Quality prediction model creation system, quality prediction system, quality prediction model creation method, and quality prediction method

Abstract

To create a quality prediction model for predicting the quality of a long product.SOLUTION: A quality prediction model creation system includes: a first logging data acquisition unit 301 that acquires known logging data detected and stored at every predetermined interval by a plurality of detectors when a long product is manufactured; an average value calculation unit 302 that calculates an average value of the logging data from the pieces of logging data acquired by the first logging data acquisition unit 301; and a model creation unit 304 that creates a quality prediction models associated with the quality characteristics of the long product based on the average value of the logging data.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to a quality prediction model creation system, a quality prediction system, a quality prediction model creation method, and a quality prediction method. [Background technology]

[0002] Japanese Patent Application Laid-Open No. 2022-87429 (Patent Document 1) describes a technology that utilizes artificial intelligence to estimate the physical property values ​​of composite materials with high accuracy. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2022-87429 Summary of the Invention [Problem to be solved by the invention]

[0004] For long products, while it is possible to evaluate the quality characteristics of the end portions using destructive testing, it is not possible to evaluate the quality characteristics along the entire length. For example, it is not possible to determine whether the intermediate portion of a long product actually meets the required quality characteristics. Furthermore, unlike physical properties such as outer diameter, eccentricity, and capacitance, some quality characteristics of long products are difficult to record using a detector or other device to measure the actual values ​​during manufacturing. [Means for solving the problem]

[0005] In one embodiment, the quality prediction model creation system includes a first logging data acquisition unit that acquires known first logging data that is detected and stored by multiple detectors at predetermined intervals during the manufacture of long products, an average value calculation unit that calculates average values ​​of the first logging data from the first logging data acquired by the first logging data acquisition unit, and a model creation unit that creates a quality prediction model associated with the quality characteristics of the long product based on the average values ​​of the first logging data.

[0006] In one embodiment, the quality prediction system includes a quality prediction model memory unit that stores a quality prediction model associated with the quality characteristics of a long item created based on the average value of known first logging data of the long item, a second logging data acquisition unit that acquires second logging data during the manufacture of the long item to be predicted, and a prediction unit that substitutes the second logging data acquired by the second logging data acquisition unit into the quality prediction model to predict the quality characteristics of the long item to be predicted.

[0007] In one embodiment, a quality prediction model creation method acquires known first logging data, calculates average values ​​of the acquired first logging data from the acquired first logging data, and creates a quality prediction model associated with the quality characteristics of the long product based on the average values ​​of the first logging data.

[0008] In addition, in one embodiment, a quality prediction method acquires known first logging data that is detected and stored by multiple detectors at predetermined intervals during the manufacture of long products, calculates average values ​​of the acquired first logging data from the first logging data, creates a quality prediction model associated with the quality characteristics of the long product based on the average values ​​of the first logging data, acquires second logging data during the manufacture of the long product to be predicted, and substitutes the second logging data into the quality prediction model to predict the quality characteristics of the long product to be predicted. [Effects of the Invention]

[0009] According to one embodiment, a quality prediction model for predicting the quality of long products can be created. Moreover, according to one embodiment, the quality prediction model can be used to predict the quality characteristics of a long product over its entire length. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a schematic diagram illustrating an example of a configuration of a manufacturing system including a quality prediction system. [Figure 2] FIG. 1 is a schematic diagram illustrating an example of a configuration of a manufacturing system including a quality prediction system. [Figure 3] FIG. 1 illustrates an example of a hardware configuration of a quality prediction system. [Figure 4] FIG. 1 is a diagram illustrating an example of functional blocks of a quality prediction system. [Figure 5] 10 is a flowchart illustrating an example of a quality prediction model creation process. [Figure 6] FIG. 10 is a diagram showing an example of first logging data. [Figure 7] FIG. 10 is a diagram illustrating an example of calculation result data showing a calculation result of an average value of logging data. [Figure 8] 1 is a flowchart illustrating an example of a process for implementing a full-length quality prediction technique. [Figure 9] FIG. 1 is a diagram for explaining a full-length quality prediction technique. [Figure 10] FIG. 10 is a diagram showing an example of second logging data. [Figure 11] FIG. 10 is a diagram for explaining an outline of a simulation. [Figure 12] FIG. 1 is a diagram showing the relationship between the measured value (measured strength) and the predicted value (predicted strength) of tensile strength. [Figure 13] FIG. 1 is a diagram showing the relationship between the measured value (measured elongation) and the predicted value (predicted elongation) of tensile elongation. [Figure 14] FIG. 10 is a diagram showing second logging data of the screw rotation speed and the cylinder temperature. [Figure 15]10 is a diagram showing second logging data of a current value of a cylinder and a resin pressure value detected by a resin pressure sensor. FIG. [Figure 16] FIG. 10 is a diagram showing second logging data of the take-up speed of the take-up device and the ambient temperature. [Figure 17] FIG. 10 is a diagram showing second simulation data of maximum resin pressure and driving force. [Figure 18] FIG. 10 is a diagram showing second simulation data of residence time and average temperature. [Figure 19] FIG. 10 is a diagram showing second simulation data of strain and torque. [Figure 20] FIG. 10 is a diagram showing the predicted results of the tensile strength of a heat-resistant insulated wire. [Figure 21] FIG. 10 is a diagram showing the predicted results of tensile elongation of a heat-resistant insulated wire. DETAILED DESCRIPTION OF THE INVENTION

[0011] In all the drawings for explaining the embodiments, the same components are generally designated by the same reference numerals, and repeated explanations thereof will be omitted. In addition, hatching may be used even in plan views to make the drawings easier to understand.

[0012] The technical idea of ​​this embodiment is to create a quality prediction model and use the model to predict the quality characteristics of a long product over its entire length. For example, if the long product is an electric wire coated with a heat-resistant insulating material, CAE and AI / machine learning are used to predict the quality characteristics such as tensile strength and tensile elongation at each position on the electric wire.

[0013] The technology of this embodiment uses data that can be measured and logged during the production of long products (e.g., extruder screw rotation speed, cylinder temperature, resin pressure value, outer diameter, etc.) and data that can be calculated through numerical simulation (average resin temperature in the cylinder, maximum resin pressure value, torque value, magnitude of strain, etc.) to create a quality prediction model associated with quality characteristics through machine learning. Furthermore, the technology of this embodiment makes it possible to predict the quality characteristics of long products over their entire length using the created quality prediction model and real-time data during production.

[0014] <Description of Related Art> Let's take the case of a long product, such as a heat-resistant insulated wire. Standards have been established for the quality characteristics of heat-resistant insulated wire, such as insulation resistance, withstand voltage, tensile strength of the insulator and sheath, heat resistance, and oil resistance, and the test methods are also specified in detail in JIS C3005, Test Methods for Rubber and Plastic Insulated Wires.

[0015] As stated in 4.16 of the same, the mechanical properties (tensile strength, tensile elongation) of the insulator and sheath of long cables such as rubber wires are to be measured by taking three or more dumbbell-shaped or tubular test pieces from the finished product and measuring them at a specified tensile speed (25 to 500 mm / min). For this reason, three or more pieces are cut out from the end of a finished long cable, and the tensile strength and tensile elongation values ​​are calculated using a tensile tester, which are then used as the quality characteristics of that long cable.

[0016] With related technology, quality characteristics can only be evaluated by destructive testing of the ends of long products, making it impossible to determine whether the middle part of a long product truly meets the standards. Also, unlike outer diameter, eccentricity, and capacitance, there are some quality characteristics for which it is difficult to constantly log the actual measured values ​​of quality characteristics of long products during manufacturing using a detector or other device.

[0017] <Manufacturing system configuration> 1 and 2 are schematic diagrams showing an example of the configuration of a manufacturing system 1 including a quality prediction system 100. FIG. 1 is a schematic diagram showing an example of the configuration of an extruder 10 of the manufacturing system 1 as viewed from the Y direction. FIG. 2 is a schematic diagram showing an example of the configuration of the extruder 10 of the manufacturing system 1 as viewed from the X direction in the head 31 of FIG. 1. In this embodiment, a case will be described in which the quality prediction system 100 includes a quality prediction model creation system, but the quality prediction system 100 and the quality prediction model creation system may be separate systems.

[0018] 1 and 2, the X direction, Y direction, and Z direction are defined. The X direction, Y direction, and Z direction are perpendicular to each other, but may intersect at an angle other than a perpendicular angle.

[0019] The manufacturing system 1 includes an extruder 10 and a quality prediction system 100. The extruder 10 has a single screw (not shown). In this embodiment, the extruder 10 will be described in the case of predicting the quality characteristics of a heat-resistant insulated electric wire EW as an example of a long product. Note that the long product is not limited to an electric wire such as the heat-resistant insulated electric wire EW. For example, the long product may be a pipe, a sheet, or a film. The quality prediction system 100 is, for example, a computer system. Details of the quality prediction system 100 will be described later.

[0020] As shown in Fig. 1, the extruder 10 produces a heat-resistant insulated wire EW by coating a core wire CW with a resin, for example, a heat-resistant insulating material. As shown in Fig. 1, the extruder 10 has a drawing device 20, a head 31, a die 32, a water tank 40, and a take-up device 50. The drawing device 20, the head 31, the die 32, the water tank 40, and the take-up device 50 form an extrusion line. As shown in Fig. 2, the extruder 10 has a cylinder 33 connected to the head 31, and a material input section 34.

[0021] The drawing device 20 accommodates the core wire CW wound thereon. The drawing device 20 rotates in the direction of the arrow AR1 shown in the figure, thereby drawing out the core wire CW from the drawing device 20. The core wire CW drawn in the direction of the arrow AR2 shown in the figure passes through the head 31 and the die 32. The heat-resistant insulated electric wire EW that has passed through the head 31 and the die 32 passes through the water tank 40 and is taken up by the take-up device 50. The heat-resistant insulated electric wire EW is wound and accommodated in the take-up device 50. The arrow AR3 shown in the figure indicates the direction in which the heat-resistant insulated electric wire EW is wound.

[0022] The head 31 is connected to a cylinder 33. A heat-resistant insulating material is supplied to the head 31 from the cylinder 33. The heat-resistant insulating material supplied to the head 31 is attached to the core wire CW drawn out from the drawing device 20. The die 32 has a hole (not shown) of a predetermined size. The core wire CW with the heat-resistant insulating material attached passes through the hole of the die 32. As a result, a heat-resistant insulating material of a thickness corresponding to the size of the hole is formed on the core wire CW, and a heat-resistant insulated electric wire EW is produced.

[0023] The heat-resistant insulated wire EW that has passed through the die 32 passes through the water tank 40. In the water tank 40, the heat-resistant insulated wire EW is cooled to a predetermined temperature. The heat-resistant insulated wire EW cooled in this manner is taken up by the take-up device 50. In the take-up device 50, the heat-resistant insulated wire EW is wound at a take-up speed (hereinafter also referred to as "linear speed") and stored in the take-up device 50. Information indicating the take-up speed and a take-up current value that indicates the value of a current supplied to the take-up device 50 is transmitted to the quality prediction system 100 at predetermined intervals, for example, every one second.

[0024] As shown in FIG. 2, a head 31 is connected to one end of a cylinder 33, and a material feeder 34 is connected to the top of the other end. Heat-resistant insulating material is fed into the material feeder 34. As indicated by the dashed arrow AR4 in the figure, the heat-resistant insulating material is fed into the cylinder 33 from the material feeder 34. The cylinder 33 has a single-axis screw. The cylinder 33 is configured to rotate the screw at a predetermined rotation speed depending on the amount of current supplied. As the screw rotates, the heat-resistant insulating material fed from the material feeder 34 is gradually extruded toward the head 31, as indicated by the dashed arrow AR5 in the figure. The extruded heat-resistant insulating material is attached to the core wire CW within the head 31 and is extruded from the die 32 to the outside as part of the heat-resistant insulated wire EW. Information indicating the screw rotation speed, acceleration, current value, etc., is transmitted to the quality prediction system 100 at predetermined intervals, for example, every one second.

[0025] Here, the screw has, for example, three roles. The first role is to push the heat-resistant insulating material supplied from the material input section 34 toward the head 31 so that it reaches the head 31. At this time, residual heat is applied to the heat-resistant insulating material. The second role is to change the heat-resistant insulating material from a solid state to a molten state. The third role is to stably extrude a constant amount of heat-resistant insulating material from the die 32.

[0026] A core wire temperature sensor S11, which is one of the detectors, is provided between the drawing device 20 and the head 31. The core wire temperature sensor S11 detects the temperature of the core wire CW between the drawing device 20 and the head 31. The temperature detected by the core wire temperature sensor S11 is transmitted to the quality prediction system 100 at predetermined intervals, for example, at intervals of one second.

[0027] As shown in FIG. 1, the head 31 is provided with a head temperature sensor S12 and a die temperature sensor S13. The head temperature sensor S12 and the die temperature sensor S13 are each detectors. The head temperature sensor S12 is a sensor that detects the head temperature of the head 31. The die temperature sensor S13 is a sensor that detects the die temperature of the die 32. In addition, a neck temperature sensor S14, which is one of the detectors, is provided near the joint between the cylinder 33 and the head 31 (see FIG. 2). The neck temperature sensor S14 is a sensor that detects the neck temperature near the joint between the cylinder 33 and the head 31. The head temperature, die temperature, and neck temperature detected by the head temperature sensor S12, die temperature sensor S13, and neck temperature sensor S14, respectively, are each transmitted to the quality prediction system 100 at predetermined intervals, for example, at one-second intervals.

[0028] A water tank temperature sensor S15, which is one of the detectors, is provided in water tank 40. Water tank temperature sensor S15 is a sensor that detects the water temperature of water tank 40. The temperature detected by water tank temperature sensor S15 is transmitted to quality prediction system 100 at predetermined intervals, for example, at one-second intervals.

[0029] An outer diameter measuring sensor S16, which is one of the detectors, is provided between the water tank 40 and the take-up device 50. The outer diameter measuring sensor S16 is a sensor that detects the size of the outer diameter of the heat-resistant insulated electric wire EW after it has been cooled in the water tank 40. The size of the outer diameter detected by the outer diameter measuring sensor S16 is transmitted to the quality prediction system 100 at predetermined intervals, for example, at intervals of one second.

[0030] 2, a resin temperature sensor S17 and a resin pressure sensor S18 are provided near the connection portion of the cylinder 33 with the head 31. The resin temperature sensor S17 and the resin pressure sensor S18 are each detectors. The resin temperature sensor S17 is a sensor that detects the resin temperature of the heat-resistant insulating material supplied to the head 31. The resin pressure sensor S18 is a sensor that detects the resin pressure value (also referred to as the tip resin pressure value) of the heat-resistant insulating material supplied to the head 31. The resin temperature and resin pressure values ​​detected by the resin temperature sensor S17 and the resin pressure sensor S18 are each transmitted to the quality prediction system 100 at predetermined intervals, for example, at one-second intervals.

[0031] Five temperature sensors S21 to S25 are provided on the cylinder 33 along the longitudinal direction from the material input section 34 side. Each of the five temperature sensors S21 to S25 is a detector. This allows the temperature of the cylinder 33 to be detected at multiple positions along the longitudinal direction of the cylinder 33. The number of temperature sensors is not limited to five and may be, for example, three, four, or six or more. The temperatures detected by the five temperature sensors S21 to S25 are transmitted to the quality prediction system 100 at predetermined intervals, for example, at one-second intervals. As described above, the cylinder 33 has three roles. By providing multiple temperature sensors S21 to S25, the quality prediction system 100 can detect changes in the temperature of the cylinder 33 in each role.

[0032] An air temperature sensor S31 and a humidity sensor S32 are disposed around the extruder 10, for example, around the cylinder 33. The air temperature sensor S31 and the humidity sensor S32 are each detectors. The temperature detected by the air temperature sensor S31 and the humidity detected by the humidity sensor S32 are each transmitted to the quality prediction system 100 at predetermined intervals, for example, at one-second intervals.

[0033] <Hardware configuration> Next, a hardware configuration of the quality prediction system 100 in this embodiment will be described. Fig. 3 is a diagram showing an example of the hardware configuration of the quality prediction system 100 in this embodiment. Note that the configuration shown in Fig. 3 merely shows an example of the hardware configuration of the quality prediction system 100, and the hardware configuration of the quality prediction system 100 is not limited to the configuration shown in Fig. 1 and may be other configurations.

[0034] 1, a quality prediction system 100 includes a CPU (Central Processing Unit) 101 that executes a program. This CPU 101 is electrically connected to, for example, a ROM (Read Only Memory) 102, a RAM (Random Access Memory) 103, and a hard disk drive 112 via a bus 113, and is configured to control these hardware devices.

[0035] The CPU 101 is also connected to input devices and output devices via a bus 113. Examples of input devices include a keyboard 105, a mouse 106, a communication port 107, and a scanner 111. Examples of output devices include a display 104, a communication port 107, and a printer 110. The CPU 101 may also be connected to, for example, a removable disk device 108 and a CD / DVD-ROM device 109. The communication port 107 is connected to, for example, a cylinder 33, a drawing device 20, a take-up device 50, a core wire temperature sensor S11, a head temperature sensor S12, a die temperature sensor S13, a neck temperature sensor S14, a water tank temperature sensor S15, an outer diameter measurement sensor S16, a resin temperature sensor S17, a resin pressure sensor S18, temperature sensors S21 to S25, an air temperature sensor S31, and a humidity sensor S32.

[0036] The quality prediction system 100 may be connected to, for example, a network. For example, when the quality prediction system 100 is connected to other external devices via a network, a communication port 107 constituting a part of the quality prediction system 100 is connected to a LAN (Local Area Network), a WAN (Wide Area Network), or the Internet.

[0037] The RAM 103 is an example of a volatile memory, and the storage media of the ROM 102, the removable disk device 108, the CD / DVD-ROM device 109, and the hard disk device 112 are examples of a non-volatile memory. These volatile memories and non-volatile memories constitute the storage device of the quality prediction system 100.

[0038] The hard disk drive 112 stores, for example, an operating system (OS) 201, a program group 202, and a file group 203. The programs included in the program group 202 are executed by the CPU 101 using the operating system 201. The RAM 103 also temporarily stores at least some of the programs of the operating system 201 and application programs executed by the CPU 101, as well as various data required for processing by the CPU 101.

[0039] A BIOS (Basic Input Output System) program is stored in the ROM 102, and a boot program is stored in the hard disk drive 112. When the quality prediction system 100 is started up, the BIOS program stored in the ROM 102 and the boot program stored in the hard disk drive 112 are executed, and the operating system 201 is started up by the BIOS program and the boot program.

[0040] The program group 202 stores programs that realize the functions of the quality prediction system 100, and these programs are read and executed by the CPU 101. The file group 203 stores information, data, signal values, variable values, and parameters indicating the results of processing by the CPU 101 as factors of files. The program group 202 includes, for example, a quality prediction model creation program 202A and a quality prediction program 202B, which will be described later.

[0041] The file group 203 includes logging data 203A, simulation data 203B, and a quality prediction model 203C. The logging data 203A is manufacturing data when the heat-resistant insulated wire EW is manufactured. The simulation data 203B is simulation data showing the results of a simulation performed using the logging data. The quality prediction model 203C is a model for predicting the quality characteristics of the heat-resistant insulated wire EW. Details of the logging data 203A, the simulation data 203B, and the quality prediction model 203C will be described later.

[0042] The files are stored in a storage medium such as the hard disk drive 112 or memory. The information, data, signal values, variable values, and parameters stored in the storage medium such as the hard disk drive 112 or memory are read into the main memory or cache memory by the CPU 101 and used for operations of the CPU 101, such as extraction, search, reference, comparison, calculation, processing, editing, output, printing, and display. For example, during the above-mentioned operation of the CPU 101, the information, data, signal values, variable values, and parameters are temporarily stored in the main memory, registers, cache memory, buffer memory, etc.

[0043] The functions of the quality prediction system 100 may be realized by firmware stored in ROM 102, or may be realized by software alone, hardware alone such as elements, devices, boards, and wiring, a combination of software and hardware, or even a combination of firmware and firmware. The firmware and software are stored as programs in a storage medium such as a hard disk drive 112, a removable disk drive 108, or a CD / DVD-ROM drive 109. The programs are read and executed by the CPU 101. For example, the programs cause a computer to function as the quality prediction system 100.

[0044] Thus, the quality prediction system 100 is a computer equipped with a CPU 101 as a processing device, a hard disk drive 112 and memory as storage devices, a keyboard 105, a mouse 106, and a communication port 107 as input devices, and a display 104, a printer 110, and the communication port 107 as output devices. The functions of the quality prediction system 100 are realized by using the processing device, the storage device, the input device, and the output device.

[0045] <Function block> FIG. 4 is a diagram illustrating an example of functional blocks of the quality prediction system 100. As shown in FIG. 4, the quality prediction system 100 includes a first logging data acquisition unit 301, an average value calculation unit 302, a first simulation unit 303, a model creation unit 304, a data merging unit 305, a quality prediction model storage unit 306, a second logging data acquisition unit 307, a second simulation unit 308, and a prediction unit 309. Details of the functions realized by the first logging data acquisition unit 301, the average value calculation unit 302, the first simulation unit 303, the model creation unit 304, the data merging unit 305, the quality prediction model storage unit 306, the second logging data acquisition unit 307, the second simulation unit 308, and the prediction unit 309 will be described later.

[0046] <Creating a quality prediction model> Next, a quality prediction model creation process will be described, which creates a quality prediction model 203C for predicting the quality characteristics of the heat-resistant insulated electric wire EW produced by the extruder 10. Fig. 5 is a flowchart showing an example of the quality prediction model creation process. This process is realized by reading out a quality prediction model creation program 202A stored in the hard disk drive 112 and executing it with the CPU 101.

[0047] First, in step ST101, the CPU 101 acquires first logging data. The processing of step ST101 realizes the function of the first logging data acquisition unit 301 of the quality prediction system 100. The first logging data acquisition unit 301 acquires, from the hard disk drive 112, known first logging data detected and stored at predetermined intervals by multiple devices and sensors during the manufacture of the heat-resistant insulated electric wire EW. In other words, the acquired first logging data is known manufacturing data for each serial number of the same product during the manufacture of the heat-resistant insulated electric wire EW. Here, the multiple devices and sensors include, for example, the cylinder 33, the drawing device 20, the take-up device 50, the core wire temperature sensor S11, the head temperature sensor S12, the die temperature sensor S13, the neck temperature sensor S14, the water tank temperature sensor S15, the outer diameter measurement sensor S16, the resin temperature sensor S17, the resin pressure sensor S18, the temperature sensors S21 to S25, the air temperature sensor S31, and the humidity sensor S32.

[0048] FIG. 6 is a diagram showing an example of the first logging data 203A1. The first logging data 203A1 is data included in the logging data 203A. FIG. 6 shows the first logging data 203A1 for the heat-resistant insulated electric wire EW with production number AAA. Time is associated with factors such as the screw rotation speed (rpm), cylinder temperature (°C), linear velocity (m / s), etc. Note that, although not shown, the cylinder temperature factor includes the temperatures acquired by the temperature sensors S21 to S25. In this embodiment, the first logging data 203A1 is stored as part of the logging data 203A of the file group 203 at a predetermined interval of one second. Therefore, the first logging data 203A1 is acquired from the file group 203. In the figure, an example is shown in which the first logging data 203A1 includes three factors, but the first logging data 203A1 corresponds to time the logging data of a specific factor among all the logging data of factors that the quality prediction system 100 can acquire.

[0049] To create the quality prediction model 203C, for example, it is desirable that the CPU 101 acquires several tens or more pieces of known first logging data 203A1 for the same product. Furthermore, it is desirable that the number of factors acquired in the first logging data 203A1 be large. Of the factors acquired from the logging data 203A, it is desirable that the CPU 101 acquire as much logging data as possible of measurable factors, such as the screw rotation speed, cylinder temperature, and linear velocity, as shown in FIG. 6 , as the first logging data 203A1. Based on the settings, the CPU 101 can select any factor from all the factors included in the logging data 203A as the target for creating the quality prediction model 203C. The created quality prediction model 203C is stored in the hard disk drive 112.

[0050] Next, in step ST102, the CPU 101 calculates an average value. The processing of step ST102 realizes the function of the average value calculation unit 302 of the quality prediction system 100. The average value calculation unit 302 calculates an average value of the first logging data 203A1 for each factor from the first logging data 203A1 acquired by the first logging data acquisition unit 301 in step ST101. In this embodiment, if the logging data 203A includes data on the setup time and adjustment time before manufacturing the heat-resistant insulated electric wire EW, the CPU 101 performs processing to exclude the setup time and adjustment time from the first logging data 203A1. In this way, the CPU 101 can calculate the average value of the first logging data 203A1 for each factor from the start to the end of manufacturing the heat-resistant insulated electric wire EW.

[0051] 7 is a diagram showing an example of calculation result data D10 showing the calculation results of the average value of the first logging data 203A1. As shown in Fig. 7, the average screw rotation speed, average cylinder temperature, average linear velocity, etc. are calculated for each production number.

[0052] Next, the CPU 101 executes a simulation in step ST103. The processing of step ST103 realizes the function of the first simulation unit 303 of the quality prediction system 100. The first simulation unit 303 simulates the quality characteristics of the heat-resistant insulated electric wire EW using the average value of the first logging data 203A1. The CPU 101 calculates a feature quantity (described later) using the average value of the first logging data 203A1 calculated in step ST102 as a condition for CAE (Computer Aided Engineering) analysis. The feature quantity can also be considered a quality characteristic of the heat-resistant insulated electric wire EW obtained from the simulation data. It is desirable to use a large number of production numbers in the CAE analysis to improve the accuracy of correlation with the quality characteristics of the heat-resistant insulated electric wire EW. On the other hand, if the number of production numbers is large, the time required for the CAE analysis may become enormous. For this reason, it is desirable to set the number of production numbers so that the calculation load used in the CAE analysis does not exceed the limit of the CPU 101 while increasing the number of production numbers. The simulation result in step ST103 is defined as first simulation data. The first simulation data is stored in the hard disk drive 112 as simulation data 203B.

[0053] The aforementioned feature quantities are, for example, data that cannot be measured directly, such as the strain and residence time of the heat-resistant insulating material inside the extruder 10, or data that would require a great deal of effort to measure. The feature quantities may also include the cylinder temperature of the cylinder 33, the temperature of the heat-resistant insulating material inside the cylinder 33, and the resin pressure value. While these can be measured directly, only values ​​near the detectors where they are installed can be detected. In other words, the "average" temperature of the heat-resistant insulating material inside the extruder 10 and the "maximum" resin pressure can be derived from the position and value through simulation. Therefore, it is desirable for the CPU 101 to set factors whose values ​​can only be detected near the detectors as feature quantities and include them in the simulation.

[0054] Next, in step ST104, CPU 101 merges the result data measured in the quality inspection with the first simulation data. The processing of step ST104 realizes the function of the data merging unit 305 of the quality prediction system 100. The data merging unit 305 acquires quality result data of a quality test performed on the end of the heat-resistant insulated electric wire EW, and merges the quality result data with the first simulation data. The quality inspection is, for example, a destructive inspection. The quality inspection targets factors that can only be evaluated by destructive inspection, such as tensile strength, tensile elongation, flame retardancy, and oil resistance. If there is a factor among these factors that can determine the validity of the first simulation data, CPU 101 preferably simulates the factor and merges it with the quality result data.

[0055] Next, in step ST105, the CPU 101 creates a quality prediction model 203C. The processing of step ST105 realizes the function of the model creation unit 304 of the quality prediction system 100. The model creation unit 304 creates a quality prediction model 203C associated with the quality characteristics of the heat-resistant insulated electric wire EW, based on the average value of the first logging data 203A1 and first simulation data indicating the results of the simulation. The model creation unit 304 creates the quality prediction model 203C through machine learning, using quality characteristic data indicating the quality characteristics of the heat-resistant insulated electric wire EW as a target variable and the first logging data and first simulation data of the heat-resistant insulated electric wire EW as explanatory variables.

[0056] Multivariate analysis is used to create the quality prediction model 203C. There are many methods for performing multivariate analysis. For example, the CPU 101 can perform multiple regression analysis using the "sklearn.Linear model.Linear Regression" class of the Python machine learning library "scikit-learn," which allows for relatively easy calculation of the regression coefficients of each data. The CPU 101 may also use a library such as "SHAP" to interpret the quality prediction model 203C, determine the contribution rate, and examine whether the quality prediction model 203C is appropriate. In this way, the quality prediction model 203C is created using machine learning. The quality prediction model 203C created in this way is stored in the hard disk drive 112 by the function of the quality prediction model storage unit 306.

[0057] <Full length quality prediction> Next, a full-length quality prediction technique for predicting the quality characteristics of the heat-resistant insulated electric wire EW over its entire length using the created quality prediction model 203C will be described. Fig. 8 is a flowchart showing an example of processing for implementing the full-length quality prediction technique. This processing is implemented by reading out the quality prediction program 202B stored in the hard disk drive 112 and executing it with the CPU 101. Fig. 9 is a diagram for explaining the full-length quality prediction technique.

[0058] 8, the CPU 101 acquires second logging data in step ST201. The processing of step ST201 realizes the function of the second logging data acquisition unit 307 of the quality prediction system 100. For example, during the manufacture of the heat-resistant insulated electric wire EW, the second logging data acquisition unit 307 acquires second logging data 203A2 (see FIG. 10) during the manufacture of the heat-resistant insulated electric wire EW that is the prediction target.

[0059] FIG. 10 is a diagram showing an example of the second logging data 203A2. FIG. 10 shows the second logging data 203A2 for the heat-resistant insulated electric wire EW with production number FFF. Time is associated with factors such as the screw rotation speed (rpm), cylinder temperature (°C), linear velocity (m / s), etc. Note that the factor of the cylinder temperature, although not shown, includes the temperatures acquired by the temperature sensors S21 to S25. In this embodiment, during the manufacture of the heat-resistant insulated electric wire EW with production number FFF, the second logging data 203A2 is stored as part of the logging data 203A of the file group 203 at predetermined intervals of one second. Therefore, the second logging data 203A2 is acquired from the file group 203. In the figure, an example is shown in which the second logging data 203A2 includes three factors, but the second logging data 203A2 is data on any factor from among all logging data on factors that the quality prediction system 100 can acquire, and is, for example, logging data on the same factors as the first logging data 203A1 used to create the quality prediction model 203C.

[0060] Next, in step ST202, the CPU 101 executes continuous simulation at regular time intervals. The processing of step ST202 realizes the function of the second simulation unit 308 of the quality prediction system 100. The second simulation unit 308 executes a simulation from the second logging data 203A2 acquired by the second logging data acquisition unit 307 and the acquired second logging data 203A2 to create second simulation data.

[0061] 10, the second logging data 203A2 used for the simulation is different from the example shown in FIG. 9(a) and 9(b). Here, the CPU 101 acquires the cylinder temperature C, the screw rotation speed N, the tip resin pressure P detected by the resin pressure sensor S18, and the air temperature Ta detected by the air temperature sensor S31 as the second logging data 203A2, which are actual measurements. The CPU 101 then extracts the second logging data 203A2 acquired at regular intervals and performs continuous simulations using the extracted second logging data 203A2. The feature quantities to be simulated are the magnitude of strain γ, the average material temperature τ of the heat-resistant insulating material, and the residence time t of the heat-resistant insulating material in the cylinder 33.

[0062] In this embodiment, the second logging data 203A2 is often collected at intervals of, for example, one second, generally at intervals of one second or less. Therefore, the amount of second logging data 203A2 is enormous. Therefore, if a simulation is performed using all of the second logging data 203A2 as analysis conditions, the computational load on the CPU 101 becomes enormous. Considering the computational load on the CPU 101, it is desirable for the CPU 101 to thin out the second logging data 203A2 at regular intervals, for example, every 10 seconds or every 60 seconds, and extract the second logging data 203A2 to be used in the simulation from all of the second logging data 203A2. Furthermore, because the second simulation data is calculated at regular intervals, the second simulation data, which is the result of the simulation, is treated as continuous data rather than as individual data.

[0063] Next, in step ST203, the CPU 101 assigns the second logging data 203A2 and the second simulation data to the quality prediction model 203C. The processing of step ST203 realizes the function of the prediction unit 309 of the quality prediction system 100. The prediction unit 309 assigns the second logging data 203A2 and the second simulation data created by the second simulation unit 308 to the quality prediction model 203C to predict the quality characteristics of the heat-resistant insulated electric wire EW to be predicted.

[0064] For example, as shown in FIG. 9(c), the CPU 101 substitutes the second logging data 203A2 obtained at regular intervals and the second simulation data obtained at regular intervals into the quality prediction model 203C and performs machine learning. While advanced AI can be used for machine learning, the CPU 101 can perform multiple regression analysis relatively easily by using the aforementioned Python "scikit-learn" library, for example. Multiple regression analysis determines the regression coefficients a1, a2, ..., and intercept b of each variable. FIG. 9(c) shows an example of determining the relationship between tensile strength (predicted value) and tensile strength (measured value). In this example, the predicted values ​​and measured values ​​are concentrated near the line L1, and therefore the CPU 101 determines that the quality prediction model 203C is valid.

[0065] Next, in step ST204, the CPU 101 substitutes the data obtained in step ST203 into logging data 203A stored in the file group 203. That is, if the quality prediction model 203C is valid, the CPU 101 substitutes the obtained data, for example, the tensile strength (predicted value) shown in FIG. 9(c), into the original logging data 203A. As a result, as shown in FIG. 9(d), the CPU 101 can store the tensile strength values ​​obtained near the end of the production time, that is, values ​​that were only actually measured at a few points on the terminals of the heat-resistant insulated electric wire EW, as logging data 203A in the file group 203, in place of the tensile strength values ​​from the start to the end of production. This allows the quality prediction system 100 to predict fluctuations in the value of the quality characteristic for a set objective variable, such as tensile strength, over the entire length of the heat-resistant insulated electric wire EW.

[0066] Furthermore, the CPU 101 may calculate the objective variable, i.e., the predicted value of the quality characteristic, in near real time while the heat-resistant insulated wire EW is being manufactured. By displaying the result on a display device such as the display 104, the operator can constantly monitor fluctuations in the predicted value of the quality characteristic of the heat-resistant insulated wire EW. By further applying this system, for example, if the predicted value of the quality characteristic tends to exceed the appropriate range during the manufacture of the heat-resistant insulated wire EW, the CPU 101 can change the manufacturing parameters so that the predicted value of the quality characteristic tends to move toward the appropriate range. In other words, the CPU 101 can optimally control the manufacturing parameters.

[0067] <Example> Next, as an example, a more detailed description will be given of a process for predicting the overall length quality of the heat-resistant insulated electric wire EW manufactured by the manufacturing system 1. In this example, the quality characteristics to be predicted are the tensile strength and tensile elongation specified in JIS C3005.

[0068] First, the logging data used in this embodiment will be described. The quality prediction system 100 constantly stores 13 values ​​at 1-second intervals: the screw rotation speed of the screw in the cylinder 33, the current value for rotating the screw, the screw acceleration, the cylinder temperature at five different positions, the neck temperature, the head temperature, the die temperature, the resin temperature, and the tip resin pressure value. These 13 pieces of data are stored in a file group 203 of the hard disk drive 112 as logging data 203A.

[0069] Furthermore, a total of seven values, namely, the air temperature of the extruder 10, the humidity of the extruder 10, the core temperature of the core wire CW, the water temperature of the water tank 40, the outer diameter of the heat-resistant insulated electric wire EW, the take-up speed of the take-up device 50, and the take-up current value, are constantly stored at one-second intervals in the quality prediction system 100. These seven pieces of data are stored in a file group 203 of the hard disk drive 112 as logging data 203A.

[0070] In this embodiment, in order to take into consideration that the characteristics of the heat-resistant insulating material (resin) change due to rapid or slow cooling, the quality prediction system 100 also stores, at one-second intervals, two values: "resin temperature - air temperature," which is the resin temperature detected by the resin temperature sensor S17 minus the air temperature; and "resin temperature - water temperature," which is the resin temperature minus the water temperature. These two pieces of data are stored as logging data 203A in the file group 203 of the hard disk drive 112. As described above, in this embodiment, the CPU 101 uses data on 22 factors as the first logging data 203A1.

[0071] Next, as shown in FIG. 11, the CPU 101 executes a simulation. FIG. 11 is a diagram for explaining an outline of the simulation. As shown in FIG. 11, the CPU 101 executes the simulation using analysis conditions including screw structure information D1 indicating the structure of the screw in the cylinder 33, the screw rotation speed D2, the resin pressure value D3, the cylinder temperatures D4 of the temperature sensors S21 to S25, and the air temperature D5. The simulation software used in this embodiment is SingleScrewSimulator by HASL. Furthermore, in this embodiment, the simulation is executed for seven factors: "discharge volume," "maximum resin pressure," "driving force," "residence time," "average temperature," "strain," and "torque." The "maximum resin pressure," "driving force," "residence time," "average temperature," "strain," and "torque" are characteristic quantities of the heat-resistant insulating member in the cylinder 33 and the screw. The CPU 101 stores the seven first simulation data D11, which are the results of the simulation execution, namely, "discharge volume," "maximum resin pressure," "driving force," "residence time," "average temperature," "strain," and "torque," as simulation data 203B in, for example, a file group 203 of the hard disk device 112.

[0072] As described above, the CPU 101 creates a quality prediction model 203C for calculating the tensile strength and tensile elongation values, which are the objective variables, using a total of 29 variables, consisting of 22 pieces of first logging data 203A1 and 7 pieces of first simulation data, as explanatory variables.

[0073] In this example, 58 production numbers are prepared as learning models, and are analyzed by performing "multiple regression analysis" using the "sklearn.linear_model.LinearRecression" class of the Python machine learning library "scikitlearn," to calculate the regression coefficients a1 to a29 and intercept b of each explanatory variable (see Figure 9(c)).

[0074] FIG. 12 is a diagram showing the relationship between the measured tensile strength values ​​(measured strength) and predicted values ​​(predicted strength). FIG. 13 is a diagram showing the relationship between the measured tensile elongation values ​​(measured elongation) and predicted values ​​(predicted elongation). In FIG. 12, the average value of tensile strength is set to zero, and values ​​higher than the average are indicated by a "+" and values ​​lower than the average are indicated by a "-". The same is true for tensile elongation in FIG. 13. As shown in FIG. 12, there is a generally constant relationship between the measured strength and the predicted value. Therefore, CPU 101 was able to create a quality prediction model 203C for tensile strength as shown by dashed line L2. The coefficient of determination, which is the slope of dashed line L2, is 0.79, and the intercept is 0. Furthermore, as shown in FIG. 13, there is a generally constant relationship between the measured elongation and the predicted elongation. Therefore, CPU 101 was able to create a quality prediction model 203C for tensile elongation as shown by dashed line L3. The coefficient of determination, which is the slope, is 0.89, and the intercept is 0.

[0075] Next, the CPU 101 predicts the total length of the heat-resistant insulated electric wire EW to be predicted, which is manufactured by the manufacturing system 1, using the quality prediction model 203C for tensile strength and tensile elongation shown in FIGS.

[0076] 14 to 16 are diagrams showing examples of second logging data 203A2 acquired at one-second intervals. FIG. 14 is a diagram showing second logging data 203A2 of the screw rotation speed and cylinder temperature. The cylinder temperatures may be acquired from all temperatures detected by temperature sensors S21 to S25, or from any desired temperatures. For example, the CPU 101 acquires three temperatures detected by temperature sensors S23, S24, and S25 located close to the head 31. FIG. 15 is a diagram showing second logging data 203A2 of the current value of the cylinder 33 and the resin pressure value detected by the resin pressure sensor S18. FIG. 16 is a diagram showing second logging data 203A2 of the take-up speed of the take-up device 50 and the ambient air temperature. The CPU 101 extracts the second logging data 203A2 shown in FIGS. 14 to 16 every 60 seconds.

[0077] Next, the CPU 101 executes a simulation using the extracted second logging data 203A2. In order to evaluate the validity of the simulation results, the CPU 101 executes the simulation starting from the setup time, which should normally be excluded. Here, the setup time is the time before the actual start of manufacturing the heat-resistant insulated electric wire EW.

[0078] 17 to 19 are diagrams showing examples of second simulation data indicating the simulation results. FIG. 17 is a diagram showing second simulation data of maximum resin pressure and driving force. FIG. 18 is a diagram showing second simulation data of residence time and average temperature. FIG. 19 is a diagram showing second simulation data of strain and torque. As shown in FIGS. 17 to 19, it can be seen that the tendency of data continuity changes significantly between the setup time and the data acquisition time. Therefore, the CPU 101 can determine that the simulation results are valid.

[0079] Next, the CPU 101 substitutes the second logging data 203A2 extracted every 60 seconds from the second logging data shown in FIGS. 14 to 16 and the second simulation data every 60 seconds shown in FIGS. 17 to 19 into the quality prediction model 203C for tensile strength and tensile elongation shown in FIGS. 12 and 13. This allows the CPU 101 to calculate predicted values ​​of tensile strength and tensile elongation every 60 seconds over the entire length of the heat-resistant insulated electric wire EW. Therefore, the CPU 101 can predict the tensile strength and tensile elongation of the manufactured heat-resistant insulated electric wire EW, in other words, the quality characteristics, over the entire length of the heat-resistant insulated electric wire EW. The predicted values ​​of tensile strength and tensile elongation are stored as part of the logging data 203A in the file group 203 of the hard disk drive 112 so as to be included in the logging data 203A acquired in step ST201.

[0080] FIG. 20 is a graph showing the predicted results of the tensile strength of the heat-resistant insulated wire EW. FIG. 21 is a graph showing the predicted results of the tensile elongation of the heat-resistant insulated wire EW. As shown in FIGS. 20 and 21, the fluctuations in the predicted values ​​of tensile strength and tensile elongation are obtained from the start to the end of heat-resistant insulated wire production, that is, over the entire length of the heat-resistant insulated wire EW. In contrast, in a comparative example that does not utilize the technology of this embodiment, the tensile strength and tensile elongation values ​​can be obtained only at a few terminal points of the heat-resistant insulated wire EW. Therefore, the quality prediction system 100 of this embodiment can predict the tensile strength and tensile elongation over the entire length of the heat-resistant insulated wire EW, and therefore the quality prediction system 100 can obtain the state of variation in quality characteristics over the entire length of the heat-resistant insulated wire EW.

[0081] Furthermore, the quality prediction system 100 needs to create the quality prediction model 203C from the logging data of long products manufactured in the past, i.e., the known first logging data 203A1. Therefore, if a long product has been manufactured in the past, the technology of the quality prediction system 100 can be applied to that long product. In particular, for long products with a sufficient manufacturing history, the accuracy of the predicted value can be improved, and the quality prediction system 100 can significantly improve the effectiveness of prediction of quality characteristics.

[0082] Furthermore, the quality prediction system 100 is not limited to predicting the tensile strength and tensile elongation of long products, but can also predict the quality characteristics of long products, such as flame retardancy and oil resistance, which could only be evaluated by destructive testing of terminals.

[0083] In the above example, the quality prediction system 100 creates the quality prediction model 203C using the known first logging data 203A1 and first simulation data based on the known first logging data 203A1. The quality prediction system 100 predicts the quality characteristics of the long product using the second logging data 203A2 during production, second simulation data based on the second logging data 203A2 during production, and the quality prediction model 203C. However, the quality prediction system 100's quality prediction model creation and quality prediction process are not limited to this. For example, the quality prediction system 100 may create the quality prediction model based only on the known first logging data 203A1. The quality prediction system 100 may also predict the quality characteristics of the long product using the second logging data 203A2 during production and a quality prediction model created based only on the known first logging data 203A1. This reduces the processing load on the quality prediction system 100.

[0084] The invention made by the inventor has been specifically described above based on the embodiments thereof, but it goes without saying that the present invention is not limited to the above-described embodiments and can be modified in various ways without departing from the spirit of the invention. [Explanation of symbols]

[0085] 1. Manufacturing System 10. Extruder 20 Pull-out device 31 head 32 dice 33 cylinders 34 Material input section 40 aquarium 50 Take-off device 100 Quality Prediction System 101 CPU 112 Hard disk drive 202 Programs 202A Quality Prediction Model Creation Program 202B Quality Prediction Program 203A Logging data 203A1 First logging data 203A2 Second logging data 203B Simulation Data 203C Quality Prediction Model 301 First logging data acquisition unit 302 Average value calculation unit 303 Simulation Section 1 304 Model Creation Department 305 Data Merge Section 306 Quality prediction model memory unit 307 Second logging data acquisition unit 308 Second Simulation Section 309 Prediction Department CW core wire EW Heat-resistant insulated wire S11 core temperature sensor S12 Head temperature sensor S13 Die Temperature Sensor S14 Neck Temperature Sensor S15 Water Tank Temperature Sensor S16 outer diameter measurement sensor S17 Resin Temperature Sensor S18 Resin Pressure Sensor S21~S25 Temperature sensors S31 Air Temperature Sensor S32 Humidity Sensor

Claims

1. a first logging data acquisition unit that acquires known first logging data detected and stored by a plurality of detectors at predetermined intervals during the manufacture of the long product; an average value calculation unit that calculates an average value of each of the first logging data from the first logging data acquired by the first logging data acquisition unit; a model creation unit that creates a quality prediction model associated with quality characteristics of the long product based on an average value of the first logging data; A quality prediction model creation system comprising:

2. 2. The quality prediction model creation system according to claim 1, a first simulation unit that simulates quality characteristics of the long product using an average value of the first logging data; the model creation unit creates the quality prediction model based on an average value of the first logging data and first simulation data indicating a result of the simulation. Quality prediction model creation system.

3. 3. The quality prediction model creation system according to claim 2, a data merging unit that acquires quality result data of a quality test performed on the end of the long product and merges the quality result data with the first simulation data, the model creation unit uses the merged result of the data merging unit when creating the quality prediction model. Quality prediction model creation system.

4. 3. The quality prediction model creation system according to claim 2, the model creation unit creates the quality prediction model by machine learning using quality characteristic data indicating the quality characteristics of the long product as a target variable and the first logging data and the first simulation data of the long product as explanatory variables. Quality prediction model creation system.

5. 5. The quality prediction model creation system according to claim 4, The long product is an electric wire coated with a heat-resistant insulating material, The electric wire is manufactured by an extruder having a cylinder including a screw and a take-up device that takes up the electric wire, the explanatory variables include at least one of a screw rotation speed of the screw, a cylinder temperature of the cylinder, a linear speed at which the take-off device takes up the electric wire, a residence time of the heat-resistant insulating member in the cylinder, a pressure value of the heat-resistant insulating member, and a strain value of the heat-resistant insulating member; Quality prediction model creation system.

6. 5. The quality prediction model creation system according to claim 4, The long product is an electric wire coated with a heat-resistant insulating material, The response variable includes a tensile strength value or a tensile elongation value of the electric wire. Quality prediction model creation system.

7. a quality prediction model storage unit configured to store a quality prediction model associated with quality characteristics of the long product, the quality prediction model being created based on an average value of known first logging data of the long product; a second logging data acquisition unit that acquires second logging data during the manufacturing of the long product that is the prediction target; a prediction unit that predicts the quality characteristics of the long product to be predicted by substituting the second logging data acquired by the second logging data acquisition unit into the quality prediction model; Equipped with a quality prediction system.

8. The quality prediction system according to claim 7, a second simulation unit that creates second simulation data by executing a simulation from the second logging data acquired by the second logging data acquisition unit; the prediction unit predicts the quality characteristics of the long product to be predicted by substituting the second logging data acquired by the second logging data acquisition unit and the second simulation data created by the second simulation unit into the quality prediction model. Quality prediction system.

9. 9. The quality prediction system according to claim 8, the second simulation unit executes the simulation using logging data extracted at predetermined intervals from the second logging data acquired by the second logging data acquisition unit. Quality prediction system.

10. The quality prediction system according to claim 7, the prediction unit stores quality characteristic data indicating the quality characteristic together with the second logging data acquired by the second logging data acquisition unit; Quality prediction system.

11. Acquire known first logging data detected and stored at predetermined intervals by a plurality of detectors during the manufacture of the long product; Calculating an average value of each of the acquired first logging data; creating a quality prediction model associated with quality characteristics of the long product based on an average value of the first logging data; How to create a quality prediction model.

12. The quality prediction model creation method according to claim 11, After calculating the average values ​​of the first logging data, the average values ​​of the first logging data are used to simulate quality characteristics of the long product; creating the quality prediction model based on an average value of the first logging data and first simulation data indicating a result of the simulation; How to create a quality prediction model.

13. Acquire known first logging data detected and stored at predetermined intervals by a plurality of detectors during the manufacture of the long product; Calculating an average value of each of the acquired first logging data; creating a quality prediction model associated with quality characteristics of the long product based on an average value of the first logging data; acquiring second logging data during the manufacturing of the long product to be predicted; Substituting the second logging data into the quality prediction model to predict the quality characteristics of the long product to be predicted. Quality prediction methods.

14. 14. The quality prediction method according to claim 13, the quality prediction model is created using the first logging data and first simulation data obtained from the first logging data; After acquiring the second logging data, a simulation is performed from the second logging data to generate second simulation data; Substituting the second logging data and the second simulation data into the quality prediction model; Quality prediction methods.

Images