Injection molding condition search aid and method, program, and recording medium

Abstract

An injection molding condition search auxiliary device relating to an injection molding machine (1) that performs injection molding based on an input of an injection molding condition to produce a molded product, the injection molding condition search auxiliary device comprising: a prediction model generation unit (65) that generates a prediction model for predicting a quality with respect to an unknown injection molding condition on the basis of previous data in which injection molding condition data and quality data of the molded product are associated with each other, wherein the prediction model includes (i) a prediction model used to calculate a first probability distribution in which a prediction value and a variance of a quality continuously change depending on a change in an injection molding condition, and / or (ii) a prediction model used to calculate a second probability distribution in which an occurrence probability of a specific phenomenon with respect to a quality continuously changes depending on a change in an injection molding condition; and an injection molding condition determining unit (66) that determines a next injection molding condition to be set in the injection molding machine (1) based on evaluation of two or more second probability distributions with respect to two or more specific phenomena or evaluation of at least one first probability distribution and at least one second probability distribution.

Description

Technical FieldThe present disclosure relates to an injection molding condition search assist apparatus, an injection molding condition search assist method, a program, and a recording mediumPrior ArtJP 2020-49 843 A relates to an injection molding machine given a function for machine learning, and more particularly discloses an apparatus and method that assist in determining molding conditions. Specifically, a correlation between molding conditions and the quality of a molded product is learned, and in a case where a quality defect occurs, what molding condition should be adjusted is presented to a user to improve a quality defect (see paragraphs 0009 to 0010 and the like of JP 2020-49 843 A CN 1 12 101 630 A relates to a method for optimizing process parameters of an injection molding method for thin-wall plastic parts. JP 2020-75 385 A relates to a device for estimating the product state. US 2006 / 0 224 540 A1 relates to a control device for an injection molding machine. DE 10 2020 209 479 A1 relates to a method and an apparatus for parameterizing a casting method and for operating a casting system using methods of machine learning. EP 4 129 621 A1 relates to an apparatus for promoting resin molding comprising a prediction unit and a display processing unit.SUMMARY OF THE INVENTIONTechnical ProblemSince a technique of creating a model using machine learning of a relationship between injection molding conditions and a mold quality to obtain conditions for a non-defective product by creating a model assumes a sufficiently trained model, it is expected that the technique cannot be used at a stage where sufficient training data is not present. Further, since, even when sufficient training data is present, an influence of noise on shape quality cannot be taken into account, there is a concern that accuracy can be reduced.Examples of a situation where sufficient training data is not present include putting a new shape into operation. In a case where another mold is used, a relationship between injection molding conditions and a mold quality is different. For this reason, even if there is existing training data, the training data cannot be used as it is. Noise related to a shape quality is considered insignificant for a shape quality measured by various sensors, but tends to be significant for a quality evaluated by humans.In view of the above-mentioned problems, the present inventor has newly discovered that it is useful to provide an apparatus and a method of supporting a search for injection molding conditions that can promote reduction or avoidance of a limitation on the number of training data and / or reduction or avoidance of an influence of noise on a mold quality.Solution to the ProblemAn injection molding condition search assist apparatus according to an aspect of the present disclosure is an injection molding condition search assist apparatus related to an injection molding machine that performs injection molding based on input of an injection molding condition to produce a molded product. The injection molding condition search assist apparatus includes: a prediction model generation unit that generates a prediction model for predicting quality related to an unknown injection molding condition on the basis of previous data in which injection molding condition data and quality data of the molded product are associated with each other, wherein the prediction model includes (i) a prediction model used to calculate a first probability distribution in which a prediction value and a variance of quality continuously change depending on a change of an injection molding condition, and / or (ii) a prediction model used to calculate a second probability distribution in which an occurrence probability of a specific phenomenon related to quality continuously changes depending on a change of an injection molding condition; and an injection molding condition determination unit that determines a next injection molding condition to be set in the injection molding machine based on evaluation of two or more second probability distributions with respect to two or more specific phenomena or evaluation of at least one first probability distribution and at least one second probability distribution.An injection molding condition search assist method according to another aspect of the present disclosure is an injection molding condition search assist method with respect to an injection molding machine that performs injection molding based on input of an injection molding condition to produce a molded product. The molding condition search assist method includes: generating a prediction model for predicting a quality related to an unknown molding condition based on previous data in which molding condition data and quality data of the molded product are associated with each other, wherein the prediction model includes (i) a prediction model used to calculate a first probability distribution in which a prediction value and a variance of a quality continuously change depending on a change of a molding condition, and / or (ii) a prediction model used to calculate a second probability distribution in which an occurrence probability of a specific phenomenon related to a quality continuously changes depending on a change of a molding condition; and determining a next injection molding condition to be set in the injection molding machine based on evaluation of two or more second probability distributions with respect to two or more specific phenomena or evaluation of at least one first probability distribution and at least one second probability distribution.A program used to perform this method may also be equally understood, and the concept of the program is substantially disclosed in this specification. In addition to being downloaded from a server, the program may be recorded and distributed on a nonvolatile recording medium (for example, an optical disk, a magnetic disk, a hard disk, a semiconductor memory, or the like). The nonvolatile recording medium is a tangible object that does not include a communication line through which such a program is temporarily propagated as data.Advantageous Effects of the InventionAccording to an aspect of the present disclosure, there is provided an apparatus and a method of supporting a search for injection molding conditions, which may promote reduction or avoidance of a limitation on the number of training data and / or reduction or avoidance of an influence of noise on a mold quality.Brief Description of the DrawingsFIG. 1 is a schematic diagram showing a schematic configuration of an injection molding machine according to an aspect of the present disclosure. FIG. 2 is a schematic block diagram mainly showing a control unit of the injection molding machine. FIG. 3 is a schematic block diagram mainly showing a prediction model generation unit and an injection molding condition determination unit. FIG. 4 is a schematic diagram showing a first probability distribution. FIG. 5 is a schematic diagram of two different second probability distributions derived for two different qualities by a classification model derivation unit. FIG. 6 is a schematic flowchart showing an operation of the control unit of the injection molding machine.DESCRIPTION OF EMBODIMENTSNon-limiting embodiments and features of the present invention will be described below with reference to FIGS. 1 to 6. Those skilled in the art can combine the respective embodiments and / or the respective characteristics without requiring excessive description, and can also understand the synergistic effects of such combinations. Basically, the repeated description of the embodiments will be omitted. Reference drawings are intended to describe primarily the invention and are simplified for ease of illustration. Each characteristic is understood as a universal characteristic that is not only a characteristic effective for a search assist device disclosed in this specification, but also a characteristic available for various other search assist devices not disclosed in this specification.As shown in FIG. 1, an injection molding machine 1 includes a mold clamping / clamping unit 2 and an injection unit 3 mounted on a common base 4 or different bases. The injection molding machine 1 continuously produces molded products based on the cooperative operation of the mold clamping / clamping unit 2 and the injection unit 3. The mold clamping / clamping unit 2 is configured to repeat a loop of mold clamping, mold clamping / clamping, and mold opening. The injection unit 3 is configured to repeat a loop of a plasticizing process, a filling process, and a pressure holding process. A mold unit 5 is mounted on the mold clamping / clamping unit 2. The specific configuration of the molding unit 5 is determined depending on the molds, sizes, and number of injection molded products. The molding unit 5 may be a two-plate type or a three-plate type. In some embodiments, the mold unit 5 includes one or more stationary molds 51 and one or more movable molds 52.The configuration and operations of the mold clamping / clamping unit 2 and the injection unit 3 will be described in more detail. The mold clamping / clamping unit 2 includes a stationary plate 21, a movable plate 22, a toggle mechanism 23, a toggle bracket 24, a plurality of columns 25, a mold clamping / clamping motor 26, and a mold space adjustment mechanism 27. the toggle bracket 24 and the movable plate 22 are connected to each other via the toggle mechanism 23, and the movable plate 22 can move forward and backward with respect to the stationary plate 21 based on the operation of the toggle mechanism 23. Specifically, the state of the toggle mechanism 23 is changed by the operation of the mold clamping / clamping motor 26, and the position of the movable platen 22 is changed. In a case where a distance between the stationary platen 21 and the movable platen 22 is large, the mold unit 5 may be inserted into a space between the stationary platen 21 and the movable platen 22. In this state, the stationary mold 51 and the movable mold 52 are mounted on the stationary plate 21 and the movable plate 22, respectively. Thereafter, the movable platen 22 is moved toward the stationary platen 21, and the mold unit 5 is closed, then clamped, and finally opened. Mold clamping is a state in which a facing surface of the stationary mold 51 and a facing surface of the movable mold 52 are in contact with each other, and a half cavity of the stationary mold 51 and a half cavity of the movable mold 52 spatially communicate with each other. The mold clamping / clamping is a state in which the movable mold 52 is strongly pressed by the stationary mold 51 to resist the pressure of a material to be injected from the injection unit 3. The mold opening is a state in which the facing surface of the stationary mold 51 and the facing surface of the movable mold 52 are not in contact with each other, and a clearance is formed between the stationary mold 51 and the movable mold 52.The toggle mechanism 23 includes a crosshead 23 athat receives a driving force from the mold clamping / clamping motor 26, first and second links 23 band 23 cthat are pivotally coupled to each other between the toggle support 24 and the movable plate 22, and third links 23 dcoupled between the crosshead 23 aand the first links 23 b. A rotational force generated from the mold clamping / clamping motor 26 is converted into a linear thrust by a force converting device such as a ball screw 262 via a belt 261, and the thrust is applied to the crosshead 23a. For example, according to the normal rotation of an output shaft of the mold clamping / clamping motor 26, the crosshead 23 ais moved straight toward the stationary plate 21, an angle between the first and second links 23 band 23 cis increased, and the movable plate 22 is moved straight toward the stationary plate 21. The crosshead 23a is moved in a direction away from the stationary platen 21 according to the reverse rotation of the output shaft of the mold clamping / clamping motor 26, the angle between the first and second links 23b and 23c is reduced, and the movable platen 22 is moved straight in a direction away from the stationary platen 21. In the mold clamping / clamping unit 2, a direction in which the movable platen 22 and the movable mold 52 mounted on the movable platen 22 are moved toward the stationary platen 21 and the stationary mold 51 mounted on the stationary platen 21 is defined as a front side or an injection unit side, and a direction opposite thereto is defined as a rear side or a side opposite to the injection molding apparatus.The toggle mechanism 23 functions to double the thrust applied to the crosshead 23a and transmit the double thrust to the movable plate 22. A factor of the toggle mechanism 23 is also referred to as a toggle factor. The toggle factor is changed depending on the angle between the first and second links 23 band 23 c. As the angle between the first and second links 23 band 23 capproaches 180°, the toggle factor is also increased.The mold space adjustment mechanism 27 is configured to adjust the position of the toggle link bracket 24 with respect to the stationary plate 21 (a distance between the stationary plate 21 and the toggle link bracket 24 in a front-rear direction, so-called a mold space). The mold space adjustment mechanism 27 includes a mold space adjustment motor 27 a. A rotational force generated by the mold space adjustment motor 27 ais transmitted via a belt 271 to nuts screwed to screw shafts provided at rear end portions of the columns 25, and the position of the toggle bracket 24 is changed along the columns 25, so that the position of the toggle bracket 24 with respect to the stationary plate 21 (i.e., a distance between the stationary plate 21 and the toggle bracket 24) is changed. The rotational force of the mold space adjustment motor 27a is transmitted to the nuts (or directly transmitted to the nuts) via transmission members such as a belt and gears.The mold clamping / clamping unit 2 includes an ejector 28 which is used to discharge molded products from the molding unit 5. For example, the ejector 28 is mounted on the rear side of the movable plate 22. Ejector device 28 includes ejector rods and an ejector motor that applies force to the ejector rods. A rotational force generated by the ejector motor is converted into a linear force by a ball screw, and the linear force is transmitted to the ejector rods. In a case where the ejector rods are caused to move forward, an ejector plate of the molding unit 5 is pushed by the ejector rods. Molded products of the movable mold 52 are pushed by ejector pins and discharged from the molding unit 5. The injection molding machine 1 causes the ejector to operate in synchronism with the mold opening.The injection unit 3 supplies molten resin material to the mold unit 5 mounted on the mold clamping / clamping unit 2. The injection unit may be an in-line screw type or a preplasticization type. The injection unit is described in this specification as an in-line screw type, but is not limited thereto. The injection unit 3 includes a cylinder 31, a screw 32, a heater 33, a plasticizing motor 34, an injection motor 35, a moving motor 36, guide rails 37, a first movable support 38, and a second movable support 39.The cylinder 31 is a tubular member that receives the worm 32 and is made of metal, and includes a cylinder body portion 31 aand a nozzle portion 31 b. The cylinder body portion 31 aaccommodates the worm 32. The nozzle portion 31 bincludes a linear flow passage having a flow passage diameter smaller than a flow passage diameter of the cylinder body portion 31 a, and includes a discharge port that discharges molten resin material supplied from the cylinder body portion 31 a. The cylinder body portion 31 aincludes a material supply port 31 cthat receives a resin material, for example, pellets, supplied from a hopper 31 fof an automated resin material feeder. The pellets are melted by heat transferred from the heater 33 via the cylinder body portion 31 aand are transported toward the front side, that is, toward the nozzle portion 31 bin accordance with the rotation of the screw 32. As will be apparent from the following description, a moving direction of the screw 32 during filling is the front side, and a moving direction of the screw 32 during plasticizing is the back side.The screw 32 includes a shaft portion and a flight provided on an outer periphery of the shaft portion in a spiral shape, and transports a solid resin material and the molten resin material to the front side of the cylinder 31 according to the rotation thereof. For example, an output shaft of the plasticizing motor 34 and the screw 32 are mechanically connected to each other via a belt 341. Further, the scroll 32 in the stationary cylinder 31 can be moved to the front side (a side approaching the nozzle portion 31 b) and the rear side (a side away from the nozzle portion 31 b) by a driving force received from the injection motor 35. For example, an output shaft of the injection motor 35 is connected to a spindle shaft of a ball screw 351 via a belt 353. The first movable support 38 is fixed to a nut 352 of the ball screw 351. The worm 32 is rotatably mounted on the first movable support 38. Also, a main body of the plasticizing motor 34 is fixed to the first movable support 38. The first movable support 38 is moved according to the operation of the injection motor 35, and the screw 32 and the plasticizing motor 34 are moved. The first movable support 38 is movably mounted on guide rails 37 fixed to the base 4. A direction toward the mold clamping / clamping unit 2 may be referred to as a front side, and a direction away from the mold clamping / clamping unit 2 may be referred to as a back side.The cylinder 31 receives a driving force from the moving motor 36 to move forward toward the mold clamping / clamping unit 2 and move backward away from the mold clamping / clamping unit 2. For example, an output shaft of the moving motor 36 is connected to a spindle shaft of a ball screw 361. The second movable support 39 is coupled to a nut 362 of the ball screw 361 via an elastic member (for example, a spring) 363. A rear end portion of the cylinder 31 is fixed to the second movable support 39. The second movable support 39 and the cylinder 31 are moved according to the operation of the moving motor 36. The second movable support 39 is movably mounted on the guide rails 37 fixed to the base 4. An instrument such as an encoder may be incorporated into each motor. The motor is feedback controlled based on outputs from the encoder.A backflow prevention ring (not shown) is attached to a tip end (front end) of the scroll 32. In a case where the screw 32 is moved toward the nozzle portion 31 bin the cylinder 31, the backflow prevention ring suppresses the backflow of molten resin material stored in a storage space 31 e.The heater 33 is attached to an outer periphery of the cylinder 31 and generates heat in a case where feedback-controlled energization is performed, for example. The heater 33 is attached to an outer periphery of the cylinder body portion 31 aand / or the nozzle portion 31 bin any manner.An outline of the operation of the injection unit 3 will be described. Heat is applied from the heater 33 to the cylinder 31, and pellets conveyed into the cylinder body portion 31 avia the hopper 31 fare melted. The screw 32 is rotated in the cylinder body portion 31a according to a rotational force generated by the plasticizing motor 34, the resin material is fed to the front side along a spiral groove of the screw 32, and the resin material is gradually melted in this process. When the molten resin material is supplied to the front of the screw 32, the screw 32 moves backward, and the molten resin material is stored in the storage space 31 e(referred to as the "plasticizing process"). The number of rotations of the screw 32 is measured using the encoder of the plasticizing motor 34. In order to limit the sudden backward movement of the screw 32, the injection motor 35 may be driven to apply a back pressure to the screw 32 in the plasticizing process. The back pressure applied to the screw 32 is measured using, for example, a pressure detector. The screw 32 moves backward to a plasticizing completion position, a predetermined amount of molten resin material is accumulated in the storage space 31 ein front of the screw 32, and the plasticizing process is completed.Subsequent to the plasticizing process, the screw 32 is moved from a filling start position toward the nozzle portion 31 bto a filling completion position according to a driving force generated by the injection motor 35, and the molten resin material stored in the storage space 31 eis supplied to the molding unit 5 (referred to as the "filling process") via the discharge opening of the nozzle portion 31 b. The position or speed of the screw 32 is measured using, for example, the encoder of the injection motor 35. In a case where the position of the scroll 32 reaches a setting position, the switching of the filling process to the pressure holding process (so-called V / P switching) is performed. A position where the V / P switching is performed is also referred to as a V / P switching position. A set speed of the screw 32 may be changed depending on the position of the screw 32, a time, or the like.In a case where the position of the screw 32 reaches a setting position in the filling process, the screw 32 may be temporarily stopped at the setting position, and the V / P switching may then be performed. Immediately before the V / P switching, the screw 32 may move forward at a very low speed instead of stopping the screw 32, or may move backward at a very low speed. Further, a screw position detector for measuring the position of the screw 32 and a screw speed detector for measuring the speed of the screw 32 are not limited to the encoder of the injection motor 35, and other types of detectors may be used.Subsequent to the filling process, when the screw 32 is moved to the front side, a holding pressure of the resin material in front of the screw 32 is maintained at a setting pressure, and the remaining resin material is extruded into the molding unit 5 (referred to as the "pressure holding process"). Insufficient plastic material may be replenished due to cooling contraction in the mold unit 5. The holding pressure is measured using, for example, a pressure detector. A setting value of the holding pressure may be changed depending on a time elapsed since the start of the pressure holding process. The resin material filled in a cavity formed in the mold unit 5 is gradually cooled in the pressure holding process, and an inlet of the cavity is closed by the solidified resin material at the completion time of the pressure holding process. This condition is referred to as a sprue seal and the backflow of the plastic material from the cavity is prevented. After the pressure holding process, a cooling process is started. During the cooling process, the plastic material filled into the cavity is solidified. A plasticizing process of the next molding cycle may be performed in the cooling process to shorten a molding cycle time.Following the pressure holding process, the above-mentioned plasticizing process is performed.The injection molding machine 1 includes a control board 7 (see FIG. 1 ) on which a control system for controlling the mold clamping / clamping unit 2 and / or the injection unit 3 is stored. The control system stored on the control board 7 controls the mold clamping / clamping motor 26, the ejector motor, the plasticizing motor 34 and the injection motor 35 in sequence. The control system performs mold clamping, mold clamping / clamping, and mold opening based on the control of the mold clamping / clamping motor 26. The control system performs plasticizing, filling and holding pressure based on the control of the plasticizing motor 34 and the injection motor 35. The control system can eject molded products from the movable mold 52 of the molding unit 5 based on the control of the ejector motor. The control system may position the cylinder 31 at an appropriate position based on the control of the motion motor 36. The control system may also control the temperatures of the heater 33 and the molding unit 5 in addition to the above-mentioned control.For example, in a molding cycle, a plasticizing process, a mold clamping process, a mold clamping / clamping process, a filling process, a pressure holding process, a cooling process, a mold opening process, and an ejection process are performed in this order. The order mentioned here is an order in which the start times of the respective processes are earlier. The filling process, the pressure holding process, and the cooling process are performed between the start of the mold clamping / clamping process and the end of the mold clamping / clamping process. The end of the mold clamping / clamping process coincides with the start of the mold opening process. Multiple processes may be performed simultaneously to shorten the molding cycle time. The plasticizing process may be performed, for example, during a cooling process of a previous molding cycle. In this case, the mold clamping process may be performed at the beginning of the molding cycle. Further, the filling process may be started during the mold clamping process. Further, the ejection process may be started during the mold opening process.As shown in FIG. 2, an injection molding machine control unit 60 is provided so as to be capable of communicating with an injection molding machine body 1' including the mold clamping / clamping unit 2 and the injection unit 3 described above. The injection molding machine body 1' performs injection molding based on input of injection molding conditions (i.e., a combination of two or more individual setting conditions) to produce molded products. The control unit 60 for injection molding machine may be incorporated in the above-mentioned control board 7 or may be provided separately from the above-mentioned control board 7. The control unit 60 for injection molding machine may be embodied by a computer. For example, at least one central processing unit (CPU) and at least one memory (a hard disk or a semiconductor memory) are provided, and a program read from the memory is executed by the CPU so that desired functions (for example, program modules such as a prediction model generation unit 65 and an injection molding condition determination unit 66) are embodied. Some or all of the functions of the injection molding machine control unit 60 may also be placed in a network or a cloud.The injection molding machine control unit 60 includes a data storage unit 61, an injection molding condition search assist unit 64, an injection molding condition setting unit 67, and a buffer unit 68. For example, the previous data is formed of (i) injection molding condition data and (ii) at least one of quality data (data represented by consecutive values and / or discrete values) associated with each other. All of the injection molding condition data (individual setting conditions) and the quality data assume values of real numbers, but are not necessarily limited thereto. Quality values that are consecutive values may be included as the quality data. Quality indices that are discrete values may be included as the quality data. As mentioned, the injection molding conditions include two or more individual conditions (individual conditions represented by successive values and / or discrete values).As shown in Table 1, injection molding condition data X (X 1 to X 4) and quality data Y (Y 1 to Y 4) may be illustrated as a non-limiting example. Of course, injection molding condition data X and quality data Y other than those shown in Table 1 may also be used. That is, each of the setting conditions X 1 to X 4 and each of the grades Y 1 to Y 4, which are shown in Table 1, can be excluded. As a descriptive variable, various molding conditions may be used, and similar various qualities may be used as a target variable. Typically, the injection molding condition data X is formed of a combination of setting conditions (or sub-conditions) X 1 to X n( n is a natural number of 2 or more) (that is, X=(X 1, X 2 and X 3 to X n)). X 1 to X 4, shown in Table 1, are understood as non-limiting examples with respect to individual contents and orders thereof. The number of setting conditions is also set to four for ease of understanding. In the injection molding machine, since the number of setting conditions is generally ten or more, the trouble of searching for injection molding conditions is enormous. [Table 1] Table 1] [Table 1] Table 1]X 1: Filling position [mm]101416X 2: Filling rate [mm / s]98,57X 3: Pressure-holding time [s]202530X 4: Holding pressure [MPa]567Quality Y (successive values)Y 1: weight [g]566,5Y 2: dimension [mm]109,58Quality Y (discrete values)Y 3: Presence or absence of burr110Y 4: Presence or absence of sink marks001In Table 1, the occurrence probability of a specific phenomenon with respect to the quality of an injection molded product is represented by discrete values, a first real number "1" is assigned to the presence of burrs or sink marks, and a second real number "0" is assigned to the absence of burrs or sink marks. A discrete quality evaluation is not limited to an aspect in which two evaluation values are used, and three or more evaluation values may be used.The injection molding condition search assist unit 64 includes a prediction model generation unit 65 and an injection molding condition determination unit 66. The search utility 64 performs sequential model-based optimization (SMBO) (for example, Bayesian optimization) using a probabilistic prediction model with respect to a quality represented by consecutive values. In short, the search assist unit 64 ( 1) generates a prediction model used for calculating a first probability distribution, and ( 2) optimizes (maximizes or minimizes) a value (i.e., a value representing the probability of promising first injection molding condition) of a target function that is a function of an expected value (or an average value) and a variance (for example, a standard deviation) of the first probability distribution. An injection molding condition in a case where a value of the objective function is an optimum value is selected as the next injection molding condition. Additionally or alternatively, the search utility 64 performs classification prediction such as logistic regression or Gaussian process discrimination. In short, the search assist unit 64 ( 1) generates a model used for calculating a probability distribution with respect to the occurrence probability of a specific phenomenon with respect to the quality of a molded product, and ( 2) optimizes (maximizes or minimizes) a value (i.e., a value representing the probability of a promising second injection molding condition) of a target function that is a function of the occurrence probabilities of two or more probability distributions with respect to two or more specific phenomena. An injection molding condition in which the value of the objective function is an optimum value is selected as the next injection molding condition. A target function that is a function of an expected value (or an average value) and a variance (for example, a standard deviation) of the first probability distribution and a function of the occurrence probability of a probability distribution with respect to a specific phenomenon may also be used.The prediction model generation unit 65 generates a prediction model predicting quality related to an unknown injection molding condition on the basis of previous data in which injection molding condition data and quality data of molded products are associated with each other. The quality includes a quality represented by successive values and a quality represented by discrete values. Accordingly, different prediction models are generated depending on whether the quality is consecutive values or discrete values. Specifically, a model used to calculate (i) a first probability distribution in which a prediction value and a variance of quality continuously change depending on a change of an injection molding condition, or (ii) a second probability distribution in which the occurrence probability of a specific phenomenon continuously changes depending on a change of an injection molding condition is derived as the prediction model. A prediction model for (i) may be referred to as a regression model, and a prediction model for (ii) may be referred to as a classification model so as to be distinguished from each other.The injection molding condition determination unit 66 determines the next injection molding condition to be set in the injection molding machine based on the evaluation of the first probability distributions with respect to one or more qualities and / or the evaluation of two or more second probability distributions with respect to two or more specific phenomena and / or the evaluation of at least one of the first probability distributions and at least one of the second probability distributions. Specifically, the injection molding condition determination unit 66 evaluates a probability distribution using a target function that calculates a value representing the probability of a promising injection molding condition from parameters (for example, an expected value (or an average value) and a variance (for example, a standard deviation)) or a variable (for example, the occurrence probability) related to the first and / or second probability distribution. The injection molding condition determination unit 66 determines an injection molding condition obtained in a case where the value of the target function is an optimum value (the maximum or minimum value) as the next injection molding condition to be set in the injection molding machine body 1'. It is possible to determine a probabilistically promising next injection molding condition (i.e., the next injection molding condition under which molded products of better quality are highly likely to be obtained) in view of the former injection molding condition and the former quality) using such cooperation between the prediction model generation unit 65 and the injection molding condition determination unit 66.In a case where the probabilistic prediction model is used as described above, it is possible to obtain a predicted distribution of a mold quality corresponding to an unknown injection molding condition. Accordingly, it is possible to specify a shape condition under which it is expected that a quality exceeding the best quality of previous data can be probabilistically obtained. Therefore, the present disclosure is useful for searching for injection molding conditions even in a state where there is a small amount of previous data, and is particularly useful for a person who has no knowledge and experience in searching for injection molding conditions. It should be noted that the next injection molding condition can be determined independently of or with a low correlation with a quality at a stage where there is a small amount of previous data.Advantageously, as described above, the injection molding condition determination unit 66 evaluates, but is not limited to, a probability distribution using a target function that calculates a value representing the probability of a promising injection molding condition from parameters or a variable related to the first and / or second probability distributions. The evaluation of a probability distribution can also take place in a plurality of stages. For example, it is also possible to obtain a range of promising injection molding condition and then specify an injection molding condition within the range. It is also possible to obtain a plurality of promising injection molding conditions and select a desired promising injection molding condition from among the plurality of promising injection molding conditions. Various calculation methods, such as weighting, may be used to determine a value representing the probability of a promising injection molding condition.It is likely that noise is included in the value of quality. For example, the evaluation results of quality (in particular, the presence or absence of external abnormalities such as burr or sink marks) are likely to vary between a person skilled in the art and a beginner. In a case where a probabilistic prediction model is used, it is possible to determine a probabilistically promising next injection molding condition also based on an influence of such noise. Accordingly, a possibility that molded products of better quality can be obtained is increased.As shown in FIG. 3, the prediction model generation unit 65 includes a regression model derivation unit 71, a classification model derivation unit 72, and a model storage unit 73. Also, a classification model derived by the classification model deriving unit 72 is stored in the model storage unit 73. Both the regression model deriving unit 71 and the classification model deriving unit 72 may be used, or only the regression model deriving unit 71 or the classification model deriving unit 72 may be used.The regression model deriving unit 71 may generate a model used for calculating a probability distribution based on / according to a Gaussian process (accordingly, an expected value (or an average value) and a variance of quality related to a certain injection molding condition are obtained using a model). In addition to or as an alternative to the expected value, an average value may also be obtained. In a case where a Gaussian process is used, an unknown quality y with respect to an arbitrary injection molding condition x can be obtained as a probability distribution of a Gaussian distribution. That is, an expected value μ(x) and a variance σ(x) of prediction values of quality are obtained.Equation 1 shows a probability distribution model of a mold quality y new. corresponding to an unknown injection molding condition x new. D denotes known (x, y).Here,A distribution of the mold quality y new corresponding to the unknown injection molding condition x new is represented by an average (expected value) and a variance on the right side of Equation 1. k * denotes a relationship between x new and known x 1 to x N, which was expressed by a kernel function as a vector. k ** denotes a relationship between x new and x new, which was expressed by a kernel function as a scalar. K denotes a relationship between known x 1 to x N, which has been expressed as a matrix by a kernel function.An expected value of a Gaussian distribution that is predicted may be represented as in Equation 2. Likewise, a variance as shown in Equation 3 can be represented.A kernel function (e.g., a Gaussian kernel of Equation 4) is used for the calculation of a predicted distribution. A value point-estimated as a value at which a marginal probability of a Gaussian process reaches its maximum may be used as a hyperparameter θ of the kernel function, but the hyperparameter is not limited thereto. The hyperparameter may also be estimated as a random variable.An example of an image of a probability distribution calculated from the model is shown in Fig. 4. Molding conditions X 10 to X 30 and quality values corresponding to these molding conditions X 10 to X 30 are included in previous data. A solid line in FIG. 4 shows that an expected value of quality continuously changes depending on a change in the injection molding condition. A one-dot chain line in FIG. 4 schematically shows a Gaussian distribution that peaks at an expected value of quality. Two dotted lines shown on both upper and lower sides of the solid line in FIG. 4 determine a confidence interval based on a predetermined probability and determine a confidence interval with, for example, a probability of 95%. The quality has a value within the confidence interval with a predetermined probability (e.g., a probability of 95%). In a case where a Gaussian process is appropriately performed, a variance of a prediction value of quality with respect to a known injection molding condition is small, and a variance of a prediction value of quality with respect to an unknown injection molding condition is large. In a case where such a probability distribution is used (specifically, in a case where an expected value (or an average value) and a variance of quality are used), quality related to an unknown injection molding condition can be predicted. More specifically, a molding condition X next may be selected as a promising molding condition because a quality value exceeding a molding quality target value is more likely to be obtained under the molding condition X next than under the molding condition X 10. In FIG. 4, the injection molding condition looks like a variable. However, the injection molding condition is actually determined from a plurality of setting conditions. It is possible to obtain an injection molding condition under which a value to be searched is high (for example, an injection molding condition in a case where a value of a target function is an optimum value) using the target function (for example, a function for calculating a value representing the probability of a promising injection molding condition as a function of an expected value and a variance of quality to be predicted).The classification model deriving unit 72 is configured to generate a model that calculates the occurrence probability of a specific phenomenon with respect to a quality of a molded product from injection molding conditions (two or more individual setting conditions), and generates such a model based on, for example, logistic regression or a Gaussian process discriminator. In the case of logistic regression, for example, a logistic function is used to apply the occurrence probability, so that a regression curve shown in FIG. 5 is obtained. As shown in FIG. 5, the occurrence probability of a specific phenomenon continuously changes depending on a change in the injection molding condition. For example, the regression curve represents a range in which the occurrence probability of the specific phenomenon changes from a low value to a high value, which can be used for evaluating the probability distribution. It is possible to obtain an injection molding condition under which a value to be searched is high (for example, an injection molding condition in a case where a value of an objective function is an optimum value) using the objective function (for example, a function for calculating the sum or logarithmic sum of occurrence probabilities with respect to two or more specific phenomena).An example of a probability distribution related to the occurrence of an incident spot is shown in an upper row in FIG. 5. A solid line shown in the upper row in FIG. 5 indicates the occurrence probability of an incident point. With respect to an incidence point, a range determined from the shape condition value X 20 and the shape condition value X 30 is a range in which the presence of an incidence point is changed to the absence of an incidence point, and a range in which a value of quality with respect to an incidence point is changed from a defective product value (indicating a defective product) to a non-defective product value (indicating a non-defective product).An example of a probability distribution related to occurrence of a burr is shown in a lower row in FIG. 5. A solid line shown in the lower row in FIG. 5 indicates the occurrence probability of a burr. With respect to a burr, a range determined from the molding condition value X 20 and the molding condition value X 30 is a range in which the absence of a burr is changed to the presence of a burr, and a range in which a value of quality with respect to a burr is changed from a non-defective product value (indicating a non-defective product) to a defective product value (indicating a defective product).As known from FIG. 5, it may be effective to select an injection molding condition belonging to both a range in which the occurrence probability of a certain specific phenomenon is changed from a low value (for example, 0.2 or less) to a high value (for example, 0.8 or more) and a range in which the occurrence probability of another specific phenomenon is changed from a high value (for example, 0.8 or more) to a low value (for example, 0.2 or less). The quality is not limited to a quality having a correlation similar to the correlation of a burr and an sink.As described above, the injection molding condition determination unit 66 determines the next injection molding condition to be set in the injection molding machine body 1' using a target function that calculates a value representing the probability of a promising injection molding condition from parameters (for example, an expected value (or an average value) and a variance (for example, a standard deviation)) or a variable (for example, the occurrence probability) related to the first and / or second probability distributions. Specifically, a shape condition is selected in a case where a value of the objective function is an optimum value. In order to optimize (maximize or minimize) the value of the objective function, a gradient method or a Metropolitan Sharing (MCMC) method may be used as an optimization tool.For the evaluation of the first probability distribution of (i), an objective function with respect to an upper confidence bound (UCB) can be used.Here, UCB(x) denotes a value of UCB under an injection molding condition x, μ(x) denotes an expected value of the quality predicted under the injection molding condition x, σ(x) denotes a standard deviation of the quality predicted under the injection molding condition x, and k denotes a hyperparameter.Also, the following objective function (i.e., a function regarding a probability of updating an existing maximum value regarding quality (probability of improvement (PI))) may be used in addition to or as an alternative to the above-mentioned Equation 5.Here, PI(x) denotes a value of PI under the injection molding condition x, μ(x) denotes an expected value of the quality predicted under the injection molding condition x, σ(x) denotes a standard deviation of the quality predicted under the injection molding condition x, y denotes a value of the quality, and y max denotes the maximum value of the quality.In the case of Equation 5, maximizing UCB(x) means increasing the sum of an expected value and a variance (for example, a standard deviation) of the quality. Since an injection molding condition x is calculated in a case where UCB(x) has its maximum, it is expected that probabilistic molded products of better quality can be obtained. In a case where the value of k, which is a hyperparameter, is increased, a search for an unknown shape condition is prioritized. Also in Equation 6, as in Equation 5, since an injection molding condition x is obtained in a case where PI(x) has its maximum, probabilistic molded products of better quality can be obtained. Determining the injection molding condition x in this manner is equivalent to evaluating a probability distribution based on the objective function.The following objective function can be used for evaluating the second probability distribution of (ii). The product (P all( x)) of individual occurrence probabilities (P n( x)) of N (N is a natural number equal to or greater than 2) qualities under the injection molding condition x is calculated in Equation 7. A logarithmic value of the product of the occurrence probabilities of equation 7 is calculated in equation 8 and can be understood as in equation 7.In the case of Equation 7, maximizing P all( x) means that, in the N qualities, the occurrence probabilities (P n( x)) are increased as a total tendency. Since an injection molding condition x is calculated in a case where P all( x) has its maximum, it is expected that molded products of better quality with respect to the N qualities can be obtained. The same applies to equation 8 and equation 7.As known from the above description, in the present embodiment, in the case of the first probability distribution of (i), an expected value and a variance (for example, a standard deviation) are integrated into a target function (accordingly, a value of the target function (a value representing the probability of promising injection molding condition) is a function of the expected value and the variance), and in the case of the second probability distribution of (ii), the occurrence probability is integrated into a target function (accordingly, a value of the target function (a value representing the probability of promising injection molding condition) is a function of the occurrence probability). Since an injection molding condition x is calculated in a case where a value of the objective function is optimized (for example, maximized or minimized), a promising injection molding condition x that probabilistically results in molded products of better quality is determined. In a case where an objective function whose value is increased with high quality of a molded product and is reduced with low quality of a molded product is used, an injection molding condition x in a case where a value of the objective function has its maximum is calculated. In a case where an objective function whose value is reduced with high quality of a molded product and is increased with low quality of a molded product is used, an injection molding condition x in a case where a value of the objective function has its minimum is calculated.An optimization tool is used to optimize the value of the objective function. For example, the gradient method or the Metropolitan Hastings (MCMC) method may be used. In the objective function, a search range can be designated. In a case where there is a target value for a certain quality, the target value may be defined in the objective function. In a case where optimization is to be performed for plural types of qualities, the importance degrees of these qualities may be adjusted. It is also possible to maximize or minimize a quality whose set point is not present.The evaluation of the first probability distribution of (i) and the evaluation of the second probability distribution of (ii) can also be carried out using a common target function. That is, a target function related to the first probability distribution and a target function related to the second probability distribution are combined to form a common target function. This objective function is, for example, a function of UCB(x) of equation 5 and P all( x) of equation 7.As shown in FIG. 3, the injection molding condition determination unit 66 includes a target function storage unit 74 and a calculation execution unit 75. The calculation execution unit 75 integrates parameters or a variable related to a probability distribution calculated from the prediction model stored in the model storage unit 73 into the objective function, and uses the optimization tool to obtain an injection molding condition x in a case where a value of the objective function (a value representing the probability of the promising injection molding condition) is an optimum value. In this way, the next injection molding condition is determined. Specifically, an objective function that is a function of an expected value and a variance (for example, a standard deviation) is used in the case of the first probability distribution of (i), and an objective function that is a function of the occurrence probability is used in the case of the second probability distribution of (ii). A molding condition in a case where a value of the objective function is an optimum value is not necessarily a molding condition under which it is expected that molded products of better quality can be obtained. For example, a shape condition useful for improving prediction accuracy is selected at a stage where there is a small amount of previous data used to derive a prediction model.A shape condition under which quality is highly likely to exceed a target quality may be selected with respect to the first probability distribution of (i). For this purpose, for example, the objective functions shown in equations 5 and 6 are used. In the probability distribution shown in FIG. 4, an injection molding condition x next, under which quality is expected to most likely exceed the target quality, is specified and may be selected. For example, in a case where UCB is used in a state where "k=2" is satisfied, the upper limit of a confidence interval of 95% is selected. This selected condition is balanced both with respect to a distance from a known molding condition and with respect to a quality value. In a case where PI is used, an injection molding condition is selected under which quality is highly likely to exceed a known maximum value with respect to quality. As a preferable characteristic, the molding condition determination unit 66 may select a molding condition under which a variance of a prediction value of quality is relatively large as the next injection molding condition. Since a molding condition under which variance of quality is large is selected, selection of an injection molding condition that is close to a known injection molding condition with respect to a space or a coordinate space is suppressed. Accordingly, it is possible to efficiently search for an injection molding condition.A shape condition under which the number of qualities having higher occurrence probabilities is likely to increase may be selected with respect to the second probability distribution of (ii). For this purpose, the objective functions shown in equations 7 and 8 are used. An injection molding condition x under which the objective functions are maximized may be an injection molding condition belonging to both a range in which the occurrence probability of a certain specific phenomenon is changed from a low value (for example, 0.2 or less) to a high value (for example, 0.8 or more) and a range in which the occurrence probability of another specific phenomenon is changed from a high value (for example, 0.8 or more) to a low value (for example, 0.2 or less).In a case where an injection molding condition is selected for a search candidate, this injection molding condition is transmitted from the injection molding condition determination unit 66 to the injection molding condition setting unit 67 and the buffer unit 68. The injection molding condition setting unit 67 instructs the injection molding machine body 1' to operate based on the received injection molding condition for a search candidate. The injection molding machine body 1' operates based on an injection molding condition designated by the injection molding condition setting unit 67 to produce molded products. Injection molding may be performed after user confirmation. Subsequently, a user inspects molded products and inputs the quality data of the molded products into the buffer unit. In this way, injection molding condition data and quality data are stored in the buffer unit 68 in association with each other. Subsequently, a user uses input means (not shown) to store the data stored in the buffer unit 68 in the data storage unit 61. In this way, new data is added. The prediction model generation unit 65 may regenerate a prediction model using a database to which new data is added in this manner and which is updated.Finally, the operation of the injection molding machine control unit 60 will be described with reference to FIG. 6. First, a user sets a target function (S1). Next, the user operates the injection molding machine 1 to obtain and register previous information in which injection molding condition data and quality data are associated with each other (S 2). For example, the previous data shown in Table 1 is registered in a database. Next, the prediction model generation unit 65 generates a prediction model on the basis of the previous data prepared at S 2 (S 3). As described above, the prediction model may be for calculating one or both of the first or second probability distributions. Thereafter, the injection molding condition determination unit 66 determines an injection molding condition for the next search candidate using a target function that is a function of parameters or a variable of a probability distribution (S 4). Specifically, an injection molding condition x in a case where a value of the objective function is an optimum value is selected as the next injection molding condition. The injection molding condition thus determined is set by the injection molding condition setting unit 67 as an operating condition for the injection molding machine body 1' (S5), and the injection molding machine body 1' operates under this condition (S6). The qualities of molded products manufactured in this manner are evaluated by the user or a quality determination device (S7). Then, such data as previous data is added to which quality data is associated together with injection molding condition data (S8). After new previous data is added, a prediction model is again derived, a loop from S3 to S8 is repeated, and the prediction model is further updated.List of reference characters1 Injection molding machine 61 Data storage unit 65 Prediction model generation unit 66 Injection molding condition determination unit

Claims

An injection molding condition search assist apparatus related to an injection molding machine (1) that performs injection molding based on input of an injection molding condition to produce a molded product, the injection molding condition search assist apparatus comprising: a prediction model generation unit (65) that generates a prediction model for predicting quality related to an unknown injection molding condition based on previous data in which injection molding condition data and quality data of the molded product are associated with each other, wherein the prediction model (i) a prediction model used to calculate a first probability distribution in which a prediction value and a variance of quality continuously change depending on a change of an injection molding condition, and / or (ii) a prediction model used to calculate a second probability distribution, wherein a specific phenomenon occurrence probability with respect to quality continuously changes depending on a change in an injection molding condition includes; and an injection molding condition determination unit ( 66) that determines a next injection molding condition to be set in the injection molding machine ( 1) based on evaluation of two or more second probability distributions with respect to two or more specific phenomena or evaluation of at least one first probability distribution and at least one second probability distribution.The search assist apparatus for injection molding condition according to claim 1, wherein the injection molding condition determination unit (66) determines the next injection molding condition using a target function for calculating a value representing probability of promising injection molding condition from a parameter and / or a variable relating to the first and / or second probability distribution.The molding condition search assist apparatus according to claim 2, wherein the parameter includes a prediction value and a variance of the quality at the first probability distribution.The search assist apparatus for injection molding condition according to claim 2 or 3, wherein the variable includes the occurrence probability of the specific phenomenon at the second probability distribution.The search assist apparatus for injection molding condition according to any one of claims 2 to 4, wherein the value representing probability of the promising injection molding condition is a function of prediction values and variances of the qualities at the first probability distributions with respect to one or more qualities, and is a function of the occurrence probabilities at the two or more second probability distributions with respect to the two or more specific phenomena.The search assist apparatus for injection molding condition according to any one of claims 2 to 5, wherein the objective function refers to an upper confidence bound (UCB) or a probability of updating an existing maximum value with respect to a quality (probability of improvement (PI)).The search assist apparatus for injection molding condition according to any one of claims 2 to 6, wherein the objective function refers to a product of the occurrence probabilities at the two or more second probability distributions with respect to the two or more specific phenomena or to a logarithmic value of the product.The molding condition search assist apparatus according to any one of claims 1 to 7, wherein the prediction model generation unit (65) generates the first probability distribution based on a Gaussian process.The search assist apparatus for injection molding condition according to claim 8, wherein the injection molding condition determining unit (66) is configured to determine the next injection molding condition based on an upper confidence limit (UCB) or a probability of updating an existing maximum value with respect to a quality (probability of improvement (PI)).The search assist apparatus for injection molding condition according to claim 8 or 9, wherein the injection molding condition determination unit (66) is configured to select a molding condition under which a variance of a prediction value of quality is relatively large as the next injection molding condition.The injection molding condition search assist apparatus according to any one of claims 1 to 10, wherein the prediction model generation unit (65) is configured to generate a classification model that calculates a specific phenomenon occurrence probability from an injection molding condition on the basis of logistic regression or a Gaussian process discriminator.The search assist apparatus for injection molding condition according to claim 11, wherein the injection molding condition determination unit (66) is configured to determine the next injection molding condition based on a product of the occurrence probabilities at the two or more second probability distributions with respect to the two or more specific phenomena or a logarithmic value of the product.The search assist apparatus for injection molding condition according to claim 11 or 12, wherein the injection molding condition determination unit (66) is configured to select, as the next injection molding condition, an injection molding condition belonging to both a range in which an occurrence probability of a certain specific phenomenon is changed from a value of 0.2 or less to a value of 0.8 or more and a range in which an occurrence probability of another specific phenomenon is changed from a value of 0.8 or more to a value of 0.2 or less.An injection molding condition search assist method with respect to an injection molding machine (1) performing injection molding based on input of an injection molding condition to produce a molded product, the injection molding condition search assist method comprising: generating a prediction model for predicting quality with respect to an unknown injection molding condition based on previous data in which injection molding condition data and quality data of the molded product are associated with each other, wherein the prediction model comprises (i) a prediction model used to calculate a first probability distribution in which a prediction value and a variance of quality continuously change depending on a change of an injection molding condition, and / or (ii) a prediction model used to calculate a second probability distribution, wherein a probability of occurrence of a specific phenomenon with respect to quality continuously changes depending on a change of an injection molding condition includes; and determining a next injection molding condition to be set at the injection molding machine (1) based on evaluation of two or more second probability distributions with respect to two or more specific phenomena or evaluation of at least one first probability distribution and at least one second probability distribution.A program for causing a computer to perform processing of: generating a prediction model for predicting a quality related to an unknown injection molding condition based on previous data in which injection molding condition data and quality data of the molded product are associated with each other, wherein the prediction model includes (i) a prediction model used to calculate a first probability distribution in which a prediction value and a variance of a quality continuously change depending on a change of an injection molding condition, and / or (ii) a prediction model used to calculate a second probability distribution in which an occurrence probability of a specific phenomenon related to a quality continuously changes depending on a change of an injection molding condition; and determining a next injection molding condition to be set in an injection molding machine (1) based on evaluation of two or more second probability distributions with respect to two or more specific phenomena or evaluation of at least one first probability distribution and at least one second probability distribution.A nonvolatile recording medium on which the program according to claim 15 is stored.

Images