Production management system

Abstract

To optimize manufacturing parameters to maintain a state in which non-defective products are produced, and to quickly restore conditions for producing the non-defective products even if a defective product occurs due to a sudden disturbance or the like.SOLUTION: A dataset classification part 22 extracts data and measurement values during production of products determined to be non-defective as a non-defective dataset. Furthermore, the dataset classification part 22 extracts data and measurement values during production of products determined to be defective as a defective dataset. A regression tree processing part 24 creates a regression tree for the non-defective dataset and extracts explanatory variables with relatively high importance. Furthermore, the regression tree processing part 24 creates a regression tree for the defective dataset and extracts explanatory variables with relatively high importance. A parameter discrimination part 26 extracts, from the extracted explanatory variables, those explanatory variables that are correlated with increases and decreases in control setting values of an injection molding device 60 as control parameters.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] This specification discloses a system for managing production at a manufacturing site. [Background technology]

[0002] For example, Patent Document 1 discloses a control device that controls a manufacturing device. This control device includes a molding condition calculation unit and an acquisition unit. The molding condition calculation unit derives input target variables that are input to the manufacturing device in order to control the manufacturing device. The acquisition unit acquires non-input target variables, which are variables other than the input target variables. The non-input target variables include values ​​related to the production environment during the manufacturing of the object to be manufactured (disturbance values, etc.) and actual measured values ​​of the manufacturing conditions of the manufacturing device. The molding condition calculation unit derives the input target variables by referring to the non-input target variables.

[0003] Patent Document 2 discloses an information processing device including a derivation unit and an update unit. The derivation unit derives input target variables to be input to a manufacturing device in order to control the manufacturing device. During this derivation, the derivation unit derives the input target variables using a prediction model that receives the values ​​of the input target variables and outputs predicted values ​​of quality information related to the quality of the object to be manufactured, so that the predicted values ​​of the quality information satisfy predetermined conditions. The update unit updates the prediction model by referring to the actual measured values ​​of quality information related to the actual quality of the object to be manufactured using the input target variables derived by the derivation unit and the predicted values ​​of the quality information.

[0004] Patent Document 3 discloses an injection molding condition optimization system. This system comprises a control device, a measurement device, and an optimization calculation device. The control device automatically changes several setting condition variables (plasticizing screw rotation speed, injection speed, injection hold pressure, etc.) within a specified range for the injection molding machine while the injection molding machine is operating. The measurement device automatically measures molding evaluation items (resin temperature, cycle time, power consumption, etc.) when molding under the set conditions. The optimization calculation device accumulates data by repeatedly changing the setting condition variables a specified number of times, and performs multiple regression analysis on the data to calculate and set the optimal conditions. Patent Documents 4-6 also disclose injection molding machines. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Patent Publication No. 2021-12475 [Patent Document 2] Patent Publication No. 2021-11046 [Patent Document 3] Japanese Patent Application Publication No. 8-156060 [Patent Document 4] Japanese Patent Application Publication No. 2017-065016 [Patent Document 5] Japanese Patent Application Publication No. 2016-083775 [Patent Document 6] Japanese Patent Application Laid-Open No. 2016-107511 Summary of the Invention [Problem to be solved by the invention]

[0006] In order to reduce the rate of defective products, it is necessary not only to optimize manufacturing parameters and maintain a state in which good products are produced, but also to quickly restore the conditions for producing good products in the unlikely event that a defective product occurs due to a sudden disturbance or the like. [Means for solving the problem]

[0007] The production management system disclosed in this specification includes a production device, a collection unit, a measurement unit, a judgment unit, a classification unit, a regression tree processing unit, a parameter discrimination unit, and a display unit. The production device produces products. The collection unit collects various data during production by the production device. The measurement unit measures the products produced by the production device. The judgment unit judges the quality of the products based on the measurements by the measurement unit. The classification unit extracts data and measurement values ​​during production of products judged to be good as a good product dataset. The classification unit further extracts data and measurement values ​​during production of products judged to be defective as a defective product dataset. The regression tree processing unit creates a regression tree for the good product dataset using the measurement values ​​as the response variable and performs good product regression tree processing to extract explanatory variables with relatively high importance. The regression tree processing unit also creates a regression tree for the defective product dataset using the measurement values ​​as the response variable and performs defective product regression tree processing to extract explanatory variables with relatively high importance. The parameter determination unit extracts, as control parameters, explanatory variables correlated with increases and decreases in the control setting values ​​of the production equipment from among the explanatory variables extracted by the good product regression tree processing and the defective product regression tree processing. The display unit displays the control parameters.

[0008] According to the above configuration, by monitoring the control parameters obtained by the good product regression tree processing, it is possible to maintain a good product production state. Furthermore, when a defective product occurs, it is possible to quickly return to the conditions for producing good products by taking measures such as moving the control parameters obtained by the defective product regression tree processing from the defective product occurrence area to outside that area based on the correlation of the control setting values.

[0009] In the above configuration, when there are multiple types of explanatory variables extracted by the good product regression tree processing and the defective product regression tree processing, the parameter determination unit may exclude some of the explanatory variables from the management parameters. In this case, the parameter determination unit excludes from the management parameters explanatory variables of relatively low importance that have an inverse correlation with the correlation between the relatively high importance explanatory variables and the control setting values ​​of the production devices.

[0010] According to the above configuration, process management does not become complicated.

[0011] In the above configuration, the production device may be an injection molding device. In this case, the measurement unit measures the chip depth of the molded product. The judgment unit judges the molded product to be defective if the chip depth is equal to or greater than a first threshold. The classification unit extracts various data of the injection molding device and the chip depth as a small data set when the chip depth is less than a second threshold that is smaller than the first threshold. The regression tree processing unit performs small regression tree processing, creating a regression tree for the small data set using the chip depth as a target variable and extracting explanatory variables with relatively high importance. The parameter discrimination unit extracts explanatory variables correlated with increases and decreases in the control setting values ​​of the injection molding device as control parameters from the explanatory variables extracted by the small regression tree processing.

[0012] If small chip depth is a condition for a good product, adjusting the control settings to reduce chip depth too much may result in an excessive amount of resin being injected, leading to overpacking. By extracting control parameters from a small data set obtained from production conditions where overpacking is possible, it is possible to prevent overpacking when improving chip depth.

[0013] The above configuration may further include an update unit that outputs an update command to the classification unit, regression tree processing unit, and parameter determination unit when the number of shots by the injection molding device exceeds an update threshold.

[0014] According to the above configuration, it is possible to extract a control parameter that reflects the most recent operating state of the injection molding machine. [Effects of the Invention]

[0015] According to the production management system disclosed in this specification, manufacturing parameters are optimized to maintain a state in which good products are produced, and even if a defective product occurs due to a sudden disturbance or the like, the system can quickly return to conditions for producing good products. [Brief explanation of the drawings]

[0016] [Figure 1]FIG. 1 is a diagram illustrating a production management system according to an embodiment of the present invention. [Figure 2] FIG. 1 illustrates an injection molding apparatus. [Figure 3] FIG. 2 is a diagram illustrating an example of functional blocks of a management device. [Figure 4] FIG. 10 is a diagram illustrating a flow for extracting control parameters from a non-defective product data set. [Figure 5] FIG. 10 is a diagram illustrating an example of a regression tree for a non-defective data set. [Figure 6] FIG. 10 is a diagram illustrating a flow for extracting management parameters from a defective product data set. [Figure 7] FIG. 10 is a diagram illustrating a flow for extracting management parameters from a small data set. [Figure 8] FIG. 10 is a diagram illustrating an example of a management screen on which management parameters are displayed. DETAILED DESCRIPTION OF THE INVENTION

[0017] 1. Overall structure The configuration of the production management system will be described below with reference to the drawings. This production management system includes a management device 10 and an injection molding device 60.

[0018] The injection molding machine 60 is an example of a production machine that produces products. As will be described later, the production management system according to this embodiment can acquire management parameters if it can acquire measurement values ​​that serve as reference values ​​for determining whether a product is good or bad and data during production.

[0019] For example, as will be described later, the production management system according to this embodiment can generate a regression tree in which the measurement values ​​that serve as the reference values ​​for pass / fail judgment are used as the objective variables and production data are used as the explanatory variables. Furthermore, the importance of each explanatory variable is calculated. The explanatory variables with the highest importance are set as management parameters. In other words, the production management system according to this embodiment can manage production as long as the production equipment is capable of acquiring the measurement values ​​that serve as the reference values ​​for pass / fail judgment and production data. For example, the production management system according to this embodiment can be equipped with various production equipment, such as a press molding machine, a laser processing machine, or a coating machine, instead of the injection molding machine 60.

[0020] 2.Injection molding equipment 2 illustrates an injection molding apparatus 60. The injection molding apparatus 60 includes an injection unit 61 and a mold clamping unit 71. The mold clamping unit 71 includes a fixed mold 72 and a movable mold 73. When the fixed mold 72 and the movable mold 73 close together, a cavity 77 is formed. When resin is injected into the cavity 77, a molded product 35 (see FIG. 1) is obtained.

[0021] The injection unit 61 includes a cylinder 62 , a screw 63 , a motor 64 , a hopper 65 , a heater 66 , and a nozzle 67 .

[0022] A screw 63 is housed in a cylindrical cylinder 62. The rear end of the screw 63 (the side opposite to the nozzle 67) is connected to a motor 64. The motor 64 rotates the screw 63 to move it forward and backward.

[0023] Resin pellets are charged into a hopper 65. The resin pellets are sent from the hopper 65 into a cylinder 62. The cylinder 62 is surrounded by a plurality of heaters 66. The heaters 66 are annular, and for example, three heaters 66 are provided along the longitudinal direction of the cylinder 62. The heaters 66 melt the resin pellets in the cylinder 62. The molten resin is injected into a cavity 77 from a nozzle 67.

[0024] The position where the screw 63 is closest to the nozzle 67 is screw position 0. When the screw 63 retracts from screw position 0, the molten resin fills the cylinder 62. This process is called a metering process. The position where the screw 63 retracts from screw position 0 and stops is called a metering position.

[0025] As the screw 63 advances from the metering position, the molten resin is injected into the cavity 77. The forward speed of the screw 63 is controlled by the motor 64. When the screw 63 advances further to a position called the VP switching position, the process transitions to the pressure holding process. In other words, at the VP switching position, the control of the screw 63 transitions from speed control to pressure control. In the pressure holding process, the screw 63 pushes the resin material into the cavity 77 to maintain a constant pressure inside the cavity 77. In the pressure holding process, the motor 64 is torque controlled.

[0026] When the resin cools and hardens in the cavity 77, the movable mold separates from the fixed mold. The molded product 35 is then removed from the mold clamping unit 71. In preparation for the next shot, the movable mold 73 and fixed mold 72 are closed again. At the same time, the above-mentioned metering process is carried out in the injection unit 61.

[0027] 3.Management device 1, management device 10 performs production management of injection molding device 60. Management device 10 acquires various data of injection molding device 60 from sensor units 75 and 76. The contents of the data will be described later. Management device 10 also transmits control signals to injection molding device 60. The contents of this control will be described later.

[0028] The management device 10 is connected to cameras 31A-31D. The cameras 31A-31D capture images of the molded product 35. The cameras 31A-31D capture images of the molded product 35 from different angles. The cameras 31A-31D capture four images per molded product 35. These four captured images are sent to the management device 10.

[0029] The management device 10 is also connected to an operation device 30. The operation device 30 may be a keyboard or a mouse. Alternatively, the management device 10 may be provided with a touch panel display in which the operation device 30 and a display unit are integrated.

[0030] As will be described later, the control setting values ​​set by the management device 10 are changed by the operation device 30. By this operation, the occurrence of defective molded products 35 is suppressed.

[0031] The management device 10 is configured by, for example, a computer and includes a CPU 11 (Central Processing Unit), a RAM 12 (Random Access Memory), a ROM 13 (Read Only Memory), a storage 14, an input / output controller 15, and a display unit 16.

[0032] The CPU 11 is a central processing unit, also called a processor. The RAM 12 is a volatile storage device that temporarily stores data during operation. The ROM 13 is a storage device that can read data. The storage 14 is a storage device that can write and read data. The storage 14 is configured, for example, from an HDD (Hard Disk Drive) or an SSD (Solid State Drive).

[0033] 3 is configured in the management device 10 as a result of the CPU 11 (processor) executing a program stored in the ROM 13 or the storage 14. That is, the management device 10 includes, as processing units, a control unit 20, a data collection unit 21, a data set classification unit 22, a variable selection unit 23, a regression tree processing unit 24, a quality determination unit 25, a parameter discrimination unit 26, a parameter setting unit 27, and an update unit 28.

[0034] The quality assessment unit 25 assesses the quality of the molded product 35 based on the images captured by the cameras 31A-31D. The quality assessment unit 25 measures the depth of chips in the molded product 35 based on the captured images. A chip refers to a missing part that occurs in the molded product 35. Chips are mainly caused by an insufficient amount of resin. In other words, chips occur due to so-called short shots.

[0035] The quality determination unit 25 compares the image of the chip depth 0 with the captured image of the molded product 35, for example, using a known pattern matching method. Through this comparison, the quality determination unit 25 identifies the chipped area (missing area). Furthermore, the quality determination unit 25 measures the chip depth based on the chipped area image. As described above, four captured images are obtained from one molded product 35, and therefore, the chip depth is measured for four different parts of one molded product 35. In other words, the quality determination unit 25 also has an element as a measurement unit that measures the chip depth of the molded product 35.

[0036] The quality judgment unit 25 judges the quality of the molded product 35 based on the calculated chip depth. For example, if at least one of the four chip depth values ​​(measured values) for one molded product 35 is equal to or greater than a predetermined first threshold, the quality judgment unit 25 judges the molded product 35 with such chips as defective. Alternatively, if the average value of the four chip depth values ​​(measured values) for one molded product 35 is equal to or greater than the first threshold, the quality judgment unit 25 judges the molded product 35 with such chips as defective. For example, the first threshold is set to 50 μm. The calculated chip depth value is included in the datasets (good product dataset, defective product dataset, and minute dataset) described below.

[0037] The operating device 30 is capable of changing the control set values ​​for the control unit 20. The control set values ​​are set values ​​for controlling the operation, pressure, temperature, etc. of each device of the injection molding apparatus 60. For example, the control set values ​​include the metering position, VP switching position, and motor torque. The control set values ​​can be changed from the operating device 30.

[0038] The data collection unit 21 collects various data during production of the injection molding apparatus 60 via the sensor units 75 and 76. For example, the data collection unit 21 collects one-shot data and time-series data from the sensor units 75 and 76.

[0039] One-shot data is data that can be acquired once when one molded product 35 is produced. For example, 55 types of one-shot data, as exemplified in Table 1 below, are collected by the data collection unit 21. In Table 1, chip depth measurement values ​​1-4 are values ​​calculated by the quality determination unit 25 based on images captured by cameras 31A-31D.

[0040] [Table 1]

[0041] For each item in Table 1, the cycle time indicates the time from when the movable mold 73 opens to when the movable mold 73 opens again after one shot of molding. The injection time indicates the sum of the filling time and the pressure holding time. The cushion position indicates the most advanced position of the tip of the screw 63 after the resin is filled. The number of connected dryers indicates the number of injection molding devices 60 connected to a dryer that dries the resin pellets.

[0042] Furthermore, the nozzle, screw 1, screw 2, screw 3, fixed mold temperature, and movable mold temperature all indicate the temperature settings during molding. Screw 1-screw 3 indicate the set temperatures at three locations on the screw 63. For example, if the screw 63 is divided into a metering section, compression section, and supply section in that order from the tip, the set temperatures for each section are determined by screw 1-screw 3.

[0043] Furthermore, position 1st, position 2nd, position clamping, speed 1st, speed 2nd, and speed clamping indicate the set values ​​for the clamping position and speed of the movable mold 73. For example, in speed 1st, the speed set value at position 1 of the movable mold 73 is input. In speed clamping, the speed set value of the movable mold 73 when it abuts against the fixed mold 72 (position clamping) is input.

[0044] Further, the subsequent position type opening, position 2nd, position 1st, speed type opening, speed 2nd, and speed 1st indicate the set values ​​of the position and speed of the movable mold 73 when the mold is opened. For example, for the speed type opening, the speed set value of the movable mold 73 at the open position (position type opening) is input.

[0045] Additionally, for the low pressure clamping position and low pressure clamping pressure, the clamping position and pressure setting values ​​for low pressure are input. Speed ​​during dwelling, time 3rd, time 2nd, time 1st, pressure 3rd, pressure 2nd, and pressure 1st indicate the speed, time, and pressure setting values ​​during dwelling. For example, for pressure 1st, the pressure setting value for the 1st dwelling time is input.

[0046] Furthermore, the position VP switch, the second position, the first position, the speed VP switch, the second speed, the first speed, and the pressure indicate the set values ​​of the position, the speed, and the pressure when the resin is filled. For example, the speed of the screw 63 at the first position is set to the first speed.

[0047] The flush time indicates the time setting value for the flush process. The flush process is a process carried out between injection and pressure holding, and refers to a process in which restrictions are placed on the injection operation so that the resin compressed by injection can fill the space under its own pressure. Furthermore, the metering back pressure refers to the force that pushes the screw 63 from the rear to the front during metering.

[0048] Furthermore, the suck back position and the suck back speed are set values ​​during suck back. Suck back refers to the operation of slightly retracting the screw 63 after dwelling or measuring in injection molding.

[0049] The time-series data is measurement data that is measured multiple times at predetermined intervals when producing one molded product 35. For example, each of the following variables is measured 350 times at 20 ms intervals.

[0050] [Table 2]

[0051] Here, screws 1-3 refer to the measured values ​​of each section when, for example, screw 63 is divided into a metering section, a compression section, and a supply section in that order from the tip, as described above. For fixed molds 1 and 2 and movable molds 1 and 2, two temperature sensors are attached to each of fixed mold 72 and movable mold 73. These four temperature measurement points are obtained for fixed molds 1 and 2 and movable molds 1 and 2. Injection torque indicates the torque value of motor 64 during injection. Mold opening / closing torque indicates the drive torque of movable mold 73. Metering torque indicates the torque value (actual measured value) of motor 64 during the metering process.

[0052] When one molded product 35 is produced, each of the 55 types of one-shot data shown in Table 1 is measured once. When one molded product 35 is produced, each of the 12 types of one-shot data shown in Table 2 is measured 350 times at 20 ms intervals. In other words, when one molded product 35 is produced, 55 + 12 × 350 = 4255 pieces of data are collected by the data collection unit 21 from the sensor units 75 and 76.

[0053] 4. Control parameter extraction process The management device 10 narrows down the management data to be displayed on the display unit 16 (see FIG. 8) based on the data shown in Tables 1 and 2 obtained from the sensor units 75 and 76.

[0054] Referring to FIG. 4, the data set classification unit 22 classifies the data collected by the data collection unit 21 into a non-defective product data set and a defective product data set based on the pass / fail judgment of the molded product 35 (S12).

[0055] That is, the dataset classification unit 22 refers to the judgment result of the quality judgment unit 25 and extracts, as a good-product dataset, a dataset at the time of production of a molded product 35 that has been judged to be a good product. The dataset classification unit 22 also extracts, as a defective-product dataset, a dataset at the time of production of a molded product 35 that has been judged to be a defective product by the quality judgment unit 25. The good-product dataset and the defective-product dataset each include four chip depth values ​​(measured values) measured from images of the molded product 35 captured by the cameras 31A-31D.

[0056] 4-1. Extraction of control parameters from good product data sets (S14-S24) 3 and 4, the variable selection unit 23 selects explanatory variables from the non-defective product data set, with the chip depth value (measured value) as the dependent variable. The chip depth value, which is the dependent variable, may be the average value of four chip depth values ​​measured from the images captured by the cameras 31A-31D. Alternatively, the chip depth value, which is the dependent variable, may be the maximum value of the four chip depth values ​​measured from the images captured by the cameras 31A-31D.

[0057] The variable selection unit 23 extracts explanatory variables from a data set consisting of the above-mentioned 4255 types of data (more specifically, the chunk depth value, which is the objective variable, is excluded). When extracting explanatory variables, the variable selection unit 23 narrows down candidate variables. For example, the variable selection unit 23 narrows down the data (features) from the 4255 types of data until they are fewer than 20 types (S14).

[0058] To narrow down the data, the variable selection unit 23 uses at least one of the following methods: filter method, wrapper method, embedding method, and random forest. The target variable is the average or maximum value of the chip depth values ​​(measured values) of the four points. The above narrowing down methods are well known, so a description thereof will be omitted here.

[0059] When the data (features) are further narrowed down to less than 20 types, the regression tree processing unit 24 creates a regression tree using the objective variables in the non-defective product data set. Here, the objective variable is the average or maximum value of the chip depth values ​​(measured values) of the above four points.

[0060] For example, the regression tree processing unit 24 creates a decision tree for less than 20 types of data narrowed down within the non-defective product data set. In this embodiment, the objective variable is a block depth value, which is a numerical value, so the decision tree is a regression tree (S16). The depth of the regression tree is arbitrary, but is set to, for example, 3 to 6. This type of regression tree processing is also called non-defective product regression tree processing.

[0061] Figure 5 shows an example of a regression tree with a depth of 3. Each box 50-56 displays the explanatory variables (candidate control parameters), mean squared error, the number of data points for the objective variable, and the average value of the chip depth value. For example, taking box 50 at the top of the regression tree as an example, "screw position 65" is selected as the explanatory variable. 3.85 mm is also selected as the threshold. The mean squared error is 2.026, and the number of data points for the objective variable is 79,774. The average value of the 79,774 objective variables is 10.752 μm.

[0062] "Screw position 65 <= 3.85" refers to the feature of whether the screw position is 3.85 mm or greater on the 65th measurement (i.e., 20 ms x 65 = 1300 ms after the start of injection) out of 350 measurements taken at 20 ms intervals.

[0063] Data for which the conditions indicated by the explanatory variables and thresholds in box 50 are true is allocated to the left side of the regression tree. Therefore, box 51 branching to the left shows the characteristics of data for which the screw position was 3.85 mm or greater 20 ms x 65 = 1,300 ms after the screw started to retract. According to this, the number of dependent variables is 36,741, and the average value of these 36,741 dependent variables is 11.364 μm.

[0064] Data for which the condition indicated by the explanatory variables and threshold in box 50 is false is assigned to the right side of the regression tree. Therefore, box 52 branched to the right shows the characteristics of data in which the screw position was less than 3.85 mm 20 ms x 65 = 1300 ms after the start of screw retraction. According to this, the number of dependent variables is 43,033, and the average value of these 43,033 dependent variables is 10.229 μm.

[0065] Qualitatively, the lower the chip depth value, the less likely it is to result in a defective product. In other words, in this regression tree, a screw position of less than 3.85 mm 20 ms x 65 = 1300 ms after the screw starts to retract is less likely to result in a defective product (it can be a good product) than a screw position of 3.85 mm or more 20 ms x 65 = 1300 ms after the screw starts to retract.

[0066] When such a significant difference in the objective variable is observed, the explanatory variable "screw position 65" becomes more important. In regression tree algorithms, importance indicates how much Gini impurity can be reduced by splitting the objective variable by a certain feature (explanatory variable). Gini impurity is an index that measures how poorly the target (objective variable) is classified at a certain node. For example, in the example in Figure 5, box 51 can be interpreted as containing data with relatively large block depth values, while box 52 contains data with relatively small block depth values. Qualitatively, the greater the difference in the values ​​of the two split boxes, the higher the importance value.

[0067] For example, the regression tree processing unit 24 creates multiple regression trees so that all of the explanatory variables narrowed down in step S14 are reflected in the regression trees (S16). Furthermore, the importance of each box is calculated. Furthermore, the parameter discrimination unit 26 extracts explanatory variables (features) with high importance (S18). For example, the parameter discrimination unit 26 extracts the top three explanatory variables (features) with high importance.

[0068] In this way, when extracting the control parameters in this embodiment, variable selection such as random forest is performed as the primary screening, and a regression tree algorithm is used as the secondary screening.

[0069] In contrast to primary screening, which selects variables based solely on importance, regression tree algorithms, as illustrated in Figure 5, indicate the number of data samples for the dependent variable. For example, even if a certain explanatory variable is assigned a high importance, it may be inappropriate as an explanatory variable if the data is divided in an extreme manner. For example, an explanatory variable that divides 10,000 pieces of data into 9,999 boxes and one box is considered inappropriate even if it has a high importance. In this way, regression tree algorithms are highly appropriate as an explanatory variable extraction method, as they allow operators to verify the appropriateness of the data group division in addition to the importance.

[0070] The parameter determination unit 26 determines whether the extracted explanatory variables (feature amounts) can be adjusted by changing the control setting values ​​for the injection molding device 60 (S20). In other words, the parameter determination unit 26 determines whether the extracted explanatory variables are correlated with an increase or decrease in the injection molding device 60.

[0071] For example, when the control setting value of the injection molding device 60 is only the VP switching position, the parameter determining unit 26 determines whether the explanatory variable (feature amount) extracted in step S18 changes depending on whether the VP switching position increases or decreases.

[0072] Here, the control set value refers to a set value that can be changed by an operator's operation to change the behavior of the injection molding apparatus 60. For example, a set value that can be changed by input from the operating device 30 is a control set value.

[0073] The VP switching position refers to the position of the screw 63 at the switching point between the injection process in which the screw 63 is speed-controlled and the pressure-holding process in which the screw 63 is pressure-controlled (torque-controlled).

[0074] For example, in step S20, the parameter discriminator 26 determines the correlation between the VP switching position and each explanatory variable by referring to the data set collected by the data collector 21. Alternatively, the parameter discriminator 26 may store in advance the correlation between the 4255 types of data and the VP switching position.

[0075] If there is an explanatory variable that cannot be adjusted (has no correlation) by the control setting value, the parameter discriminator 26 excludes the explanatory variable from the management parameters (S26). An example of such an explanatory variable (feature amount) is the number of connected dryers in Table 1.

[0076] In step S20, once only multiple types of explanatory variables (features) that are correlated with increases and decreases in the control setting value are extracted, the parameter discrimination unit 26 then determines whether or not there is an explanatory variable that exhibits a trend opposite to the fluctuation trend of the explanatory variable with the greatest importance (S22).

[0077] For example, suppose that the explanatory variable with the highest importance (e.g., screw position 65) tends to increase when the VP switching position, which is a control setting value, is increased, that is, when the VP switching position is moved away from the nozzle 67. At this time, if an explanatory variable with a relatively low importance tends to decrease, the parameter discriminator 26 excludes such an explanatory variable with a relatively low importance from the management parameters.

[0078] Through steps S18, S20, and S22, the parameter discriminator 26 extracts explanatory variables that affect the chip depth value, are correlated with increases or decreases in the control setting value, and have the same correlation as the explanatory variable with the highest importance. The parameter setter 27 sets these explanatory variables as control parameters (S24). The set control parameters are displayed on the display unit 16, as shown in FIG. 8. In this example, screw position 144 and injection torque 54 are displayed as control parameters extracted from the non-defective product data set. Note that the solid lines in each graph indicate actual measured values. The dashed lines in each graph indicate threshold values. For example, referring to FIG. 5, the threshold value corresponds to 3.85 mm for screw position 65 in box 50 and -85.35 N·m for injection torque 41 in box 55. In other words, the threshold value is the boundary value that determines whether the chip depth value becomes shallower or deeper.

[0079] If only one parameter with high importance is extracted from the non-defective product data set in step S18, step S22 is omitted. In this case, if the extracted parameter with the highest importance is excluded in step S20 and then step S26, no management parameter is set from the non-defective product data set. In this case, the process waits for the update unit 28 (see FIG. 3) to update the management parameter.

[0080] 4-2. Extraction of control parameters from defective product dataset Figure 6 shows an example of a flow branched from Figure 5. In the flow of Figure 6, control parameters are set from a defective product dataset. In other words, in a production environment where defective products occur, explanatory variables that affect chip depth values ​​are narrowed down and extracted as control parameters.

[0081] Steps S34-S46 are the same as steps S14-26 in Fig. 4 in terms of processing content, although the base (good / defective product data set) is different. That is, the variable selection unit 23 narrows down the data (features) from a data set consisting of 4,255 types of data (more specifically, the chip depth value, which is the objective variable) in the defective product data set, as exemplified in Tables 1 and 2, until the number of types is less than 20 (S34). As described above, the narrowing down is performed using at least one of the filter method, wrapper method, embedding method, and random forest method.

[0082] When the data (features) are further narrowed down to less than 20 types, the regression tree processing unit 24 creates a regression tree using the average or maximum value of the above four chip depth values ​​(measured values) as the objective variable (S36). This type of regression tree processing is also called defective product regression tree processing. For example, in the defective product regression tree processing, the regression tree processing unit 24 creates multiple regression trees so that all of the explanatory variables narrowed down in step S34 are reflected in the regression tree.

[0083] Next, the parameter discriminator 26 extracts explanatory variables (feature amounts) with high importance (S38). For example, the parameter discriminator 26 extracts the top three types of explanatory variables (feature amounts) with high importance.

[0084] Thereafter, the parameter discriminator 26 determines whether or not the extracted explanatory variables (feature amounts) are correlated with an increase or decrease in the injection molding device 60 (S40). If there is an explanatory variable that cannot be adjusted (has no correlation) by the control setting value, the parameter discriminator 26 excludes the explanatory variable from the management parameters (S46).

[0085] In step S40, when only explanatory variables (features) correlated with increases and decreases in the control setting value are extracted, the parameter discriminator 26 then determines whether or not there is an explanatory variable that exhibits a fluctuation trend opposite to that of the explanatory variable with the greatest importance (S42).If there is an explanatory variable that exhibits a fluctuation trend opposite to that of the explanatory variable with the greatest importance, the parameter discriminator 26 excludes such an explanatory variable from the management parameters (S46).

[0086] After steps S38, S40, and S42, the parameter determination unit 26 extracts explanatory variables that affect the chip depth value, are correlated with increases or decreases in the control setting value, and have the same correlation as the explanatory variable with the highest importance. The parameter setting unit 27 sets such explanatory variables as management parameters (S44).

[0087] The set control parameters are displayed on the display unit 16, as shown in FIG. 8. In this example, only the measurement torque 169 is displayed as the control parameter extracted from the defective product data set. The solid lines in each graph indicate actual measured values. The dashed lines in each graph indicate threshold values. Taking the regression tree in FIG. 5 as an example, the threshold value corresponds to 3.85 mm at the screw position 65 of box 50. In other words, a value is set as the threshold value such that the chip depth increases when the control parameter exceeds the threshold value.

[0088] 4-3. Extraction of control parameters from small data sets Since chipping of the molded product 35 occurs due to so-called short shots, the chip depth can be reduced by increasing the amount of resin injected. However, if the amount of resin injected is excessive, so-called overpacking occurs, which may cause burrs.

[0089] Therefore, control parameters for monitoring the presence or absence of overpacks may be displayed on the display unit 16. For example, a data set obtained during production of a molded product 35 with a relatively small chip depth value is used as a data set for extracting control parameters for overpack monitoring. This data set is called a minute data set.

[0090] When generating the minute dataset, a second threshold smaller than the first threshold used for determining acceptability is used. The second threshold is set to, for example, 10 μm. Referring to FIG. 7, the dataset classification unit 22 extracts various data during production of molded product 35, for which the average value of the chip depth values ​​at four points taken by cameras 31A-31D is less than the second threshold (10 μm), i.e., 4,255 types of data shown in Tables 1 and 2, and creates a minute dataset (S52).

[0091] Although the process of extracting control parameters from a small dataset differs (non-defective product / small dataset), the processing content is the same as the flow in Figure 5. That is, the variable selection unit 23 narrows down the data (features) from a dataset consisting of 4,255 types of data in the small dataset (more specifically, the chip depth value, which is the objective variable, is excluded) as exemplified in Tables 1 and 2, until the number of types is less than 20 (S54). As described above, at least one of the filter method, wrapper method, embedding method, and random forest method is used for narrowing down the data.

[0092] When the data (features) are further narrowed down to less than 20 types, the regression tree processor 24 creates a regression tree using the average or maximum value of the block depth values ​​of the four points as the objective variable (S56). This type of regression tree processing is also called micro-regression tree processing. For example, the regression tree processor 24 creates multiple regression trees so that all of the explanatory variables narrowed down in step S34 are reflected in the regression tree.

[0093] Next, the parameter discriminator 26 extracts explanatory variables (feature amounts) with high importance (S58). For example, the parameter discriminator 26 extracts the top three types of explanatory variables (feature amounts) with high importance.

[0094] Thereafter, the parameter discriminator 26 determines whether or not the extracted explanatory variables (feature amounts) are correlated with an increase or decrease in the injection molding device 60 (S60). If there is an explanatory variable that cannot be adjusted (has no correlation) by the control setting value, the parameter discriminator 26 excludes the explanatory variable from the management parameters (S66).

[0095] Once only explanatory variables (features) correlated with increases and decreases in the control setting value have been extracted, the parameter discriminator 26 then determines whether or not there are explanatory variables that exhibit a trend opposite to the trend of fluctuations in the explanatory variable with the greatest importance (S62).If there are explanatory variables that exhibit a trend opposite to the trend of fluctuations in the explanatory variable with the greatest importance, the parameter discriminator 26 excludes such explanatory variables from the management parameters (S66).

[0096] Through the above steps, the parameter determination unit 26 extracts explanatory variables that affect the chip depth value, are correlated with increases or decreases in the control setting value, and have the same correlation as the explanatory variable with the highest importance. The parameter setting unit 27 sets such explanatory variables as management parameters (S64).

[0097] Referring to FIG. 8, in addition to the non-defective product data set and the defective product data set, a control parameter (light torque 149) extracted from the minute data set is displayed on the display unit 16. The solid line in the graph of the control parameter "light torque 149" indicates the actual measured value. The dashed line in the graph indicates the threshold value. Specifically, if the actual measured value exceeds the threshold value, the risk of overpacking increases. In other words, the control setting value is adjusted so that the actual measured value is kept below the threshold value.

[0098] 5. Production management using management data 5-1. When good products are produced The operator of the injection molding machine 60 checks that the conditions for producing non-defective products are being maintained based on the chip measurement values ​​displayed on the display unit 16. When the chip measurement values ​​approach the threshold values, the operator adjusts the control setting values ​​so that the control parameters extracted from the non-defective product data set are moved away from the threshold values.

[0099] In this case, due to the correlation described above, moving the control parameters away from the thresholds reduces the chip depth values. Excessive reduction in the chip depth values ​​can lead to overpacking. Therefore, the operator moves the control parameters extracted from the non-defective data set away from the thresholds, as long as the control parameters extracted from the small data set do not exceed the thresholds.

[0100] 5-2.If a defective product occurs When a defective molded product 35 occurs due to a sudden disturbance, the control parameter extracted from the defective product data set exceeds the threshold. The operator adjusts the control setting value so that the control parameter is below the threshold. By making such adjustments, the chip depth is reduced, and the production of non-defective products can be quickly restored.

[0101] 6. Update control parameters 3, the update unit 28 updates (re-executes) the process of extracting control parameters from the non-defective data set, the defective data set, and the minute data set at predetermined intervals. For example, when the number of shots exceeds a predetermined update threshold (e.g., 10,000 shots), the control parameters are updated. For example, the update unit 28 outputs update commands to the data set classification unit 22, the variable selection unit 23, the regression tree processing unit 24, the parameter discrimination unit 26, and the parameter setting unit 27.

[0102] The explanatory variables that affect the chip depth value may vary depending on the operating environment and number of years of operation of the injection molding machine 60. Therefore, the control parameters are extracted and set at predetermined intervals. This allows the control parameters that reflect the most recent operating status to be extracted. [Explanation of symbols]

[0103] 10 management device, 20 control unit, 21 data collection unit, 22 data set classification unit, 23 variable selection unit, 24 regression tree processing unit, 25 quality judgment unit (measurement unit and judgment unit), 26 parameter discrimination unit, 27 parameter setting unit, 28 update unit, 30 operation device, 31A-31D cameras, 35 molded product (product), 60 injection molding device (production device), 75, 76 sensor unit.

Claims

1. A production device that produces the product; a collection unit that collects various data during production of the production device; a measuring unit that measures the product produced by the production device; a determining unit that determines whether the product is good or bad based on the measurement value obtained by the measuring unit; a classification unit that extracts data and the measurement values ​​at the time of production of the products that have been determined to be non-defective as a non-defective data set, and extracts data and the measurement values ​​at the time of production of the products that have been determined to be defective as a defective data set; a regression tree processing unit that executes a good product regression tree processing that creates a regression tree for the good product data set using the measurement value as a response variable and extracts explanatory variables with relatively high importance, and a defective product regression tree processing that creates a regression tree for the defective product data set using the measurement value as a response variable and extracts explanatory variables with relatively high importance; a parameter determination unit that extracts, from the explanatory variables extracted by the non-defective product regression tree processing and the defective product regression tree processing, explanatory variables that are correlated with an increase or decrease in a control setting value of the production device as control parameters; a display unit that displays the management parameters; A production management system equipped with:

2. The production management system according to claim 1, When there are a plurality of types of explanatory variables extracted by the good product regression tree processing and the defective product regression tree processing, the parameter determination unit excludes, from the management parameters, explanatory variables of relatively low importance that have an inverse correlation with the correlation between the explanatory variables of relatively high importance and the control setting values ​​of the production devices. Production management system.

3. 3. The production management system according to claim 1 or 2, The production device is an injection molding device, The measuring unit measures the depth of a chip in the molded product, The determination unit determines the molded product to be defective when the chip depth is equal to or greater than a first threshold value, the classification unit extracts, as a minute data set, various data of the injection molding device and the chip depth when the chip depth is less than a second threshold value that is smaller than the first threshold value; the regression tree processing unit executes a micro regression tree process to create a regression tree for the micro dataset using the chunk depth as a target variable and extract explanatory variables with relatively high importance; the parameter determination unit extracts, from the explanatory variables extracted by the small regression tree processing, explanatory variables that are correlated with an increase or decrease in a control setting value of the injection molding device as the management parameters; Production management system.

4. The production management system according to claim 3, an update unit that outputs an update command to the classification unit, the regression tree processing unit, and the parameter determination unit when the number of shots by the injection molding device exceeds an update threshold; Production management system.

Images