Apparatus for quality evaluation, method for quality evaluation, program, and recording medium.

Abstract

This quality prediction device comprises: a probabilistic prediction model generation unit (66) that generates a probabilistic prediction model on the basis of a plurality of pieces of learning data in which a log related to the molding operation conditions or the state of a molding machine and molded-product quality values corresponding to the log are associated; and a quality prediction unit (61) that uses the probabilistic prediction model to calculate, from the log, a quality index for the molded product.

Description

Technical Field

[0001] The present disclosure relates to a quality prediction device, its method, program, and recording medium.

Background Art

[0002] Patent Document 1 discloses that a microcomputer built into an injection molding machine captures monitor items, compares measured values with upper and lower limit values of the monitor items, and performs quality determination of molded products.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] It is beneficial to provide an index that contributes to determining the necessity of inspecting molded products. Merely point-estimating the quality of molded products (e.g., weight, presence or absence of sink marks) may lead to confusion in determining the necessity of inspection, and there is a risk that the judgment burden will not be reduced.

Means for Solving the Problems

[0005] A quality prediction device according to one aspect of the present disclosure includes a probabilistic prediction model generation unit that generates a probabilistic prediction model based on a plurality of learning data in which logs related to molding operation conditions or states of a molding machine are associated with quality values of molded products corresponding to the logs, and a quality prediction unit that calculates a quality index of a molded product from the logs using the probabilistic prediction model.

[0006] The molding machine may be an injection molding machine, but may also be other types of molding machines such as a die-casting molding machine.

[0007] The quality indicator may represent the expected value and variance of the molded product, the probability of the desired quality occurring in the molded product, or the probability density distribution of the quality of the molded product.

[0008] Quality predictions can be made each time the molding machine produces a molded product. Therefore, a probabilistic prediction model with low computational cost can be adopted. From this perspective, a probabilistic prediction model calculated based on Bayesian linear regression is preferably adopted.

[0009] A quality prediction method according to another aspect of this disclosure generates a probabilistic prediction model based on a plurality of training data in which logs relating to the molding operation conditions or state of a molding machine and the quality values ​​of molded products corresponding to said logs are associated, and calculates a quality index of molded products from the logs using the probabilistic prediction model.

[0010] A program for implementing this method is similarly understandable, and its concept is substantially disclosed herein. The program may be downloaded from a server or recorded and distributed on non-temporary recording media (e.g., optical discs, magnetic discs, hard disks, semiconductor memory, etc.). Non-temporary recording media are tangible objects that do not involve communication lines through which such programs are temporarily propagated as data.

[0011] In some embodiments, a determination unit is provided that determines the need for inspection of a molded product based on quality indicators of the molded product. The determination parameters used in the determination unit may be adjustable. A confusion matrix may be used to adjust the determination parameters. The confusion matrix may be a classification of the predicted molding quality results based on a probabilistic prediction model against the true molding quality. [Effects of the Invention]

[0012] According to one aspect of this disclosure, it is possible to facilitate the determination of the necessity of inspecting molded products. [Brief explanation of the drawing]

[0013] [Figure 1] This is a schematic diagram showing the general configuration of an injection molding machine according to one aspect of the present disclosure. [Figure 2] This is a schematic block diagram mainly showing the control unit of an injection molding machine. [Figure 3] This figure shows the relationship between probability density and threshold. [Figure 4] This figure shows an example of a probability density distribution. [Figure 5] This is a schematic flowchart illustrating the operation of the control unit of an injection molding machine. [Figure 6] This is a schematic flowchart illustrating the simulation procedure. [Modes for carrying out the invention]

[0014] The following describes non-limiting embodiments and features of the present invention with reference to Figures 1 to 6. Those skilled in the art will understand that each embodiment and / or feature can be combined without needing excessive explanation, and that the synergistic effects of such combinations will also be understandable. Duplication of explanation between embodiments will be omitted in principle. The reference drawings are primarily for describing the invention and have been simplified for ease of drawing. Each feature is not valid only for the quality prediction apparatus and method disclosed herein, but is understood to be a universal feature applicable to various other quality prediction apparatus and method not disclosed herein.

[0015] As shown in Figure 1, the injection molding machine 1 has a clamping device 2 and an injection device 3 mounted on a common or different base 4. The injection molding machine 1 continuously produces molded products based on the coordinated operation of the clamping device 2 and the injection device 3. The clamping device 2 is configured to repeat a loop of mold closing, clamping, and mold opening. The injection device 3 is configured to repeat a loop of metering, filling, and holding pressure processes. A mold device 5 is attached to the clamping device 2. The specific configuration of the mold device 5 is determined by the shape, size, and number of injection molded products. The mold device 5 may be a two-plate or three-plate type. In some forms, the mold device 5 has one or more fixed molds 51 and one or more movable molds 52.

[0016] The configuration and operation of the clamping device 2 and the injection device 3 will be described in more detail below. The clamping device 2 includes a fixed platen 21, a movable platen 22, a toggle mechanism 23, a toggle support 24, multiple tie bars 25, a clamping motor 26, and a mold thickness adjustment mechanism 27. The toggle support 24 and the movable platen 22 are connected via the toggle mechanism 23, and the movable platen 22 can move forward and backward relative to the fixed platen 21 based on the operation of the toggle mechanism 23. Specifically, the operation of the clamping motor 26 changes the state of the toggle mechanism 23, and the position of the movable platen 22 changes. When the distance between the fixed platen 21 and the movable platen 22 is large, the mold device 5 can be introduced into the space between the fixed platen 21 and the movable platen 22. In this state, the fixed and movable molds 51 and 52 are attached to the fixed and movable platens 21 and 22, respectively. After this, the movable platen 22 is moved toward the fixed platen 21, the mold device 5 is closed, then clamped, and finally opened. In the closed state, the opposing surfaces of the fixed mold 51 and the movable mold 52 are in contact, and the semi-cavities of the fixed mold 51 and the movable mold 52 are spatially connected. In the clamped state, the movable mold 52 is pressed firmly against the fixed mold 51 to withstand the injection pressure of the material from the injection device 3. In the open state, the opposing surfaces of the fixed mold 51 and the movable mold 52 are not in contact, and a gap is left between them.

[0017] The toggle mechanism 23 has a crosshead 23a that receives a driving force from the clamping motor 26, first and second links 23b, 23c pivotally coupled between the toggle support 24 and the movable platen 22, and a third link 23d that couples between the crosshead 23a and the first link 23b. The rotational force generated by the clamping motor 26 is converted into a linear thrust by a force conversion device such as a ball screw 262 via a belt 261 and applied to the crosshead 23a. For example, as the output shaft of the clamping motor 26 rotates forward, the crosshead 23a moves straight toward the fixed platen 21, the angle formed by the first link 23b and the second link 23c increases, and the movable platen 22 moves straight toward the fixed platen 21. As the output shaft of the clamping motor 26 rotates reversely, the crosshead 23a is moved in a direction away from the fixed platen 21, the angle formed by the first link 23b and the second link 23c decreases, and the movable platen 22 moves straight in a direction away from the fixed platen 21. In the clamping device 2, the direction in which the movable platen 22 and the movable mold 52 attached thereto move toward the fixed platen 21 and the fixed mold 51 attached thereto can be defined as the front side or the injection device side, and the opposite direction can be defined as the rear side or the reflected molding device side.

[0018] The toggle mechanism 23 operates to multiply the thrust applied to the crosshead 23a and transmit it to the movable platen 22. The magnification is also called the toggle ratio. The toggle ratio changes according to the angle formed by the first link 23b and the second link 23c. As the angle formed by the first link 23b and the second link 23c approaches 180°, the toggle ratio also increases.

[0019] The mold thickness adjustment mechanism 27 is configured to adjust the position of the toggle support 24 relative to the fixed platen 21 (the front-back distance between the two, so-called mold thickness). The mold thickness adjustment mechanism 27 includes a mold thickness adjustment motor 27a. The rotational force generated by the mold thickness adjustment motor 27a is transmitted via a belt 271 to a nut screwed onto the screw shaft at the rear end of the tie bar 25, changing the position of the toggle support 24 along the tie bar 25, and changing the position of the toggle support 24 relative to the fixed platen 21 (that is, the distance between the two). The rotational force of the mold thickness adjustment motor 27a is transmitted to the nut via transmission elements such as belts and gears (or directly).

[0020] The mold clamping device 2 includes an ejector device 28 for discharging the molded product from the mold device 5. The ejector device 28 is attached, for example, behind the movable platen 22. The ejector device 28 includes an ejector rod and an ejector motor that supplies power to the ejector rod. The rotational force generated by the ejector motor is converted into a linear force by a ball screw and transmitted to the ejector rod. When the ejector rod is advanced, the ejector plate of the mold device 5 is thereby pushed. The molded product of the movable mold 52 is pushed by the ejector pins and discharged from the mold device 5. The injection molding machine 1 operates the ejector device in synchronization with mold opening.

[0021] The injection device 3 supplies a molten resin material to the mold device 5 attached to the mold clamping device 2. The injection device can be an in-line screw type or a pre-plastification type. In this specification, the injection device is described as being of the in-line screw type, but it should not be limited to this. The injection device 3 has a cylinder 31, a screw 32, a heater 33, a metering motor 34, an injection motor 35, a movement motor 36, a guide rail 37, a first movable support 38, and a second movable support 39.

[0022] The cylinder 31 is a metal cylindrical material that houses the screw 32 and has a cylinder body 31a and a nozzle section 31b. The cylinder body 31a houses the screw 32. The nozzle section 31b has a straight flow path with a flow path diameter smaller than the flow path diameter of the cylinder body 31a and has a discharge port for discharging molten plastic material supplied from the cylinder body 31a. The cylinder body 31a has a material supply port 31c for receiving plastic material, such as pellets, supplied from a hopper 31f or an automated plastic material supply device. The pellets melt in response to the heat transmitted from the heater 33 via the cylinder body 31a and are conveyed to the front, i.e., towards the nozzle section 31b, in response to the rotation of the screw 32. As can be seen from the description below, the direction of movement of the screw 32 during filling is towards the front, and the direction of movement of the screw 32 during metering is towards the rear.

[0023] The screw 32 has a shaft and helically arranged flights on the outer circumference of the shaft, and as it rotates, it conveys solid and molten resin material to the front of the cylinder 31. The screw 32 can rotate by receiving rotational force from the metering motor 34. For example, the output shaft of the metering motor 34 and the screw 32 are mechanically connected via a belt 341. The screw 32 can also move forward (towards the nozzle 31b) and backward (away from the nozzle 31b) within the stationary cylinder 31 by receiving driving force from the injection motor 35. For example, the output shaft of the injection motor 35 is connected to the screw shaft of the ball screw 351 via a belt 353. The first movable support 38 is fixed to the nut 352 of the ball screw 351. The screw 32 is rotatably mounted on the first movable support 38. Similarly, the main body of the metering motor 34 is fixed to the first movable support 38. The first movable support 38 moves in response to the operation of the injection motor 35, causing the screw 32 and metering motor 34 to move. The first movable support 38 is movably mounted on a guide rail 37 fixed to the base 4. The direction toward the clamping device 2 can be called forward, and the direction toward the clamping device 2 can be called backward.

[0024] The cylinder 31 moves forward toward the clamping device 2 and backward toward the clamping device 2, receiving driving force from the mobile motor 36. For example, the output shaft of the mobile motor 36 is connected to the screw shaft of the ball screw 361. The second movable support 39 is coupled to the nut 362 of the ball screw 361 via an elastic member (e.g., a spring) 363. The cylinder 31 is fixed to the second movable support 39 at its rear end. The second movable support 39 and the cylinder 31 move in accordance with the operation of the mobile motor 36. The second movable support 39 is movably mounted on a guide rail 37 fixed to the base 4. Each motor may be equipped with an instrument such as an encoder. The motor is feedback controlled based on the output signal of the encoder.

[0025] A backflow prevention ring (not shown) is attached to the tip (front end) of the screw 32. The backflow prevention ring prevents the molten plastic material stored in the storage space 31e from flowing back when the screw 32 is moved towards the nozzle portion 31b within the cylinder 31.

[0026] The heater 33 is mounted on the outer circumference of the cylinder 31 and generates heat, for example, by feedback-controlled energization. The heater 33 is mounted in any manner on the outer circumference of the cylinder body 31a and / or nozzle 31b.

[0027] In summary, the operation of the injection device 3 is as follows: Heat is supplied to the cylinder 31 from the heater 33, and the pellets supplied to the cylinder body 31a via the hopper 31f are melted. The screw 32 rotates within the cylinder body 31a in response to the rotational force from the metering motor 34, and the plastic material is fed forward along the helical groove of the screw 32, in which case the plastic material is gradually melted. As the molten plastic material is supplied to the front of the screw 32, the screw 32 retracts, and the molten plastic material is stored in the storage space 31e (called the "metering process"). The rotational speed of the screw 32 is detected using the encoder of the metering motor 34. In the metering process, the injection motor 35 may be driven to apply back pressure to the screw 32 in order to limit the rapid retraction of the screw 32. The back pressure on the screw 32 is detected, for example, using a pressure detector. The screw 32 retracts to the metering completion position, a predetermined amount of molten plastic material is accumulated in the storage space 31e in front of the screw 32, and the metering process is completed.

[0028] Following the metering process, the screw 32 moves toward the nozzle section 31b from the filling start position to the filling completion position in accordance with the driving force from the injection motor 35, and the molten plastic material stored in the storage space 31e is supplied into the mold device 5 through the discharge port of the nozzle section 31b (this is called the "filling process"). The position and speed of the screw 32 are detected, for example, using the encoder of the injection motor 35. When the position of the screw 32 reaches the set position, a switch from the filling process to the holding pressure process (so-called V / P switching) is performed. The position where the V / P switching takes place is also called the V / P switching position. The set speed of the screw 32 may be changed according to the position of the screw 32, time, etc.

[0029] During the filling process, when the screw 32 reaches a set position, the screw 32 may be temporarily stopped at that set position, and then the V / P switching may be performed. Immediately before the V / P switching, instead of stopping the screw 32, the screw 32 may be moved forward or backward at a slow speed. Furthermore, the screw position detector that detects the position of the screw 32 and the screw speed detector that detects the speed of the screw 32 are not limited to the encoder of the injection motor 35, but other types of detectors can be used.

[0030] Following the filling process, the holding pressure of the plastic material in front of the screw 32 is maintained at a set pressure as the screw 32 moves forward, and the remaining plastic material is pushed into the mold device 5 (called the "holding pressure process"). This allows for the replenishment of plastic material lost due to cooling shrinkage within the mold device 5. The holding pressure is detected, for example, using a pressure detector. The set value of the holding pressure may be changed according to the elapsed time from the start of the holding pressure process. During the holding pressure process, the plastic material in the cavity within the mold device 5 is gradually cooled, and when the holding pressure process is completed, the cavity entrance is sealed with solidified plastic material. This state is called a gate seal, and prevents backflow of plastic material from the cavity. After the holding pressure process, the cooling process is started. During the cooling process, the plastic material in the cavity is solidified. To shorten the molding cycle time, the metering process for the next molding cycle may be started during the cooling process.

[0031] Following the pressure holding process, the aforementioned weighing process is carried out.

[0032] The injection molding machine 1 has a control panel 7 (see Figure 1) which houses a control system for controlling the clamping device 2 and / or the injection device 3. The control system housed in the control panel 7 sequence-controls the clamping motor 26, the ejector motor, the metering motor 34, and the injection motor 35. Based on the control of the clamping motor 26, the control system performs mold closing, clamping, and mold opening. Based on the control of the metering motor 34 and the injection motor 35, the control system performs metering, filling, and holding pressure. Based on the control of the ejector motor, the control system can eject the molded product from the movable mold 52 of the mold device 5. Based on the control of the moving motor 36, the control system can position the cylinder 31 in the appropriate position. In addition to the above controls, the control system can also control the temperature of the heater 33 and the mold device 5.

[0033] For example, in a single molding cycle, the metering process, mold closing process, mold clamping process, filling process, holding pressure process, cooling process, mold opening process, and ejection process are performed in this order. The order described here is from earliest to latest start time for each process. The filling process, holding pressure process, and cooling process are performed between the start and end of the mold clamping process. The end of the mold clamping process coincides with the start of the mold opening process. In order to shorten the molding cycle time, multiple processes may be performed simultaneously. For example, the metering process may be performed during the cooling process of the previous molding cycle, in which case the mold closing process may be performed at the beginning of the molding cycle. Also, the filling process may be started during the mold closing process. Also, the ejection process may be started during the mold opening process.

[0034] As shown in Figure 2, a log containing one or more log values ​​is input to the injection molding machine control unit 60 from the injection molding machine body 1', which includes the mold clamping device 2 and injection device 3 described above. The log value is a variable related to the injection molding conditions or state, and includes, for example, a set value set in the injection molding machine body 1', a sensing value acquired by a sensor provided in the injection molding machine body 1', or a combination thereof. It may also include data from peripheral equipment and external environmental data such as utilities, ambient temperature, and differences in resin rods.

[0035] The injection molding machine control unit 60 may be incorporated into the control panel 7 described above, or it may be provided separately from the control panel 7. The injection molding machine control unit 60 can be implemented using a computer. For example, at least one CPU (Central Processing Unit) and at least one memory (hard drive, semiconductor memory) may be provided, and a program read from the memory may be executed by the CPU to implement desired functions (for example, program modules such as the quality prediction unit 61, judgment unit 62, probabilistic prediction model generation unit 66, and judgment condition calculation unit 67 described later). It is also possible to place some or all of the functions of the injection molding machine control unit 60 on a network or cloud.

[0036] The injection molding machine control unit 60 includes a quality prediction unit 61, a determination unit 62, a data storage unit 63, a buffer unit 64, and a calculation processing unit 65. The calculation processing unit 65 includes a probabilistic prediction model generation unit 66 and a determination condition calculation unit 67. The quality prediction unit 61 predicts the quality of the molded product from logs (i.e., multiple log values) received from the injection molding machine body 1' using a probabilistic prediction model generated by the probabilistic prediction model generation unit 66. This prediction is performed, for example, by calculating the expected value and variance for one or more quality, or by calculating the probability of one or more desired quality occurring, or by calculating the probability density distribution for one or more quality. The determination unit 62 makes a determination based on the prediction results of the quality prediction unit 61 (e.g., expected value and variance, probability of desired quality occurring, and / or probability density distribution). The determination unit 62 may make a determination by selecting one of three categories: good product, defective product, and product requiring inspection, or by selecting one of two categories: whether or not inspection is required. The determination parameters (e.g., thresholds) of the determination unit 62 are appropriately set, and a determination is made in response to the user request (described later). A notification unit 68 is provided to inform the user of the prediction result of the quality prediction unit 61 and / or the determination result of the determination unit 62. The notification unit 68 informs the user of the prediction and / or determination result using images, sounds, flashing lights, etc. The notification unit 68 can be configured, for example, with various displays (e.g., liquid crystal displays), speakers, or LED array flashers.

[0037] The quality prediction performed in the quality prediction unit 61 is based on the probabilistic prediction model generated in the probabilistic prediction model generation unit 66. The judgment parameters (e.g., thresholds) used in the judgment of the judgment unit 62 may, but are not limited to, those adjusted according to the simulation results in the judgment condition calculation unit 67.

[0038] The probabilistic prediction model generation unit 66 generates a probabilistic prediction model based on the training data (and, if applicable, user requirements) stored in the data storage unit 63. The probabilistic prediction model calculates quality indicators such as expected value and variance, the probability of occurrence of the desired quality, and / or the probability density distribution. Various types of models can be used to this extent. For example, if quality is represented by continuous values, a model is generated by Bayesian linear regression. If quality is represented by discrete values, a model is generated by logistic regression. If quality is represented by a mixture of continuous and discrete values, a mixed model can be used.

[0039] For reference, here is an example of a Bayesian linear regression model. In this example of a Bayesian linear regression model, where the dependent variable y can be represented by a model in which noise with a variance of 1 / β is added to a linear combination of the dependent variable y and the explanatory variable vector x, and further assuming a multivariate Gaussian distribution as the prior distribution of the coefficients, the predictive distribution p(y|x) of the dependent variable y corresponding to the explanatory variable vector x can be calculated using a Gaussian distribution as shown in Equation 1. In Equation 1, y represents the dependent variable, x represents the explanatory variable vector, μ represents the mean, and σ represents the standard deviation. The mean μ is calculated by Equation 2. The standard deviation σ is calculated by Equation 3. However, in Equation 2, m and x are column vectors. In Equation 3, x is also a column vector. TIFF0007901065000001.tif1924 shows the mean of the posterior distribution of the coefficients calculated through learning. TIFF0007901065000002.tif1920 shows the covariance matrix of the posterior distribution of the coefficients calculated through learning.

number

[0040]

number

[0041]

number

[0042] The judgment condition calculation unit 67 performs a simulation based on the training data stored in the data storage unit 63, the probabilistic prediction model generated by the probabilistic prediction model generation unit 66, and user requests, and determines the judgment parameters (e.g., thresholds) used for the judgment of the judgment unit 62. In some cases, the judgment condition calculation unit 67 estimates the distribution of the molding quality of the next molded product using the distribution of molding quality from the training data (i.e., logs of measured molding quality). The judgment condition calculation unit 67 also estimates the judgment result for the molding quality of the next molded product based on the judgment parameters (optionally, in addition to the judgment parameters, the uncertainty of the prediction). The judgment condition calculation unit 67 can simulate a confusion matrix based on the estimated molding quality, the judgment result estimated therefor, and the true molding quality (e.g., based on the training data), and can estimate the inspection rate, good product discard rate, and defective product outflow rate.

[0043] For example, the confusion matrix is ​​represented by the following table. Cell address (1:1) A shows the number of samples in which true good products were correctly estimated as good products. Cell address (1:2) B shows the number of samples in which true defective products were incorrectly estimated as good products. Cell addresses (2:1) and (2:2) C and D show the number of samples in which inspection should be performed to determine whether they are good or bad. Cell address (1:3) E shows the number of samples in which true good products were incorrectly estimated as defective products. Cell address (2:3) F shows the number of samples in which true defective products were correctly estimated as defective products. [Table 1]

[0044] The inspection rate, the rate of discarded good products, and the rate of defective products being shipped out can be calculated as follows: Test rate = (C + D) / (Sum of A to F) Good product discard rate = E / (E+F) Outflow rate of defective products=B / (A+B)

[0045] When the judgment parameters are changed, the judgment results are changed accordingly, and the inspection rate, good product discard rate, and defective product outflow rate are changed accordingly. The judgment condition calculation unit 67 can change the judgment parameters until one or more of the inspection rate, good product discard rate, or defective product outflow rate satisfies the user requirements, and can determine the judgment parameters that satisfy the user requirements. In this way, the judgment condition calculation unit 67 calculates judgment parameters that achieve the production process required by the user by probabilistically handling the distribution of molding quality and the predicted distribution of molding quality, and the specific method is not limited to the method described above.

[0046] The data storage unit 63 is a database in which training data is stored. Log values ​​and molding quality included in the log are associated with each other and stored in the data storage unit 63. For example, training data consists of (i) one or more log values ​​and (ii) at least one quality value that are associated with each other. Both log values ​​and quality values ​​take real values, but are not necessarily limited to real values. Log values ​​can be continuous or discrete values. Quality can be continuous or discrete values.

[0047] Table 2 shows examples of non-limiting log X and quality value Y. Of course, log X and quality value Y other than those shown in Table 2 can also be used. For example, any of the log values ​​X1-X4 and quality values ​​Y1-Y4 shown in Table 2 can be excluded. Various log values ​​can be used as explanatory variables, and similarly, various quality values ​​can be used as dependent variables. Typically, log X is log values ​​X1-X n It is constructed as a combination of (where n is a natural number greater than or equal to 2) (i.e., X = (X1, X2, X3 ~ X) nThe X1-X4 shown in Table 2 are understood as non-restrictive examples regarding their individual contents and order. The number of log values ​​is also set to four for ease of understanding. In injection molding machines, the number of log values ​​is generally 10 or more, and the direction of quality changes in response to changes in log values ​​is also diverse. [Table 2]

[0048] In Table 2, for discrete quality values, the first real number "1" is assigned to the presence of burrs or sink marks, and the second real number "0" is assigned to the absence of burrs or sink marks. Discrete quality evaluation is not limited to the use of two evaluation values; it is also possible to use three or more evaluation values.

[0049] In the cases shown in Figures 3(A) and (B), the quality prediction unit 61 generates a probability density distribution of quality using a probabilistic prediction model and optionally calculates the good product rate. In Figures 3(A) and (B), the mass (g) of the molded product is selected as the quality, the lower threshold TH1 is set to 4 (g), and the upper threshold TH2 is set to 5 (g). In the case of Figure 3(A), a unimodal distribution of probability density exists between the lower and upper thresholds. In the case of Figure 3(B), the unimodal distribution of probability density and the upper threshold are superimposed. The good product probability obtained by integrating the probability density between the lower and upper thresholds is 99.999% or higher in the case of Figure 3(A), and lower in the case of Figure 3(B) (for example, 80%). The prediction results of the quality prediction unit 61 (probability density distribution and / or good product probability) are notified to the user via the notification unit 68. The user can decide whether to inspect the molded product based on this information.

[0050] The determination unit 62 makes a determination based on the prediction results of the quality prediction unit 61 (expected value and variance, probability of occurrence, probability density distribution, and / or probability of good product). A simple determination that the determination unit 62 can make is to evaluate the prediction results of the quality prediction unit 61 against one or more thresholds. For example, the probability of good product is compared with one threshold, and if it is below the threshold, it is determined that inspection is required. Alternatively, the probability of good product is compared with upper and lower thresholds, and if it exceeds the upper threshold, it is determined that it is a good product; if it is between the upper and lower thresholds, it is determined that inspection is required; and if it is below the lower threshold, it is determined that it is a defective product (i.e., no inspection is required). If the user interface is set up appropriately, the thresholds can be adjusted by the user.

[0051] The judgment parameters used in the judgment unit 62 are set to respond to user requirements regarding three aspects: (i) the percentage of defective products that end up on the market (i.e., defective product outflow rate = number of defective products / total production quantity), (ii) the percentage of molded products that are inspected (i.e., inspection rate = number of inspected products / total shipment quantity), and (iii) the percentage of good products that are discarded as defective despite having relatively good molding quality (i.e., good product discard rate = number of good products / total discard quantity). An optional feature is that the judgment condition calculation unit 67 is used to determine or adjust the judgment parameters used in the judgment unit 62.

[0052] The judgment condition calculation unit 67 calculates the probability density distribution of quality using the learning data stored in the data storage unit 63 (and optionally, past quality prediction results) and the probabilistic prediction model generated by the probabilistic prediction model generation unit 66. It then simulates what values ​​one or all of the above-mentioned defective product outflow rate, inspection rate, and good product discard rate will be by applying thresholds to the probability density distribution. The confusion matrix described above can be used to calculate the defective product outflow rate, inspection rate, and good product discard rate. Log data that formed the basis of past quality prediction results may also be used.

[0053] For example, the judgment condition calculation unit 67 receives a user request relating to one, two, or all of the three perspectives described above. The judgment condition calculation unit 67 simulates whether it is possible to set a judgment parameter for the judgment unit 62 that satisfies the user request. If it is possible, that parameter will be used in the judgment unit 62. If it is not possible, the judgment condition calculation unit 67 prompts the user to change the user request.

[0054] For example, the gap between the upper and lower thresholds is narrowed to satisfy the user's requirements regarding the inspection rate, and the defective product outflow rate at that point is calculated. If the calculated defective product outflow rate satisfies the user's requirements, it is determined that such upper and lower thresholds can be adopted.

[0055] Instead of automatically setting the judgment parameters of the judgment unit 62 using the judgment condition calculation unit 67, the probability density distribution calculated by the judgment condition calculation unit 67 can be presented to the user, allowing the user to adjust the threshold (see Figure 4). An appropriate user interface (e.g., a touch panel) is provided for adjusting the threshold. The thresholds TH1 and TH2 are adjusted by the user, and the defective product outflow rate, inspection rate, and good product discard rate are calculated accordingly.

[0056] When the determination unit 62 determines that inspection is necessary, the user inspects the molded product and measures or determines its molding quality. The molding quality measured or determined in this way is input to the buffer unit 64, stored in association with the log of when the molded product was manufactured, and then registered in the data storage unit 63. When the molding quality is registered in the data storage unit 63, the prediction model and determination parameters may be updated. In order to reduce the bias in the training data registered in the data storage unit 63, molded products that have been determined to be good or defective may also be inspected regularly or randomly. This reduces the bias in the training data registered in the data storage unit 63 and improves the reliability of the probabilistic prediction model generated by the probabilistic prediction model generation unit 66. If these regularly or randomly performed inspections are carried out based on a set inspection rate, an increase in the overall inspection rate can be prevented.

[0057] The operation of the injection molding machine control unit 60 described above will be explained with reference to Figure 5. First, a database is constructed (S1). Specifically, the learning data necessary for generating a probabilistic prediction model is stored in the data storage unit 63. Next, a probabilistic prediction model is generated (S2). Specifically, the probabilistic prediction model generation unit 66 generates a probabilistic prediction model using the learning data (data in which log values ​​and quality values ​​are associated) stored in the data storage unit 63. The probabilistic prediction model can be generated using various methods such as Bayesian linear regression and logistic regression. Next, judgment parameters are set (S3). The judgment parameters of the judgment unit 62 (e.g., thresholds) may be those specified by the user or adjusted according to the simulation results in the judgment condition calculation unit 67. In this way, the preliminary preparations for mass production of molded products are completed.

[0058] Next, injection molding is performed (S4). Specifically, the injection molding machine body 1' operates and a molded product is obtained. At the same time, a log is output from the injection molding machine body 1'. Next, the quality of the molded product is predicted (S5). Specifically, the quality prediction unit 61 predicts the quality of the molded product from the log received from the injection molding machine body 1' using the probabilistic prediction model generated by the probabilistic prediction model generation unit 66. For example, the quality prediction unit 61 calculates the expected value and variance of the quality of the molded product. In point estimation, only the expected value is estimated, but by adopting a probabilistic method, information on probability, such as variance, can also be obtained. The quality prediction unit 61 can also calculate the probability density distribution using known methods.

[0059] Next, a determination is made (S6). Specifically, the determination unit 62 makes a determination based on the prediction results of the quality prediction unit 61 (for example, the expected value and variance, the probability of the desired quality occurring, and / or the probability density distribution).

[0060] If the determination result (S7) indicates that inspection is necessary, the user inspects the molded product (S8). By inspecting the molded product, the user obtains a quality value associated with the log. Therefore, the user can register this data in the data storage unit 63 as needed (S9).

[0061] If the result of the judgment (S7) indicates that inspection is not required, the count is increased and the next injection molding (S4) is performed. Molded products are mass-produced through this process. In order to eliminate bias in the data registered in the data storage unit 63, molded products that have been regularly judged as not requiring inspection (good or defective products) may be inspected. For example, the number of cycles judged as not requiring inspection is counted with a counter, and an inspection is performed when the count value reaches a predetermined value. For example, if inspection is not required, the counter value is incremented (S10), and when the count N=10 (S11), the count value is reset to the initial value (S12), and the molded product is inspected (S8). In this way, bias in the data registered in the data storage unit 63 is reduced. As a result, the reliability of the probabilistic prediction model may be improved.

[0062] Referring to Figure 6, it is shown that the determination parameter of the determination unit 62 is determined by the determination condition calculation unit 67 to be configurable. First, the determination condition calculation unit 67 performs a simulation based on the training data stored in the data storage unit 63, the probabilistic prediction model generated by the probabilistic prediction model generation unit 66, and the user request (S11). For example, the confusion matrix described above can be used to calculate the defective product outflow rate, inspection rate, and good product discard rate. As a result, if the determination parameter (e.g., threshold) used for the determination of the determination unit 62 is configurable, it is used for the determination of the determination unit 62 (S12). If the determination parameter is not configurable, an update of the user request is requested (S13). [Explanation of Symbols]

[0063] 60: Injection molding machine control unit 61: Quality Forecasting Department 62: Judgment section 63: Data storage unit 64: Buffer section 65: Arithmetic Processing Unit 66: Probabilistic Prediction Model Generation Unit 67: Judgment condition calculation section 68: Hochi Department

Claims

1. A probabilistic prediction model generation unit generates a probabilistic prediction model based on multiple training data sets that associate logs related to the molding operation conditions or state of a molding machine with the corresponding quality values ​​of the molded product. A quality evaluation apparatus comprising a quality prediction unit that calculates a predicted distribution of quality indicators for a molded product from a log using the aforementioned probabilistic prediction model, wherein the predicted distribution of quality indicators for the molded product includes a probability density distribution of the quality of the molded product, A notification unit that informs the user of at least the probability density distribution of the quality of the molded product; A determination unit that determines whether a molded product is good, defective, or requires inspection based on the probability density distribution of the quality of the molded product and the determination parameters; and A quality evaluation apparatus further comprising one of the following: a calculation processing unit that calculates at least one of a defective product outflow rate, an inspection rate, and a good product discard rate by applying a decision parameter to a probability density distribution of quality calculated from the training data or past log data using the aforementioned probabilistic prediction model.

2. The quality evaluation apparatus according to claim 1, characterized in that the quality prediction unit further calculates the expected value and variance of the molded product, or the probability of the molded product having a desired quality.

3. The quality evaluation apparatus according to claim 1, characterized in that the judgment parameter is adjustable.

4. The quality evaluation apparatus according to any one of claims 1 to 3, characterized in that it is configured to update the probabilistic prediction model in accordance with newly acquired training data.

5. A step of generating a probabilistic predictive model based on multiple training data sets that associate logs related to the molding operation conditions or state of a molding machine with the corresponding quality values ​​of the molded product, A method for quality evaluation comprising the step of calculating a predicted distribution of quality indicators for a molded product from a log using the aforementioned probabilistic prediction model, wherein the predicted distribution of quality indicators for the molded product includes a probability density distribution of the quality of the molded product, A step of informing the user of at least the probability density distribution of the quality of the molded product; A process of determining whether a molded product is good, defective, or requires inspection based on the probability density distribution of the quality of the molded product and the judgment parameters; and A method for quality evaluation, further comprising one of the steps of calculating at least one of a defective product outflow rate, an inspection rate, and a good product discard rate by applying a decision parameter to a probability density distribution of quality calculated based on the training data or past log data using the aforementioned probabilistic prediction model.

6. A process for generating a probabilistic predictive model based on multiple training data sets that associate logs related to the molding operation conditions or state of a molding machine with the corresponding quality values ​​of the molded product, A program for causing a computer to perform a process that calculates a predicted distribution of quality indicators for molded products from logs using the aforementioned probabilistic prediction model, wherein the predicted distribution of quality indicators for molded products includes the probability density distribution of the quality of the molded product. A process that informs the user of at least the probability density distribution of the quality of the molded product; A process for determining whether a molded product is good, defective, or requires inspection based on the probability density distribution of the quality of the molded product and the judgment parameters; and A program for causing a computer to further perform one of the following processes: applying a decision parameter to a probability density distribution of quality calculated based on the training data or past log data using the aforementioned probabilistic prediction model, thereby calculating at least one of the defective product outflow rate, inspection rate, and good product discard rate.

7. A non-temporary recording medium storing the program described in claim 6.

Images