Examples 1 and 2. Furthermore, Example 7, which used model condition information, was able to reduce errors compared to Examples 3 to 6.

[0099] The configurations of the prediction device 100 and the prediction system described above are the main configurations described in order to explain the features of the above-mentioned embodiments and examples, but are not limited to the above configurations and may be modified in various ways within the scope of the claims. Furthermore, configurations that are included in general prediction systems are not excluded.

[0100] For example, the prediction system may include at least two sensors, and may include four or more sensors.

[0101] Furthermore, the prediction device 100 may include components other than the above-described components, or may not include some of the above-described components.

[0102] Furthermore, the prediction device 100, the first sensor 200, and the second sensor 300 may each be configured by a plurality of devices, or may each be configured by a single device.

[0103] Furthermore, the functions of each component may be realized by other components. For example, the first sensor 200 or the second sensor 300 may be integrated into the prediction device 100, and some or all of the functions of the first sensor 200 and the second sensor 300 may be realized by the prediction device 100.

[0104] Furthermore, the processing units of the flowcharts in the above embodiments are divided according to the main processing content in order to facilitate understanding of each process. The classification of the processing steps does not limit the scope of the present invention. Each process can be divided into more processing steps. Furthermore, one processing step may execute more processes.

[0105] The means and methods for performing various processes in the systems according to the above-described embodiments can be realized by either dedicated hardware circuits or programmed computers. The programs may be provided, for example, on computer-readable recording media such as flexible disks and CD-ROMs, or online via a network such as the Internet. In this case, the programs recorded on the computer-readable recording media are typically transferred to and stored in a storage unit such as a hard disk. The programs may also be provided as standalone application software, or may be incorporated into the software of the device as a function of the system.

[0106] While embodiments of the present invention have been described and illustrated in detail, the disclosed embodiments are made for purposes of illustration and example only and are not intended to be limiting, and the scope of the present invention should be construed by the language of the appended claims.

[0107] This application is based on a Japanese patent application (Patent Application No. 2024-113846) filed on July 17, 2024, the disclosure of which is incorporated herein by reference in its entirety.

[0108] REFERENCE SIGNS LIST 100 Prediction device, 110 CPU, 111 Acquisition unit, 112 Partition unit, 113 Determination unit, 114 Extraction unit, 115 Prediction unit, 116 Output unit, 120 ROM, 130 RAM, 140 Storage, 150 Communication interface, 160 Display unit, 170 Operation acceptance unit, 200 First sensor, 300 Second sensor, 400 Third sensor.

Claims

1. A prediction device comprising: an acquisition unit that acquires, when injection molding a resin material, first condition information relating to injection molding conditions detected by a first sensor over a predetermined detection period, and second condition information relating to injection molding conditions detected by a second sensor different from the first sensor over the detection period; a partitioning unit that partitions each of the acquired first condition information and second condition information into divided regions each having a predetermined time width; an extraction unit that extracts a first feature amount from the first condition information for each divided region, and extracts a second feature amount from the second condition information for each divided region; and a prediction unit that predicts the characteristics of a molded product made from the resin material by injection molding, based on the extracted first feature amount and second feature amount.

2. The prediction device described in claim 1, wherein the resin material is injection molded by an injection molding device including a cylinder and a mold, the first condition information is information regarding changes in physical quantities within the cylinder over the detection period, and the second condition information is information regarding changes in physical quantities within the mold over the detection period.

3. The prediction device according to claim 1, further comprising a determination unit that determines the first feature amount and the second feature amount, and the extraction unit extracts the determined first feature amount and second feature amount.

4. The prediction device according to claim 3, wherein the determination unit determines at least one of the first feature and the second feature using at least one of principal component analysis, kernel principal component analysis, partial least squares regression, a statistical method, a deterministic algorithm, and a deep learning model.

5. The prediction device according to claim 1, wherein at least one of the first feature amount and the second feature amount is a parameter relating to the length of time determined for each of the divided regions.

6. The prediction device according to claim 1, wherein the acquisition unit further acquires model condition information generated using a model-based development technique, the partitioning unit partitions the acquired model condition information into divided regions each having a predetermined time width, the extraction unit extracts model features for each divided region from the model condition information, and the prediction unit predicts the characteristics of a molded product made of the resin material based on the first feature, the second feature, and the model feature.

7. The prediction device described in claim 1, wherein the acquisition unit further acquires third condition information detected by a third sensor different from the first sensor and the second sensor over the detection period during the injection molding, the partitioning unit partitions the third condition information into divided areas each having a predetermined time width, the extraction unit extracts a third feature amount for each divided area from the third condition information, and the prediction unit predicts the characteristic based on the extracted first feature amount, second feature amount, and third feature amount.

8. The prediction device according to claim 1, wherein the prediction unit includes a machine learning model.

9. The prediction device according to claim 1, wherein the resin material includes a fiber-reinforced resin material.

10. The prediction device according to claim 1, wherein the properties of the molded product include at least one of flexural modulus and weather resistance.

11. The prediction device according to claim 1, further comprising an output section for outputting characteristic information relating to the predicted characteristics of the molded product.

12. A prediction system comprising the prediction device according to claim 1, the first sensor and the second sensor.

13. A prediction program for causing a computer to execute a process including: acquiring first condition information relating to injection molding conditions detected by a first sensor over a predetermined detection period when injection molding a resin material, and second condition information relating to injection molding conditions detected by a second sensor different from the first sensor over the detection period; dividing the acquired first condition information and second condition information into divided regions each having a predetermined time width; extracting a first feature amount for each divided region from the first condition information, and extracting a second feature amount for each divided region from the second condition information; and predicting the characteristics of a molded product made from the resin material by injection molding based on the extracted first feature amount and second feature amount.

Images