Estimation method, estimation device, and computer program

Abstract

The present invention obtains physical quantity data relating to the state of a component constituting a molding machine, prepares a plurality of estimation models for estimating the life or degree of abnormality of the component or the quality of a molded article by different algorithms, and uses one or more of the estimation models selected from the plurality of estimation models to estimate the quality of the molded article. On the basis of the acquired physical quantity data, the life of the component, the degree of abnormality, or the quality of the molded article is estimated.

Description

Technical Field

[0001] The present invention relates to an estimation method, an estimation device and a computer program. Background Art

[0002] Patent document 1 discloses a life prediction device for predicting the life of a rotating part in an injection molding machine, etc. The life prediction device of Patent document 1 accumulates the rotation speed of the rotating part rotated by a drive motor, and calculates the fatigue life of the rotating part based on the accumulated rotation speed and the torque required to rotate the rotating part.

[0003] Patent Document 2 discloses an abnormality inspection device including a vibration sensor for detecting vibration of a ball screw provided in an injection molding machine, and inspecting abnormality of the ball screw by analyzing the intensity of vibration detected by the vibration sensor.

[0004] Prior art literature

[0005] Patent Literature

[0006] Patent Document 1: Japanese Patent Application Laid-Open No. 6-91683

[0007] Patent Document 2: Japanese Patent Application Publication No. 2021-74917 Summary of the invention

[0008] Problems to be solved by the invention

[0009] It is considered to predict the life of components constituting a molding machine by collecting physical quantity data related to the states of the components and performing machine learning.

[0010] Although there are various life prediction models that can be used to predict the life of components constituting a molding machine, it is not always easy to determine an optimal life prediction model and collect sufficient data for learning the model. The same problem also occurs when estimating the degree of abnormality of components constituting a molding machine or the quality of molded products.

[0011] The purpose of the present disclosure is to provide an estimation method, an estimation device, and a computer program that can perform machine learning for estimating the life, etc. of a component constituting a molding machine, regardless of the amount of data that can be collected to estimate the life, degree of abnormality, or quality of a molded product, and estimate the life, degree of abnormality, or quality of the component.

[0012] Means for solving problems

[0013] According to an estimation method of one aspect of the present disclosure, physical quantity data related to the state of components constituting a molding machine are acquired, a plurality of estimation models are prepared for estimating the life or abnormality degree of the components or the quality of molded products by different algorithms, and the life or abnormality degree of the components or the quality of molded products are estimated based on the acquired physical quantity data using one or more estimation models selected from the plurality of estimation models.

[0014] According to one aspect of the present disclosure, an estimation device includes: an acquisition unit, which acquires physical quantity data related to the state of components constituting a molding machine; multiple estimation models, which estimate the life or abnormality of the components or the quality of molded products through different algorithms; and a processing unit, which uses one or more estimation models selected from the multiple estimation models to estimate the life or abnormality of the components or the quality of molded products based on the physical quantity data acquired by the acquisition unit.

[0015] A computer program (computer program product) according to one aspect of the present disclosure is used to cause a computer to perform the following processing: acquiring physical quantity data related to the state of components constituting a molding machine, using one or more estimation models selected from a plurality of estimation models that estimate the life or degree of abnormality of the components or the quality of molded products through different algorithms, and estimating the life or degree of abnormality of the components or the quality of molded products based on the acquired physical quantity data.

[0016] Effects of the Invention

[0017] According to the present disclosure, regardless of the amount of data that can be collected to estimate the life or abnormality of a component constituting a molding machine or the quality of a molded product, machine learning for estimating the life and the like can be performed to predict the life of the component. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a block diagram showing a configuration example of a molding machine system according to the present embodiment.

[0019] Figure 2 It is a schematic diagram showing a configuration example of the molding machine 1 according to the present embodiment.

[0020] Figure 3 It is a block diagram showing a configuration example of the data collection device 3 according to the present embodiment.

[0021] Figure 4 It is a block diagram showing a configuration example of a life prediction device according to an embodiment.

[0022] Figure 5 This is a conceptual diagram showing an example of a record layout of a collected data DB.

[0023] Figure 6This is a conceptual diagram showing an example of the record layout of the model configuration DB.

[0024] Figure 7 This is a flowchart showing the learning process procedure of the life prediction model.

[0025] Figure 8 is a conceptual diagram showing a first life prediction model.

[0026] Fig. 9 1 is a flowchart showing the learning process procedure of the first life prediction model.

[0027] Fig. 10A is a graph showing the learning results of the first life prediction model.

[0028] Fig. 10B : is a graph showing the inference result using the first life prediction model.

[0029] Fig.11 is a conceptual diagram showing a second life prediction model.

[0030] Fig.12 : is a flowchart showing the learning process procedure of the second life prediction model.

[0031] Fig.13A : is a graph showing the learning results of the second life prediction model.

[0032] Fig. 13B : is a graph showing the inference result using the second life prediction model.

[0033] Fig.14 : is a conceptual diagram showing the third life prediction model.

[0034] Fig.15 : is a flowchart showing the learning process procedure of the third life prediction model.

[0035] Fig.16A : is a graph showing the learning results of the third life prediction model.

[0036] Fig. 16B : is a graph showing the inference result using the third life prediction model.

[0037] Fig.17 This is a flowchart showing the processing procedure of life prediction.

[0038] Fig.18 This is a flowchart showing the processing procedure of life prediction.

[0039] Fig.19 This is a schematic diagram showing an example of a remaining life estimation result display screen displayed on a portal site.

[0040] Fig. 20 This is a schematic diagram showing an example of a dashboard screen displayed on a portal site.

[0041] Fig.21 This is a schematic diagram showing an example of a search display screen displayed on a portal site.

[0042] Fig. 22 This is a schematic diagram showing an example of a monthly report screen displayed on a portal site. DETAILED DESCRIPTION

[0043] The estimation method, life prediction device and computer program according to the embodiments of the present disclosure are described with reference to the following drawings. It should be noted that the present disclosure is not limited to these examples, but is intended to be shown by the claims and includes all changes within the meaning and scope equivalent to the claims. In addition, at least part of the embodiments described below can be combined arbitrarily.

[0044] Figure 1 1 is a block diagram showing a configuration example of a molding machine system according to the present embodiment. The molding machine system includes a molding machine 1, a plurality of sensors 2, a data collection device 3, a router 4, a life prediction device 5, and terminal devices 6a and 6b. The molding machine 1 includes an injection molding machine and an extruder. The following description takes the case where the molding machine 1 is an extruder as an example.

[0045] exist Figure 1 , one molding machine 1 and a data collection device 3 are shown, however, a plurality of data collection devices 3 not shown are connected to a life prediction device 5 via a network. The data collection device 3 is connected to one or more molding machines 1. The life prediction device 5 can collect information of each of the plurality of molding machines 1 and predict the remaining life of one or more components constituting each molding machine 1. The plurality of molding machines 1 and the data collection device 3 are installed in respective factories of a plurality of users who own the molding machines 1. The user is not an operator but an organization such as a legal person who owns the molding machine 1.

[0046] The terminal devices 6a and 6b are communication terminals with display units such as computers, tablet terminals, and smartphones. The terminal device 6a is a terminal used by the user. The terminal device 6b is a terminal used by service providers such as sales managers and maintenance managers associated with the user's molding machine 1.

[0047] <Molding machine 1>

[0048] Figure 2 1 is a schematic diagram showing a configuration example of a molding machine 1 according to the present embodiment. The molding machine 1 includes: a cylinder 10 having a hopper 10a for feeding a resin raw material; two screws 11; a mold 12 (see FIG. 1 ) provided at an outlet portion of the cylinder 10; Figure 1 The two screws 11 are arranged substantially parallel to each other in a meshing state and are rotatably inserted into the hole of the cylinder 10 to push the resin raw material fed into the hopper 10a in the extrusion direction ( Figure 1 and Figure 2 The molten resin raw material is discharged from the mold 12 having a through hole.

[0049] The screw 11 is formed by combining and integrating a plurality of types of screw parts. For example, the screw 11 is formed by arranging and combining a normal screw part for conveying the resin raw material in the forward direction, a reverse screw part for conveying the resin raw material in the reverse direction, a kneading part for kneading the resin raw material, etc. in an order and position according to the characteristics of the resin raw material.

[0050] In addition, the molding machine 1 includes: a motor 13 that outputs a driving force for rotating the screw 11; a reducer 14 that reduces the driving force of the motor 13; and a control device 15. The screw 11 is connected to the output shaft of the reducer 14. The screw 11 is rotated by the driving force of the motor 13 reduced by the reducer 14.

[0051] <Sensor 2>

[0052] The sensor 2 detects a physical quantity associated with the state of the components constituting the molding machine 1, and outputs the detected physical quantity data directly or indirectly to the data collection device 3. The physical quantity data is data showing the sensor value of the time series of the detected physical quantity. The sensor 2 includes: a sensor provided in the molding machine 1 as a sensor necessary for the operation control of the molding machine 1; and a sensor provided for estimating the life of the components. Some of the multiple sensors 2 are connected to the data collection device 3, and the data collection device 3 obtains the physical quantity data from the sensor 2. Some of the multiple sensors 2 are connected to the control device 15, and the data collection device 3 obtains the physical quantity data from the sensor 2 via the control device 15.

[0053] Physical quantities include temperature, position, velocity, acceleration, current, voltage, pressure, time, image data, torque, force, deformation, power consumption, weight, etc. These physical quantities can be measured using thermometers, position sensors, velocity sensors, acceleration sensors, ammeters, voltmeters, pressure gauges, timers, cameras, torque sensors, electricity meters, weight meters, etc.

[0054] The plurality of sensors 2 include, for example: a first sensor 21 for detecting physical quantities related to the reducer 14 ; a second sensor 22 for detecting physical quantities related to the screw 11 ; a third sensor 23 for detecting physical quantities related to the motor 13 ; and a fourth sensor 24 for detecting physical quantities related to the mold 12 .

[0055] The first sensor 21 is, for example, a vibration detector for detecting the vibration of the reducer 14. The second sensor 22 is a torque detector for detecting the shaft torque of the screw 11, a tachometer for detecting the rotation speed of the screw 11, a pressure gauge for detecting the pressure at the tip of the screw, a thermometer for detecting the temperature of the screw 11, a displacement sensor for detecting the displacement of the rotation center of the screw 11, etc. The third sensor 23 is an ammeter for detecting the motor current, a tachometer for detecting the motor rotation speed, etc. The fourth sensor 24 is a pressure gauge for detecting the die head pressure acting on the die 12.

[0056] <Control device 15>

[0057] The control device 15 is a computer that performs operation control of the molding machine 1 , and includes a transceiver (not shown) that transmits and receives information with the data collection device 3 , and a display unit.

[0058] Specifically, the control device 15 sends the operation data indicating the operation status of the molding machine 1 to the data collection device 3. The operation data include, for example, the motor current, the rotation speed of the screw 11, the pressure at the tip of the screw 11, the die head pressure, the feeder supply amount (supply amount of the resin raw material), the extrusion amount, the cylinder temperature, the resin pressure, and the like.

[0059] The control device 15 receives various chart data sent from the data collection device 3 and remaining life data showing the remaining life of the components constituting the molding machine 1. The control device 15 displays the contents of the received chart data and remaining life data. In addition, the control device 15 outputs a warning based on the remaining life shown in the received remaining life data.

[0060] <Data Collection Device 3>

[0061] Figure 3 1 is a block diagram showing a configuration example of a data collection device 3 according to the present embodiment. The data collection device 3 is a computer, and includes a control unit 31, a storage unit 32, a communication unit 33, and a data input unit 34, and the storage unit 32, the communication unit 33, and the data input unit 34 are connected to the control unit 31. The data collection device 3 is, for example, a PLC (Programmable Logic Controller).

[0062] The control unit 31 has a CPU (Central Processing Unit), a multi-core CPU, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array) and other arithmetic processing circuits, a ROM (Read Only Memory), a RAM (Random Access Memory) and other internal storage devices, I / O terminals, etc. The control unit 31 collects physical quantity data and sends it to the life prediction device 5 by executing a control program stored in the storage unit 32 described later. It should be noted that each functional unit of the data collection device 3 can be implemented in software, or partially or entirely in hardware.

[0063] The storage unit 32 is a nonvolatile memory such as a hard disk, an EEPROM (Electrically Erasable Programmable ROM), or a flash memory, and stores a control program for causing a computer to perform a collection process of physical quantity data.

[0064] The communication unit 32 is a communication circuit that sends and receives information based on a communication protocol specified by Ethernet (registered trademark) or the like. The communication unit 33 is connected to the control device 15 via a first communication network such as a LAN, and the control unit 31 can send and receive various information with the control device 15 via the communication unit 33. The control unit 31 acquires physical quantity data via the communication unit 33.

[0065] The router 4 is connected to the first network, and the communication unit 33 is connected to the life prediction device 5 on the cloud as the second communication network via the router 4. The control unit 31 can transmit and receive various information to and from the life prediction device 5 via the communication unit 33 and the router 4.

[0066] The data input unit 34 is an input interface for inputting a signal output from the sensor 2 . The sensor 2 is connected to the data input unit 34 , and the control unit 31 acquires physical quantity data via the data input unit 34 .

[0067] <Life Prediction Device 5>

[0068] Figure 4 : is a block diagram showing a configuration example of the life prediction device 5 according to the embodiment. The life prediction device 5 is a computer, and includes a processing unit 51, a storage unit 52, and a communication unit 53. The storage unit 52 and the communication unit 53 are connected to the processing unit 51.

[0069] The processing unit 51 is a processor, and has a CPU, a multi-core CPU, a GPU (Graphic Processing Unit), a GPGPU (General-purpose computing on graphics processing units), a TPU (Tensor Processing Unit), an ASIC, an FPGA, an NPU (Neural Processing Unit), and other computing circuits, an internal storage device such as a ROM and a RAM, an I / O terminal, etc. The processing unit 51 functions as the life prediction device 5 according to the present embodiment by executing a computer program (computer program product) P stored in the storage unit 52 described later. It should be noted that each functional unit of the life prediction device 5 can be implemented in software, or partially or entirely in hardware.

[0070] The communication unit 53 is a communication circuit that sends and receives information based on a communication protocol specified by Ethernet (registered trademark) or the like. The communication unit 33 is connected to the data collection device 3 and the terminal devices 6a, 6b via the second communication network, and the control unit 31 can send and receive various information with the data collection device 3 and the terminal devices 6a, 6b via the communication unit 53.

[0071] The storage unit 52 is a nonvolatile memory such as a hard disk, an EEPROM, or a flash memory. The storage unit 52 stores a computer program P for causing a computer to perform a process of estimating the life of components constituting the molding machine 1, a first life prediction model M1, a second life prediction model M2, a third life prediction model M3, a collected data DB 52a, and a model configuration DB 52b.

[0072] The computer program P, etc. may be recorded in a recording medium 50 in a form that can be read by a computer. The storage unit 52 stores the computer program P, etc. read from the recording medium 50 by a reading device not shown in the figure. The recording medium 50 is a semiconductor memory such as a flash memory. In addition, the recording medium 50 may be an optical disc such as a CD (Compact Disc: Compact Disc)-ROM, a DVD (Digital Versatile Disc: Digital Versatile Disc)-ROM, a BD (Blu-ray (registered trademark) disc), etc. In addition, the recording medium 50 may also be a magnetic disk such as a floppy disk, a hard disk, a magneto-optical disk, etc. In addition, the computer program P, etc. may be downloaded from an external server not shown in the figure connected to a communication network not shown in the figure, and the storage unit 52 may store it.

[0073] The first life prediction model M1 includes an image recognition learning model 73, which outputs data indicating the remaining life of components constituting the molding machine 1 when image data generated based on physical quantity data is input (see Figure 8 ). The image recognition learning model 73 has, for example, a CNN (Convolutional Neural Network). The image recognition learning model 73 of the first life prediction model M1 pre-stored in the storage unit 52 is an existing learned model. The processing unit 51 can generate a learning model dedicated to the residual life prediction of a specific molding machine 1 and a component by performing transfer learning or fine-tuning of the existing learned model.

[0074] It should be noted that, for the sake of convenience, the learning model and life prediction model that have undergone transfer learning or fine-tuning are also referred to as the image recognition learning model 73 and the first life prediction model M1.

[0075] The second life prediction model M2 includes: one or more state quantity estimation mathematical models 81, which show the relationship between physical quantity data representing multiple types of physical quantities that can be detected by the sensor 2 and other state quantities other than the physical quantity; and a life estimation mathematical model 82 (see Fig.11 ). The state quantity estimation mathematical model 81 of the second life prediction model M2 pre-stored in the storage unit 52 is an existing mathematical model. The coefficients of the state quantity estimation mathematical model 81 are pre-adjusted using a test molding machine or the like. The life estimation mathematical model 82 is also preferably pre-adjusted using a test molding machine or the like. The processing unit 51 can generate a mathematical model dedicated to the residual life prediction of a specific molding machine 1 and a component by performing transfer learning or fine-tuning on the state quantity estimation mathematical model 81 and the life estimation mathematical model 82.

[0076] It should be noted that, for the sake of convenience, the mathematical model and life prediction model that have undergone transfer learning or fine-tuning are also referred to as the state quantity estimation mathematical model 81, the life estimation mathematical model 82, and the second life prediction model M2.

[0077] The third life prediction model M3 has one or more explanatory variables and has a life estimation function (regression equation) 92 (see Fig.14 The coefficients of the life estimation function 92 are preferably roughly adjusted in advance using a molding machine for testing, etc. The processing unit 51 can generate an estimation function dedicated to the remaining life prediction of a specific molding machine 1 and component by performing transfer learning or fine adjustment of the life estimation function 92.

[0078] It should be noted that, for the sake of convenience, the estimation function and the life prediction model after transfer learning or fine-tuning are also referred to as the life estimation function 92 and the third life prediction model M3.

[0079] It should be noted that the first life prediction model M1 is a model for predicting the life of the components constituting the molding machine 1 based on the time series physical quantity data obtained through short-term measurement, and the second life prediction model M2 and the third life prediction model M3 are models for predicting the life of the components constituting the molding machine 1 based on the time series physical quantity data obtained through long-term measurement. Of course, the first to third life prediction models M1, M2 and M3 can also be configured to predict the life of the components constituting the molding machine 1 based on the time series physical quantity data measured and obtained during the same period. It can also be configured so that the measurement time span of the physical quantity data used for predicting the life of the components constituting the molding machine 1 can be set for each life prediction model.

[0080] It should be noted that the first to third life prediction models M1, M2, and M3 are examples of models that predict the life of components of the molding machine 1 using different algorithms, and the number of life prediction models is not limited to three. In addition, the types of life prediction models are not particularly limited to the above types.

[0081] For example, the first to third life prediction models M1, M2, and M3 can be composed of multilayer perceptron (MultilayerPerceptron: MLP), convolutional neural network (Convolutional Neural Network: CNN), graph neural network (Graph Neural Network: GNN), graph convolutional neural network (Graph Convolutional Network: GCN), RNN (Recurrent Neural Network), LSTM (Long Short Term Memory), and other neural network models. Alternatively, the first life prediction model M1 can be constructed using algorithms such as decision trees, random forests, and SVM (Support Vector Machine: Support Vector Machine).

[0082] Figure 5 : is a conceptual diagram showing an example of the record layout of the collected data DB 52a. The collected data DB 52a includes a hard disk and a DBMS (DataBase Management System), and stores various physical quantity data collected from the molding machine 1. For example, the collected data DB 52a has a "No." (record number) column, a "device ID" column, a "operation date and time" column, a "vibration data" column, and a "shaft torque" column.

[0083] The "Device ID" column stores the device identifier of the molding machine 1. The "Operation Date and Time" column stores information indicating the year, month, date and time when various data stored as records were obtained. The "Operation Data" column stores time-series physical quantities indicating the operating status of the molding machine 1, such as motor current, rotation speed of the screw 11, front end pressure of the screw 11, die head pressure, feeder supply amount (resin raw material supply amount), extrusion amount, cylinder temperature, resin pressure, etc. The "Vibration Data" column stores vibration data as time-series physical quantity data. The "Shaft Torque Data" column stores torque data of the screw 11 as time-series physical quantity data.

[0084] Figure 6 : is a conceptual diagram showing an example of the record layout of the model configuration DB52b. The model configuration DB52b includes a hard disk and a DBMS (DataBase Management System), and stores coefficients of various life prediction models for predicting the life of components constituting the molding machine 1. For example, the model configuration DB52b has a "user ID" column, a "device ID" column, a "component ID" column, a "first life prediction model" column, a "sensor ID used for the first life prediction" column, a "second life prediction model" column, a "sensor ID used for the second life prediction" column, a "third life prediction model" column, and a "sensor ID used for the third life prediction" column.

[0085] The “user ID” column stores the identifier of the user of the molding machine 1. The “device ID” column stores the device identifier of the molding machine 1. The “component ID” column stores the component identifier of the component constituting the molding machine 1 as the remaining life prediction target.

[0086] The "first life prediction model" column stores the first life prediction model M1, which is transfer-learned or fine-tuned in order to predict the remaining life of the component of the molding machine 1 represented by the device ID and the component ID. For example, the "first life prediction model" column stores the weight coefficients and various parameters of the image recognition learning model 73. The "sensor ID used for the first life prediction" column stores the identifier of the sensor 2 used to predict the remaining life of the component using the first life prediction model M1. In other words, the type of physical quantity data used to predict the remaining life of the component is stored. For example, the identifier of the sensor 2 that detects the motor current, the identifier of the sensor 2 that detects the vibration of the reducer 14, the identifier of the sensor 2 that detects the axial torque of the screw 11, etc. are stored.

[0087] The "Second Life Prediction Model" column stores the second life prediction model M2 that has been transfer-learned or fine-tuned to predict the remaining life of the component of the molding machine 1 indicated by the equipment ID and component ID. For example, the "Second Life Prediction Model" column stores the weight coefficients and various parameters of the state quantity estimation mathematical model 81 and the life estimation mathematical model 82. The "Sensor ID used for the second life prediction" column stores the identifier of the sensor 2 used to predict the remaining life of the component using the second life prediction model M2.

[0088] The "third life prediction model" column stores the third life prediction model M3 that has been transfer-learned or fine-tuned to predict the remaining life of the component of the molding machine 1 indicated by the equipment ID and the component ID. For example, the "third life prediction model" column stores the coefficients of the life estimation function 92. The "sensor ID used for the third life prediction" column stores the identifier of the sensor 2 used to predict the remaining life of the component using the third life prediction model M3.

[0089] like Figure 6 As shown, it is possible to configure that when predicting the remaining life of a component, physical quantity data obtained by different sensors 2 are used, and different life prediction models are used to predict the remaining life. For example, when predicting the remaining life of the reducer 14, it is possible to configure that the remaining life is calculated using the first life prediction model M1 based on the physical quantity data of the first part obtained using the first sensor 21, and the remaining life is calculated using the second life prediction model M2 or the third life prediction model M3 based on the physical quantity data of the second part obtained using the second sensor 22, and the average value of the calculated remaining life is calculated, etc. The physical quantity data obtained by the first sensor 21 is, for example, the vibration data of the reducer 14, and the physical quantity data obtained by the second sensor 22 is the axial torque of the screw 11.

[0090] It should be noted that if Figure 6 As shown, the model configuration DB52b can be configured to store information of a life prediction model used for prediction of the remaining life according to the molding machine 1 and the component, and not store information of a life prediction model not used for prediction of the remaining life. In addition, the model configuration DB52b can be configured to store all the information of the first to third life prediction models M1, M2, and M3 learned using the physical quantity data that can be collected, regardless of whether transfer learning or fine-tuning can be performed. In addition, the model configuration DB52b can store weight coefficients corresponding to the prediction accuracy of the first to third life prediction models M1, M2, and M3 in correspondence with the equipment ID and the component ID. The weight coefficient is a coefficient used to calculate the weighted average of the life predictions calculated by the first to third life prediction models M1, M2, and M3.

[0091] <Machine Learning Processing>

[0092] Figure 7 The processing unit 51 of the life prediction device 5 acquires physical quantity data related to the states of components constituting the molding machine 1 from the data collection device 3 (step S11). The processing unit 51 stores the acquired physical quantity data in the collected data DB 52a.

[0093] Next, the processing unit 51 performs machine learning processing of the first life prediction model M1 based on the acquired physical quantity data (step S12). The processing unit 51 performs machine learning processing of the second life prediction model M2 based on the acquired physical quantity data (step S13). The processing unit 51 performs machine learning processing of the third life prediction model M3 based on the acquired physical quantity data (step S14).

[0094] It should be noted that the processing unit 51 may be configured to selectively learn the first to third life prediction models M1, M2, and M3 according to the components of the molding machine 1 that are the life prediction targets. For example, when there is little physical quantity data for machine learning, only the third life prediction model M3 may be learned, and when sufficient physical quantity data is accumulated, all of the first to third life prediction models M1, M2, and M3 may be learned.

[0095] The processing unit 51 that has completed the learning of the first to third life prediction models M1, M2, and M3 will store the data used to construct the learned first to third life prediction models M1, M2, and M3 in the model construction DB52b in correspondence with the equipment ID and component ID of the molding machine 1 and components representing the life prediction object (step S15).

[0096] The data used to constitute the first life prediction model M1 that has been learned are, for example, the weight coefficients of the image recognition learning model 73. The data used to constitute the second life prediction model M2 that has been learned are, for example, the coefficients of the state quantity estimation mathematical model 81 and the life estimation mathematical model 82. The data used to constitute the third life prediction model M3 that has been learned are, for example, the coefficients of the life estimation function 92.

[0097] Figure 8: is a conceptual diagram showing the first life prediction model M1. The first life prediction model M1 includes a frequency analysis unit 71, an image generation unit 72 and an image recognition learning model 73. The frequency analysis unit 71 is an operation processing unit that Fourier transforms the physical quantity data of the time series into the physical quantity data of the frequency component. The frequency analysis unit 71 preferably performs the Fourier transform of the physical quantity data by short-time Fourier transform (STFT: Short-Time Fourier Transform). The image generation unit 72 is an operation processing unit that transforms the physical quantity data after the Fourier transform into image data represented by an image. For example, the physical quantity data is preferably represented in an image plane in which the horizontal axis of the image is set to the frequency and the vertical axis is set to the size of the frequency component. Below, the image obtained by Fourier transforming the physical quantity data is referred to as a Fourier transform image.

[0098] It should be noted that the frequency analysis method is not limited to STFT, and wavelet transform, Stockwell transform, Wigner distribution function, empirical mode decomposition, Hilbert-Huang transform, etc. can also be used.

[0099] The image recognition learning model 73 is a convolutional neural network (CNN), and includes an input layer 73a to which image data of a Fourier transformed image is input, an intermediate layer 73b, and an output layer 73c to output remaining life data indicating the remaining life of a component.

[0100] The input layer 73a has a plurality of nodes, and the pixel values ​​of each pixel constituting the Fourier transform image are input to these nodes. The intermediate layer 73b has a structure that is alternately connected to a convolution layer that convolves the pixel values ​​of each pixel of the Fourier transform image input to the input layer 73a and a pooling layer that maps the pixel values ​​convolved by the convolution layer. The intermediate layer 73b extracts the feature quantity of the Fourier transform image while compressing the image information of the Fourier transform image, and outputs the extracted feature quantity of the Fourier transform image, that is, the physical quantity data, to the output layer 73c.

[0101] The output layer 73 c has a node that outputs remaining life data indicating the remaining life of a component at the time point of physical quantity data measurement.

[0102] It should be noted that although CNN is shown as an example of the image recognition learning model 73, it can also be composed of a multilayer perceptron (Multilayer Perceptron: MLP), a convolutional neural network (Convolutional Neural Network: CNN), a graph neural network (Graph Neural Network: GNN), a graph convolutional neural network (Graph Convolutional Network: GCN), RNN (Recurrent Neural Network: Recurrent Neural Network), LSTM (Long Short Term Memory: Long Short Term Memory), and other neural network models. In addition, the first life prediction model M1 can be composed of algorithms such as decision trees, random forests, and SVM (Support Vector Machine: Support Vector Machine).

[0103] Fig. 9 : is a flowchart showing the learning processing sequence of the first life prediction model M1. The processing unit 51 performs frequency analysis on the physical quantity data of the time series accumulated in the collected data DB52a, converts it into the physical quantity of the frequency component (step S21), and generates a Fourier transform image based on the physical quantity data after Fourier transformation (step S22). Then, the processing unit 51 marks the remaining life of the component at the time point measured to the physical quantity data based on the Fourier transform image on the Fourier transform image (step S23). Next, the processing unit 51 uses the Fourier transform image marked with the remaining life as learning data (training data) to perform machine learning of the first life prediction model M1 (step S24). The storage unit 52 stores the existing image recognition learning model 73, and through transfer learning or fine-tuning of the image recognition learning model 73, it is possible to generate a first life prediction model M1 dedicated to the residual life prediction of a specific molding machine 1 and component. More specifically, the processing unit 51 optimizes the weight coefficients of the image recognition learning model 73 by using the error back propagation method, the error gradient descent method, or the like using the learning data, thereby causing the image recognition learning model 73 to perform machine learning.

[0104] Fig. 10A is a graph showing the learning result of the first life prediction model M1, Fig. 10B : is a graph showing the inference result using the first life prediction model M1. Fig. 10A and Fig. 10B The horizontal axis shows the number of days that have passed, and the vertical axis shows the remaining life (estimated life). The component to be predicted is, for example, a gear or a bearing of the reducer 14, and the physical quantity is the vibration of the reducer 14. Fig. 10AThe machine learning of the image recognition learning model 73 is performed using the learning data obtained during the period shown (days 0 to 7 months). The straight arrows represent the marked remaining life (lifetime mark). The dot graph shows the prediction results of the remaining life obtained by inputting the Fourier transform image of the learning data into the image recognition learning model 73. It should be noted that in the learning stage, components with a life of more than a certain amount are judged as components that will not be damaged.

[0105] exist Fig. 10B In FIG. 7 , the dot graph on the left side (days 0 to 2 months) shows the prediction results (estimated life) of the remaining life obtained by inputting the Fourier transform image obtained from the molding machine 1 including the used parts (short remaining life) into the image recognition learning model 73. Fig. 10B In the inference stage, the components whose residual life is predicted to be above a certain level (e.g. Fig. 10B The range in A) is not displayed as a dot plot. Fig. 10A and 10B As shown, it can be seen that the remaining life of the component can be predicted by the image recognition learning model 73 that has been learned.

[0106] Fig.11 : is a conceptual diagram showing the second life prediction model M2. The second life prediction model M2 includes a state quantity estimation mathematical model 81 and a life estimation mathematical model 82. The state estimation model is a function that specifies the relationship between a plurality of physical quantity data representing each of a plurality of types of physical quantities that can be detected using the sensor 2 and other state quantities other than the plurality of physical quantities. The plurality of physical quantities and other state quantities include, for example, the vibration of the reducer 14, the supply amount of the resin raw material, the motor current, the rotation speed of the screw 11, the type of the resin raw material, the screw composition, etc. The life estimation mathematical model 82 is a function that estimates the remaining life of the components of the molding machine 1 based on the physical quantity data and the physical quantity data estimated by the state quantity estimation mathematical model 81.

[0107] Fig.12It is a flowchart showing the learning processing sequence of the second life prediction model M2. The processing unit 51 marks the physical quantity data accumulated in the collected data DB52a with the remaining life of the component when the physical quantity data is measured (step S31). Next, the processing unit 51 uses the physical quantity data marked with the remaining life as learning data to optimize the coefficients of the state quantity estimation mathematical model 81 and the life estimation mathematical model 82 (step S32). The state quantity estimation mathematical model 81 and the life estimation mathematical model 82 can be expressed by regression equations, and the processing unit 51 can optimize the coefficients of the mathematical model by, for example, kernel ridge regression. The storage unit 52 stores the state quantity estimation mathematical model 81 and the life estimation mathematical model 82 that have been roughly adjusted in advance, and can generate the second life prediction model M2 by optimizing the coefficients of the mathematical model.

[0108] It should be noted that the method of optimizing the coefficients of the state quantity estimation mathematical model 81 and the life estimation mathematical model 82 is not limited to kernel ridge regression.

[0109] Fig.13A is a graph showing the learning results of the second life prediction model M2, Fig. 13B : is a graph showing the inference result using the second life prediction model. Fig.13A and Fig. 13B The horizontal axis shows the number of days that have passed, and the vertical axis shows the remaining life (estimated life). The prediction target component is, for example, a gear or a bearing of the reducer 14, and the physical quantity is the vibration of the reducer 14. Fig.13A The learning data obtained during the period shown (0 to 12 elapsed days) is used to perform machine learning of the state quantity estimation mathematical model 81 and the life estimation mathematical model 82. The solid line represents the marked remaining life (lifetime mark). The dot graph shows the remaining life predicted using the second life prediction model M2.

[0110] exist Fig. 13B In FIG. 13 , the dot graph shown in the range of the period shown in the right period (3 to 8 months elapsed) shows the remaining life predicted based on the RMS of the physical quantity data obtained from the molding machine 1 including new parts (long remaining life). In FIG. 13 , the dot graph shown in the range of the period (0 to 2 months) on the left shows the remaining life predicted based on the RMS of the physical quantity data obtained from the molding machine 1 including used parts (short remaining life). Fig.13A and Fig. 13B As shown, it can be seen that the remaining life of the component can be predicted by the state quantity estimation mathematical model 81 and the life estimation mathematical model 82 that have been learned.

[0111] Fig.14: is a conceptual diagram showing a third life prediction model M3. The third life prediction model M3 includes an RMS calculation unit 91 and a life estimation function 92. The RMS calculation unit 91 is an operation processing unit that calculates the root mean square of physical quantity data. The life estimation function 92 is a function that estimates the remaining life of a component of the molding machine 1 based on the RMS of the physical quantity data. The life estimation function 92 is a regression equation that represents the remaining life as a target variable by one or more explanatory variables.

[0112] Fig.15 : is a flowchart showing the learning processing sequence of the third life prediction model M3. The processing unit 51 calculates the RMS of the physical quantity based on the physical quantity data accumulated in the collected data DB52a (step S41), and marks the remaining life of the component at the time point when the physical quantity data is measured to the calculated RMS (step S42). Next, the processing unit 51 uses the RMS marked with the remaining life as learning data to optimize the coefficients of the life estimation function 92 (step S43). The processing unit 51 can optimize the coefficients of the life estimation function 92 by, for example, Lasso Regression. The storage unit 52 stores the life estimation function 92 that has been roughly adjusted in advance, and can generate the third life prediction model M3 by optimizing the coefficients of the life estimation function 92.

[0113] Fig.16A is a graph showing the learning results of the third life prediction model M3, Fig. 16B : is a graph showing the inference result using the third life prediction model M3. Fig.16A and Fig. 16B The horizontal axis shows the number of days that have passed, and the vertical axis shows the remaining life. The prediction target component is, for example, a gear or a bearing of the reducer 14, and the physical quantity is the vibration of the reducer 14. Fig.16A The machine learning of the service life estimation function 92 is performed using the learning data obtained during the period shown (1 to 16 elapsed days). The solid line shows the marked remaining life. The dot graph shows the remaining life predicted by the service life estimation function 92.

[0114] exist Fig. 16B In FIG. 1 , the dot graph on the upper right side shows the residual life predicted based on the RMS of the physical quantity data obtained from the molding machine 1 including the new part (long residual life). Fig. 16B , the dot graph on the lower left shows the residual life predicted based on the RMS of the physical quantity data obtained from the molding machine 1 including the used parts (short residual life). Fig.16A and Fig. 16B As shown, it can be seen that the remaining life of the component can be predicted by the learned life estimation function 92 .

[0115] <Residual life prediction>

[0116] Fig.17 and Fig.18 is a flowchart showing the processing sequence of life prediction. The processing unit 51 of the life prediction device 5 acquires physical quantity data from the data collection device 3 (step S51). Next, the processing unit 51 selects one or more life prediction models to be used for prediction of the life prediction target component (step S52). Specifically, the processing unit 51 reads the life estimation mathematical model 82 corresponding to the device ID and component ID representing the specific molding machine 1 and component from the model configuration DB 52b.

[0117] The processing unit 51 determines whether to use the first life prediction model M1 (step S53). When it is determined that the first life prediction model M1 is used (step S53: Yes), the processing unit 51 performs frequency analysis on the acquired time series physical quantity data, converts it into the physical quantity of the frequency component (step S54), and generates image data of the Fourier transformed image based on the physical quantity data after the Fourier transformation (step S55). Then, the processing unit 51 predicts the remaining life of the component by inputting the generated Fourier transformed image data into the image recognition learning model 73 (step S56).

[0118] When the processing of step S56 is finished, or when it is determined in step S53 that the first life prediction model M1 is not used (step S53: No), the processing unit 51 determines whether to use the second life prediction model M2 (step S57). When it is determined that the second life prediction model M2 is used (step S57: Yes), the processing unit 51 substitutes the acquired physical quantity data into the state quantity estimation mathematical model 81 to estimate the unknown state quantity (step S58). Next, the processing unit 51 substitutes the acquired physical quantity data and the calculated unknown state quantity into the life estimation mathematical model 82 to estimate the remaining life of the component (step S59).

[0119] When the processing of step S59 is finished, or when it is determined in step S57 that the second life prediction model M2 is not used (step S57: No), the processing unit 51 determines whether to use the third life prediction model M3 (step S60). When it is determined that the third life prediction model M3 is used (step S60: Yes), the processing unit 51 calculates the RMS of the physical quantity based on the acquired physical quantity data (step S61). Then, the processing unit 51 substitutes the calculated RMS of the physical quantity data into the life estimation function 92 to predict the remaining life of the component (step S62).

[0120] When the processing of step S62 is terminated, or when it is determined in step S60 that the third life prediction model M3 is not used (step S60: No), the remaining life of the component is calculated based on the first to third life estimation models used (step S63). For example, the processing unit 51 calculates the minimum value, average value, median value, weighted average value, etc. of the remaining life calculated by the first to third life estimation models, respectively.

[0121] The weight coefficient used for weighted averaging can be configured to be arbitrarily set by the user, or can be set based on feedback information of past life prediction accuracy (e.g., the difference between the calculated remaining life and the actual remaining life). Further, the processing unit 51 can select the remaining life calculated by a life prediction model selected by the user. The processing unit 51 can calculate the minimum value, average value, median value, weighted average value, etc. of the remaining life calculated by multiple life prediction models selected by the user.

[0122] Next, the processing unit 51 calculates the characteristic frequency intensity representing the characteristics of each component constituting the molding machine 1 at each moment of the time series based on the acquired time series physical quantity data (step S64). Then, the processing unit 51 creates a time series graph of the calculated characteristic frequency intensity (step S65). The time series graph is a graph in which the horizontal axis is time and the vertical axis is characteristic frequency intensity, and represents the characteristic frequency intensity of the above-calculated time series.

[0123] Next, the processing unit 51 performs a display process of the remaining life and the time series graph of each component constituting the molding machine 1 (step S66). The processing unit 51 displays the remaining life and the time series graph of each component on a portal website where the user can view the status of each molding machine 1 owned by the user through a browser.

[0124] In addition, the processing unit 51 transmits the calculated remaining life and time series graph data of each component constituting the molding machine 1 to the terminal devices 6a and 6b (step S67). For example, the processing unit 51 transmits the remaining life and time series graph data to the terminal device 6a of the user or operator of the molding machine 1 and the terminal device 6b of the person in charge of providing services related to the molding machine 1 of the user.

[0125] The life prediction device 5 according to the present embodiment provides a portal site that allows viewing of the remaining life of each of the plurality of molding machines 1 and components, and a time series graph of physical quantities related to the status of each component of the molding machine 1. The life prediction device 5 according to the present embodiment manages the status of the molding machines 1 respectively owned by a plurality of users, and each user can view and confirm the status of the molding machine 1 owned by the user by accessing the portal site using a browser. In addition, the person in charge of providing services related to the molding machines 1 and components can also view and confirm the status of the user or molding machine 1 in charge by accessing the portal site using a browser.

[0126] Fig.19 1 is a schematic diagram showing an example of a remaining life estimation result display screen 101 displayed on a portal website. The remaining life estimation result display screen 101 includes: a device name display unit 111, which displays the names of each of the plurality of molding machines 1 owned or used by the user; and a component name display unit, which displays the names of the components constituting each molding machine 1. In addition, the remaining life estimation result display screen 101 also includes a component status display unit 113, which displays the status of each component, especially the status based on the remaining life. The component status display unit 113 displays the status of each component by displaying icons such as "good", "need to be confirmed", and "abnormal". When the icon of the component status display unit 113 is operated, the processing unit 51 displays a time series graph of a physical quantity related to the status of the component. In addition, the remaining life estimation result display screen 101 includes a remaining life display unit 114 that displays the remaining life of the component with abnormality. The processing unit 51 displays the remaining life of the component calculated by the above-mentioned processing.

[0127] Fig. 20 1 is a schematic diagram showing an example of a dashboard screen 102 displayed on the portal site. The dashboard screen 102 is a screen that displays a time series graph of one or more representative physical quantity data indicating the state of the molding machine 1. The dashboard screen 102 has a display operation unit 121 for specifying the displayed molding machine 1 and period, etc., and a time series graph display unit 122. The display operation unit 121 has, for example, a model specifying unit for specifying the model of the molding machine 1, a period specifying unit for indicating the displayed period, a screw speed specifying unit for specifying the screw speed, a format display unit for specifying the display format of the graph displayed on the dashboard, a search button, and the like.

[0128] The user can set the display format in advance, and the life prediction device 5 stores the set display format. The display format includes information such as the model ID of the molding machine 1, the name of the display format, the type of the displayed component, and the data items displayed in the graph. The type of the component is, for example, the reducer 14, the shaft device, etc.

[0129] The data items include operation information data, frequency analysis data, torque analysis data, dual-axis related data, etc. Operation information data include, for example, screw speed, motor current, screw front end pressure, die head pressure, etc. Frequency analysis data include vibration RMS, rotation frequency intensity, gear meshing frequency intensity, bearing fault frequency intensity, etc. Torque analysis data include the average value, maximum value, average value, maximum value, main frequency, etc. of the screw torque. Dual-axis related data include the phase difference between the two screws 11 in the dual-axis extruder, etc.

[0130] When the processing unit 51 receives information input to the display operation unit 121 and the search button is operated, it reads the physical quantity data corresponding to the received information from the collected data DB 52a and creates a time series graph of the read physical quantity data. The time series graph display unit 122 displays a time series graph of the physical quantity data corresponding to the model, period, screw speed, and display format specified by the display operation unit 121.

[0131] Fig.21 1 is a schematic diagram showing an example of a search display screen 103 displayed on a portal website. The search display screen 103 includes a search operation unit 131 for searching a time series graph and a time series graph display unit 132 for displaying the searched time series graph. The search operation unit 131 includes, for example, input items for specifying a model of a molding machine 1 to be searched and displayed, a type of a component, a graph (data item), a channel of a sensor 2, a period, a screw speed, etc., and a search button.

[0132] When the processing unit 51 receives information input to the search operation unit 131 and the search button is operated, it reads the physical quantity data corresponding to the received information from the collected data DB 52a and creates a time series graph of the read physical quantity data. The time series graph display unit 132 displays the searched and created time series graph of the physical quantity data.

[0133] Fig. 22 1 is a schematic diagram showing an example of a monthly report screen 104 displayed on the portal site. The monthly report screen 104 is a screen that displays a monitoring report created by a service provider of the molding machine 1. The monthly report screen 104 has a report search unit 141 for searching reports, a comment display unit 142, and a time series graph display unit 143.

[0134] The report search unit 141 has items for specifying the model of the molding machine 1, the year and month of report issuance, and a search button, etc. The comment display unit 142 displays comments related to the status of the molding machine 1 input by the service provider. The service provider can confirm various time series charts related to the status of the molding machine 1 and components, the remaining life of each component, etc. by accessing the portal website, and create comments on the status of the molding machine 1 based on the viewed information, and send them to the life prediction device 5. The life prediction device 5 creates and stores the comments in correspondence with the device ID and the year and month of report issuance. The time series chart display unit 143 displays a representative time series chart representing the status of the molding machine 1. The service provider of the molding machine 1 can select and specify the time series chart to be displayed in the monthly report. The life prediction device 5 creates and stores the comments in correspondence with the device ID and the year and month of report issuance.

[0135] When receiving information input by the user to the report search unit 141 and the search button is operated, the processing unit 51 reads the comments and physical quantity data corresponding to the received information and creates a time series graph based on the physical quantity data. The time series graph display unit 132 displays the time series graph of the searched and created physical quantity data.

[0136] As described above, according to the life prediction method and the like according to the present embodiment, regardless of the amount of data that can be collected for life prediction of components constituting the molding machine 1 , machine learning for life prediction can be performed to predict the life of the components.

[0137] In addition, the life prediction device 5 can make the image recognition learning model 73 perform machine learning and predict the remaining life of the components constituting the molding machine 1. The life prediction device 5 can make the state quantity estimation mathematical model 81 and the life estimation mathematical model 82 perform machine learning and predict the remaining life of the components constituting the molding machine 1. The life prediction device 5 can make the life estimation function 92 perform machine learning and predict the remaining life of the components constituting the molding machine 1. The life prediction device 5 can combine these life prediction models and perform machine learning based on the acquired learning data to predict the remaining life of the components constituting the molding machine 1.

[0138] The life prediction device 5 collects a plurality of types of physical quantity data, but can perform machine learning on the first to third life prediction models M1 , M2 , and M3 by selecting or combining arbitrary physical quantity data, and can predict the remaining life of components constituting the molding machine 1 .

[0139] The life prediction device 5 can predict the remaining life of a component by combining a life prediction model that predicts the remaining life of a component based on physical quantity data obtained by short-term measurement and a life prediction model that predicts the remaining life of a component based on physical quantity data obtained by long-term measurement.

[0140] The life prediction device 5 is capable of predicting the remaining life of a specified component constituting the molding machine 1 by combining a life prediction model for predicting the remaining life of the specified component based on physical quantity data obtained by measuring physical quantities of a first portion related to the specified component and a life prediction model for predicting the remaining life of the specified component based on physical quantity data obtained by measuring physical quantities of a second portion related to the specified component, to predict the remaining life of the component.

[0141] The life prediction device 5 can use physical quantity data related to the reducer 14, physical quantity data related to the screw 11, physical quantity data related to the motor 13, and physical quantity data related to the mold 12 to predict the remaining life of parts related to the extruder, which includes the reducer 14, the screw 11, the motor 13 and the mold 12 that constitute the molding machine 1.

[0142] The life prediction device 5 can predict the remaining life of the reducer 14 by combining a life prediction model for predicting the remaining life of the reducer 14 based on physical quantity data related to the reducer 14 and a life prediction model for predicting the remaining life of the reducer 14 based on physical quantity data related to the screw 11. That is, when predicting the remaining life of the reducer 14, the accuracy of life prediction can be improved by using different types of physical quantity data or physical quantity data measured at different locations. In addition, by selecting different life prediction models that are respectively applicable to different types of physical quantity data or physical quantity data measured at different locations, and using multiple types of physical quantity data and life prediction models, the accuracy of life prediction can be improved.

[0143] The life prediction device 5 can improve the accuracy of life prediction by calculating the minimum value, average value, median value, or weighted average value of the life of the components of the molding machine 1 estimated using a plurality of life prediction models.

[0144] The life prediction device 5 can store different life prediction models for each of the different molding machines 1 and components. That is, it is possible to store a combination of life prediction models suitable for each of the different molding machines 1 and components. In addition, it is possible to store a life prediction model dedicated to each of the different molding machines 1 and components. The life prediction device 5 can use one or more life prediction models suitable for each of the different molding machines 1 and components to predict the remaining life of the component.

[0145] The life prediction device 5 can display the prediction results of the remaining life of the components constituting the molding machine 1 and a time series graph of the physical quantity data on the portal site.

[0146] The life prediction device 5 can transmit the prediction results of the remaining life of the components constituting the molding machine 1 and the time series graph of the physical quantity data to the user's terminal device 6 a and the service provider's terminal device 6 b .

[0147] It should be noted that, in the present embodiment, an example of predicting the remaining life of a component constituting the molding machine 1 is described, but similarly, it is also possible to configure the system to estimate the degree of abnormality of the component. When the first to third life prediction models M1, M2, and M3 are learned, as long as the degree of abnormality is learned instead of the remaining life, the degree of abnormality of the component can be estimated in the same manner as in the present embodiment.

[0148] In addition, it can be configured to estimate the quality of the molded product produced by the molding machine 1. There are cases where the quality of the molded product becomes abnormal due to a change in the state of the components constituting the molding machine 1. For example, due to wear of the screw parts, the molded product may not be melted. The quality of the molded product is a continuous quantity indicating the degree of quality. When the first to third life prediction models M1, M2, and M3 are learned, if the molded product quality is learned instead of the residual life, the molded product quality can be estimated in the same way as in the present embodiment.

[0149] (Supplementary Note 1)

[0150] A method of inference, in which:

[0151] Acquire physical quantity data related to the status of components constituting the molding machine,

[0152] preparing a plurality of estimation models which estimate the life span or abnormality degree of the component or the quality of the molded product by using different algorithms,

[0153] The lifetime or abnormality degree of the component or the quality of the molded product is estimated based on the acquired physical quantity data using one or more of the estimation models selected from the plurality of estimation models.

[0154] (Supplementary Note 2)

[0155] According to the estimation method described in Supplementary Note 1,

[0156] The plurality of estimation models include a learning model that outputs data indicating the life or abnormality of the component or the quality of the molded product when image data generated based on the physical quantity data is input.

[0157] The life span or abnormality degree of the component or the quality of the molded product is estimated by inputting the image data generated based on the acquired physical quantity data into the learning model.

[0158] (Supplementary Note 3)

[0159] According to the presumption method described in Supplementary Note 1 or Supplementary Note 2,

[0160] The physical quantity data includes physical quantity data representing multiple types of physical quantities,

[0161] The plurality of estimation models include a mathematical model showing a relationship between the plurality of types of physical quantity data, other state quantities other than the plurality of types of physical quantities, and the life or abnormality degree of the component or the quality of the molded product,

[0162] By substituting the acquired physical quantity data into the one or more mathematical models, the life span or abnormality degree of the component or the quality of the molded product is estimated.

[0163] (Supplementary Note 4)

[0164] According to any one of the estimation methods in Supplementary Notes 1 to 3,

[0165] The plurality of estimation models have one or more explanatory variables and include regression equations representing the relationship between the physical quantity data and the life or abnormality degree of the component or the quality of the molded product,

[0166] By substituting the acquired physical quantity data into the regression equation, the life span or abnormality degree of the component or the quality of the molded product is estimated.

[0167] (Supplementary Note 5)

[0168] According to any one of the estimation methods in Supplementary Notes 1 to 4,

[0169] The physical quantity data includes physical quantity data representing multiple types of physical quantities,

[0170] The plurality of inference models include:

[0171] a learning model that, when image data generated based on the physical quantity data is input, outputs data indicating the life or degree of abnormality of the component or the quality of the molded product;

[0172] a mathematical model representing a relationship between the plurality of types of physical quantity data, other state quantities other than the plurality of types of physical quantities, and the life or abnormality degree of the component or the quality of the molded product; and

[0173] A regression equation having one or more explanatory variables, which expresses the relationship between the physical quantity data and the life or abnormality degree of the component or the quality of the molded product,

[0174] The life or abnormality of the component or the quality of the molded product is estimated by inputting the image data generated based on the acquired physical quantity data into the learning model, substituting the acquired physical quantity data into the mathematical model, and substituting the acquired physical quantity data into the regression equation.

[0175] (Supplementary Note 6)

[0176] According to any one of the estimation methods in Supplementary Notes 1 to 5,

[0177] The plurality of inference models include:

[0178] An estimation model for estimating the life or degree of abnormality of the component or the quality of a molded product based on the physical quantity data obtained by short-term measurement; and

[0179] An estimation model for estimating the life or abnormality degree of the component or the quality of the molded product based on the physical quantity data obtained by long-term measurement.

[0180] (Supplementary Note 7)

[0181] According to any one of the estimation methods in Supplementary Notes 1 to 6,

[0182] The plurality of inference models include:

[0183] An estimation model for estimating the life or degree of abnormality of the component or the quality of a molded product based on physical quantity data obtained by measuring a physical quantity of a first portion related to the component; and

[0184] An estimation model for estimating the life or degree of abnormality of the component or the quality of the molded product based on physical quantity data obtained by measuring a physical quantity of a second location related to the component.

[0185] (Supplementary Note 8)

[0186] According to any one of the estimation methods in Supplementary Notes 1 to 7,

[0187] The molding machine is an extruder including a reducer, a screw, a motor and a mold.

[0188] The physical quantity data include:

[0189] physical quantity data related to the reducer, physical quantity data related to the screw, physical quantity data related to the motor, or physical quantity data related to the mold.

[0190] (Supplementary Note 9)

[0191] According to any one of the estimation methods in Supplementary Notes 1 to 8,

[0192] The plurality of inference models include:

[0193] An estimation model for estimating the life or abnormality degree of the reducer or the quality of a molded product based on physical quantity data related to the reducer; and

[0194] An estimation model for estimating the life or abnormality degree of the speed reducer or the quality of a molded product based on physical quantity data related to the screw.

[0195] (Supplementary Note 10)

[0196] According to any one of the estimation methods in Supplementary Notes 1 to 9,

[0197] The minimum value, average value, median value or weighted average value of the life span or abnormality degree of the component or the quality of the molded product estimated using the plurality of estimation models is calculated.

[0198] (Supplementary Note 11)

[0199] According to any one of the estimation methods in Supplementary Notes 1 to 10,

[0200] A device identifier representing each of the plurality of molding machines, a component identifier representing the components constituting the molding machine, and one or more estimation models for estimating the life or abnormality degree of the components or the quality of molded products are stored in a corresponding manner.

[0201] (Supplementary Note 12)

[0202] According to any one of the estimation methods in Supplementary Notes 1 to 11,

[0203] Information indicating the estimated life span or abnormality degree of the component or quality of the molded product can be displayed and provided.

[0204] (Supplementary Note 13)

[0205] According to the estimation method described in any one of Supplementary Notes 1 to 12,

[0206] Information indicating the estimated life span or abnormality degree of the component or quality of the molded product is transmitted to the terminal device.

[0207] Description of Reference Numerals

[0208] 1: Molding machine

[0209] 2: Sensor

[0210] 3: Data collection device

[0211] 4: Router

[0212] 5: Life prediction device

[0213] 6a: Terminal Device

[0214] 6b: Terminal Device

[0215] 10: Cylinder

[0216] 10a: Hopper

[0217] 11: Screw

[0218] 12: Mould

[0219] 13: Motor

[0220] 14: Reducer

[0221] 15: Control device

[0222] 21: First sensor

[0223] 22: Second sensor

[0224] 23: Third sensor

[0225] 24: Fourth sensor

[0226] 31: Control Department

[0227] 32: Storage

[0228] 33: Ministry of Communications

[0229] 34: Data input department

[0230] 50: Recording medium

[0231] 51: Processing Department

[0232] 52: Storage

[0233] 53: Ministry of Communications

[0234] 71: Frequency Analysis Department

[0235] 72: Image generation unit

[0236] 73: Image Recognition Learning Model

[0237] 73a: Input layer

[0238] 73b: Middle layer

[0239] 73c: Output layer

[0240] 81: Mathematical model for state quantity estimation

[0241] 82: Mathematical model for life estimation

[0242] 91: RMS calculation department

[0243] 92: Lifespan estimation function

[0244] 101: Residual life estimation result display screen

[0245] 102: Dashboard screen

[0246] 103: Search display screen

[0247] 104: Monthly Report Screen

[0248] DB52a: Collecting Data

[0249] DB52b: Model composition

[0250] M1: The first life prediction model

[0251] M2: Second life prediction model

[0252] M3: The third life prediction model

[0253] P: Computer program.

Claims

1. A method for inferring, in which: Acquire physical quantity data related to the status of components constituting the molding machine, preparing a plurality of estimation models which estimate the life span or abnormality degree of the component or the quality of the molded product by using different algorithms, The lifetime or abnormality degree of the component or the quality of the molded product is estimated based on the acquired physical quantity data using one or more of the estimation models selected from the plurality of estimation models.

2. The estimation method according to claim 1, wherein: The plurality of estimation models include a learning model that, when image data generated based on the physical quantity data is input, outputs data indicating the life or abnormality of the component or the quality of the molded product. The life span or abnormality degree of the component or the quality of the molded product is estimated by inputting the image data generated based on the acquired physical quantity data into the learning model.

3. The estimation method according to claim 1, wherein: The physical quantity data includes physical quantity data representing multiple types of physical quantities, The plurality of estimation models include a mathematical model that represents a relationship between a plurality of types of the physical quantity data, a plurality of types of state quantities other than the physical quantity, and the life or abnormality degree of the component or the quality of the molded product, By substituting the acquired physical quantity data into one or more mathematical models, the life span or abnormality degree of the component or the quality of the molded product is estimated.

4. The estimation method according to claim 1, wherein: The plurality of estimation models include a regression equation having one or more explanatory variables and representing a relationship between the physical quantity data and the life or abnormality degree of the component or the quality of the molded product, By substituting the acquired physical quantity data into the regression equation, the life span or abnormality degree of the component or the quality of the molded product is estimated.

5. The estimation method according to claim 1, wherein: The physical quantity data includes physical quantity data representing multiple types of physical quantities, The plurality of inference models include: a learning model that, when image data generated based on the physical quantity data is input, outputs data indicating the life or degree of abnormality of the component or the quality of the molded product; a mathematical model representing a relationship between a plurality of types of the physical quantity data, a plurality of types of other state quantities other than the physical quantity, and the life or abnormality degree of the component or the quality of the molded product; and A regression equation having one or more explanatory variables, which expresses the relationship between the physical quantity data and the life or abnormality degree of the component or the quality of the molded product, The life or abnormality of the component or the quality of the molded product is estimated by inputting the image data generated based on the acquired physical quantity data into the learning model, substituting the acquired physical quantity data into the mathematical model, and substituting the acquired physical quantity data into the regression equation.

6. The estimation method according to any one of claims 1 to 5, wherein: The plurality of inference models include: An estimation model for estimating the life or degree of abnormality of the component or the quality of a molded product based on the physical quantity data obtained by short-term measurement; and An estimation model for estimating the life or abnormality degree of the component or the quality of the molded product based on the physical quantity data obtained by long-term measurement.

7. The estimation method according to any one of claims 1 to 5, wherein: The plurality of inference models include: An estimation model for estimating the life or degree of abnormality of the component or the quality of a molded product based on physical quantity data obtained by measuring a physical quantity of a first portion related to the component; and An estimation model for estimating the life or degree of abnormality of the component or the quality of the molded product based on physical quantity data obtained by measuring a physical quantity of a second location related to the component.

8. The estimation method according to any one of claims 1 to 5, wherein: The molding machine is an extruder including a reducer, a screw, a motor and a mold. The physical quantity data includes physical quantity data related to the reducer, physical quantity data related to the screw, physical quantity data related to the motor, or physical quantity data related to the mold.

9. The estimation method according to claim 8, wherein: The plurality of inference models include: An estimation model for estimating the life or abnormality degree of the reducer or the quality of a molded product based on physical quantity data related to the reducer; and An estimation model for estimating the life or abnormality degree of the speed reducer or the quality of a molded product based on physical quantity data related to the screw.

10. The estimation method according to any one of claims 1 to 5, wherein: The minimum value, average value, median value or weighted average value of the life span or abnormality degree of the component or the quality of the molded product estimated using the plurality of estimation models is calculated.

11. The estimation method according to any one of claims 1 to 5, wherein: A device identifier representing each of the plurality of molding machines, a component identifier representing the components constituting the molding machine, and one or more estimation models for estimating the life or abnormality degree of the components or the quality of molded products are stored in a corresponding manner.

12. The estimation method according to any one of claims 1 to 5, wherein: Information indicating the estimated life span or abnormality degree of the component or quality of the molded product can be displayed and provided.

13. The estimation method according to any one of claims 1 to 5, wherein: Information indicating the estimated life span or abnormality degree of the component or quality of the molded product is transmitted to the terminal device.

14. A presumption device, comprising: an acquisition unit that acquires physical quantity data related to the states of components constituting the molding machine; Multiple estimation models, estimating the life or abnormality degree of the component or the quality of the molded product through different algorithms; as well as The processing unit estimates the life or abnormality degree of the component or the quality of the molded product based on the physical quantity data acquired by the acquisition unit using one or more of the estimation models selected from the plurality of estimation models.

15. A computer program for causing a computer to execute the following processing: Acquire physical quantity data related to the status of components constituting the molding machine, The life or abnormality degree of the component or the quality of the molded product is estimated based on the acquired physical quantity data using one or more estimation models selected from a plurality of estimation models that estimate the life or abnormality degree of the component or the quality of the molded product using different algorithms.

Images