Information processing method, information processing apparatus, molding machine system, program product, and storage medium

Abstract

An information processing method for detecting a state of a plurality of manufacturing devices used by a plurality of companies, characterized by acquiring sensor value data obtained by detecting a physical quantity of a manufacturing device, saving the collected sensor value data to a plurality of databases for each company, generating a plurality of learning models by machine learning based on the saved sensor value data, and calculating a state of a manufacturing device of one company by inputting sensor value data acquired from the manufacturing device of the one company to one or more learning models selected by the one company.

Description

Technical Field

[0001] This invention relates to information processing methods, information processing devices, molding machine systems, program products, and storage media. Background Technology

[0002] Patent Document 1 discloses a state judgment device that acquires data from an injection molding machine and uses a learning model to infer abnormalities in the injection molding machine. The state judgment device in Patent Document 1 stores multiple learning models, classifies the acquired injection molding machine data, and inputs the classified data into the corresponding learning models, thereby efficiently and accurately inferring abnormalities in the injection molding machine. For example, the state judgment device classifies the data for each step of the injection molding process to infer abnormalities in each step.

[0003] Existing technical documents

[0004] Patent documents

[0005] Patent Document 1: JP 2020-66178 Summary of the Invention

[0006] The purpose of this disclosure is to provide an information processing method, information processing apparatus, molding machine system, and computer program that can prepare multiple learning models with rich variations and excellent state detection accuracy by collecting sensor value data of manufacturing equipment from multiple enterprises, and can appropriately select and utilize the learning models required or optimized by each enterprise as the user of the information processing apparatus.

[0007] Furthermore, the state determination device disclosed in Patent Document 1 simply stores learning models corresponding to multiple processes, and is not configured to select and utilize a user-desired learning model from a variety of prepared learning models. Additionally, it does not disclose a structure for collecting sufficient sensor value data necessary to generate or update a variety of learning models with excellent state detection accuracy for the manufacturing apparatus.

[0008] The information processing method disclosed herein detects the status of multiple manufacturing devices used by multiple enterprises. The method is characterized by acquiring sensor value data obtained by detecting physical quantities of the multiple manufacturing devices used by the multiple enterprises; saving the sensor value data from the multiple manufacturing devices to multiple databases prepared for each enterprise using the manufacturing device; generating or updating multiple learning models for detecting the status of the manufacturing devices through machine learning based on the sensor value data stored in the multiple databases; storing the enterprise's identification information in a corresponding association with model selection information representing one or more learning models selected by the enterprise; and calculating the status of the enterprise's manufacturing device by inputting sensor value data obtained from the manufacturing device of one enterprise into one or more learning models represented by the model selection information associated with the identification information of that enterprise.

[0009] The information processing apparatus disclosed herein detects the status of multiple manufacturing devices used by multiple enterprises. The apparatus is characterized by comprising: an acquisition unit that acquires sensor value data obtained by detecting physical quantities of the multiple manufacturing devices used by the multiple enterprises; multiple databases that store the sensor value data obtained from the multiple manufacturing devices for each enterprise using the manufacturing devices; and a processing unit that, based on the sensor value data stored in the multiple databases, generates or updates multiple learning models for detecting the status of the manufacturing devices through machine learning, stores the enterprise's identification information in a corresponding association with model selection information representing one or more learning models selected by the enterprise, and calculates the status of the enterprise's manufacturing device by inputting sensor value data obtained from the manufacturing device of one enterprise into one or more learning models represented by the model selection information associated with the identification information of that enterprise.

[0010] The molding machine system disclosed herein includes the aforementioned information processing device and a molding machine, wherein the information processing device detects the status of the molding machine.

[0011] The computer program disclosed herein is used to enable a computer to perform a process of detecting the state of multiple manufacturing devices used by multiple enterprises. The computer program is characterized in that it enables the computer to perform the following processes: acquiring sensor value data obtained by detecting physical quantities of the multiple manufacturing devices used by the multiple enterprises; saving the sensor value data obtained from the multiple manufacturing devices to multiple databases prepared for each enterprise using the manufacturing devices; generating or updating multiple learning models for detecting the state of the manufacturing devices through machine learning based on the sensor value data stored in the multiple databases; storing the enterprise's identification information in a corresponding association with model selection information representing one or more learning models selected by the enterprise; and calculating the state of the enterprise's manufacturing device by inputting sensor value data obtained from the manufacturing device of one enterprise into one or more learning models represented by the model selection information associated with the identification information of that enterprise.

[0012] Invention Effects

[0013] According to this disclosure, by collecting sensor value data of manufacturing equipment from multiple enterprises to prepare multiple learning models with rich variations and excellent state detection accuracy, it is possible to appropriately select and utilize the learning models required or optimized by each enterprise as the user of the information processing equipment. Attached Figure Description

[0014] Figure 1 This is a block diagram showing the overall configuration of the molding machine system of Embodiment 1.

[0015] Figure 2 This is a block diagram showing the detailed configuration of the molding machine system of Embodiment 1.

[0016] Figure 3 This is a schematic diagram showing an example of the configuration of the molding machine according to Embodiment 1.

[0017] Figure 4 This is a block diagram illustrating a configuration example of an edge computer according to Embodiment 1.

[0018] Figure 5 This is a block diagram illustrating an example of the configuration of the information processing apparatus according to Embodiment 1.

[0019] Figure 6 This is a conceptual diagram illustrating an example of the record structure of the enterprise information DB in Implementation 1.

[0020] Figure 7 This is a conceptual diagram illustrating an example of the record structure of the machine information DB in Implementation 1.

[0021] Figure 8This is a conceptual diagram illustrating an example of the record structure of an individual database constituting an individual DB group in Implementation 1.

[0022] Figure 9 This is a concept diagram illustrating an example of model selection information.

[0023] Figure 10 This is a conceptual diagram illustrating an example of a cloud platform provided by an information processing device.

[0024] Figure 11 This is a flowchart illustrating the selection and processing order of the learning model.

[0025] Figure 12 This is a flowchart illustrating the processing sequence of the state detection of the molding machine according to Embodiment 1.

[0026] Figure 13 This is a flowchart illustrating the processing sequence of the state detection of the molding machine according to Embodiment 1.

[0027] Figure 14 This is a conceptual diagram illustrating the methods for generating, updating, and fine-tuning multiple learning models in Implementation 2.

[0028] Figure 15 This is a conceptual diagram illustrating the hierarchical structure of the fine-tuning of the learning model in Implementation 2.

[0029] Figure 16 This is a conceptual diagram illustrating model selection information for Implementation Method 2.

[0030] Figure 17 This is a flowchart illustrating the processing sequence for fine-tuning.

[0031] Figure 18 This is a flowchart illustrating the processing sequence of state detection in Implementation Method 2.

[0032] Figure 19 This is a flowchart illustrating the processing sequence of state detection in Implementation Method 2.

[0033] Figure 20 This is a flowchart illustrating the processing sequence of the proposal for the optimized learning model in Implementation 3.

[0034] Figure 21 This is a flowchart illustrating the processing sequence of the proposal for the optimized learning model in Implementation 3.

[0035] Figure 22 This is a conceptual diagram illustrating an individual DB in implementation 4.

[0036] Figure 23 This is an example of a screen displaying the status detection results of the molding machine in Embodiment 5.

[0037] Figure 24 This is a flowchart showing the category classification of charts and the processing order of labeling.

[0038] Figure 25 This is a flowchart showing the processing sequence as illustrated in the diagram.

[0039] Figure 26 This is a flowchart illustrating the predictive processing sequence as an example of additional analysis.

[0040] Figure 27 This is a schematic diagram showing the selected chart.

[0041] Figure 28 This is a schematic diagram showing the prediction results in a chart.

[0042] Figure 29 This is a flowchart illustrating the analysis process sequence as an example of additional analysis.

[0043] Figure 30 This is an example of a screen displaying the analysis results. Detailed Implementation

[0044] Specific examples of the information processing method, information processing apparatus, molding machine system, and computer program according to embodiments of the present invention will now be described with reference to the accompanying drawings. Furthermore, the present invention is not limited to these examples, and all modifications are intended to be included within the meaning and scope of the claims as shown in them. Additionally, at least some of the embodiments and modifications described below may be combined arbitrarily.

[0045] Figure 1 This is a block diagram showing the overall structure of the molding machine system according to Embodiment 1. Figure 2 This is a block diagram showing the detailed configuration of the molding machine system according to Embodiment 1. The molding machine system is as follows: Figure 1 The diagram shows multiple molding machines 1 used by multiple companies A1, A2, A3... (hereinafter appropriately referred to as Company A), and an information processing device 2 for detecting the status of these multiple molding machines 1. Furthermore, the molding machine system includes... Figure 2 The diagram shows a sensor 5, an edge computer 6, a router 7, and a terminal device 8 for each enterprise A using a molding machine 1. Each enterprise A has one or more molding machines 1, but for simplicity, in embodiment 1, it is described that each enterprise A has one molding machine 1. The information processing device 2 constitutes a cloud computer and includes multiple individual databases 31 (see reference 1) prepared for each enterprise A using the molding machine 1. Figure 6 The individual DB groups 3 consist of a DB group and an AI model group 4 consists of multiple learning models used to detect various states of the molding machine 1.

[0046] Figure 3 This is a schematic diagram showing an example of the configuration of the molding machine 1 according to Embodiment 1. The molding machine 1 is, for example, a twin-shaft compounding extruder, comprising a cylinder 10 with a hopper 10a for feeding resin raw materials, and two screw shafts (rotating shafts) 11. Furthermore, the two screw shafts 11... Figure 3 Arranged along the depth direction of the paper surface, therefore, in Figure 3 Only one screw shaft 11 is depicted. The two screw shafts 11 are arranged approximately parallel to each other in a meshing state and are rotatably inserted into the hole of the cylinder 10, pushing the resin raw material fed into the hopper 10a in the extrusion direction. Figure 2 as well as Figure 3 Transport from the right (center to right) and perform melting and mixing.

[0047] The screw shaft 11 is constructed by combining and integrating various screw components into a single screw shaft 11. For example, the screw shaft 11 is constructed by arranging and combining a threaded rod-shaped positive thread component for conveying resin raw material in the forward direction, a helical screw component for conveying resin raw material in the reverse direction, and a kneading piece for mixing resin raw material in a sequence and position corresponding to the characteristics of the resin raw material.

[0048] The cylinder block 10 is constructed by combining multiple cylinder block castings.

[0049] Additionally, molding machine 1, such as Figure 2 The device shown includes a motor 12 for outputting a driving force to rotate a screw shaft 11, a reducer 13 for reducing and transmitting the driving force of the motor 12, and a control device 14. The screw shaft 11 is connected to the output shaft of the reducer 13. The screw shaft 11 rotates by the driving force of the motor 12 transmitted by the reducer 13.

[0050] Sensor 5 detects physical quantities of the molding machine 1 and outputs sensor value data representing the detected physical quantities to the edge computer 6. The physical quantities of the molding machine 1 include those obtained from the molding machine 1 itself and those obtained from the molded product manufactured by the molding machine 1. These physical quantities include temperature, position, speed, acceleration, current, voltage, pressure, time, image data, torque, force, strain, power consumption, weight, etc. These physical quantities can be measured using thermometers, position sensors, speed sensors, acceleration sensors, ammeters, voltmeters, pressure gauges, timers, cameras, torque sensors, power meters, weight gauges, etc.

[0051] More specifically, the sensors 5 that detect the physical quantities of the action of the molding machine 1 include vibration sensors such as acceleration sensors that detect the vibration of the reducer 13, torque sensors that detect the output torque of the motor 12, torque sensors that detect the shaft torque applied to the screw shaft 11, displacement sensors that detect the displacement of the rotation center of the screw shaft 11, vibration sensors that detect the vibration of the screw shaft 11, thermometers that detect the temperature of the screw shaft 11, resin temperature sensors that detect the resin temperature, resin pressure sensors that detect the resin pressure, outlet temperature sensors that detect the outlet resin temperature, and outlet pressure sensors that detect the outlet resin pressure.

[0052] Sensor 5, which detects the physical quantities of the molded product, includes optical measuring instruments, camera devices, and weight gauges for detecting the size, color, brightness, etc. of the molded product.

[0053] The control device 14 is a computer for controlling the operation of the molding machine 1, and includes a transceiver unit (not shown), a timing unit, and a display unit for transmitting and receiving information with the edge computer 6.

[0054] Specifically, the control device 14 sends the machine ID, date and time, molding machine control parameters, machine configuration data, and basic data for identifying multiple molding machines 1 to the edge computer 6. Molding machine control parameters include, for example, the feeder supply rate (resin raw material supply rate), screw shaft 11 rotation speed, extrusion rate, cylinder temperature, resin pressure, and motor current. Machine configuration data includes information indicating the model of the molding machine 1, screw configuration, cylinder configuration, and mold shape. Basic data represents the physical properties of the resin raw material.

[0055] The control device 14 receives information indicating the status of the molding machine 1 sent from the edge computer 6 and displays the received information. Additionally, the control device 14 uses the information indicating the status of the molding machine 1 to monitor for any abnormalities, outputs warnings as needed, and executes processes to stop the operation of the molding machine 1.

[0056] Figure 4 This is a block diagram illustrating an example configuration of the edge computer 6 in Embodiment 1. The edge computer 6 includes an arithmetic unit 61, a storage unit 62, a communication unit 63, and an input unit 64, all of which are connected to the arithmetic unit 61. The edge computer 6 is a computer with lower hardware specifications compared to the information processing device 2 described later, performing simple anomaly detection processing of the molding machine 1, or processing such as sending and receiving information required for status detection processing in the information processing device 2. Furthermore, the specific circuit configuration and type of computer in the edge computer 6 are not particularly limited.

[0057] The arithmetic unit 61 includes 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, as well as internal storage devices such as ROM (Read Only Memory) and RAM (Random Access Memory) and I / O terminals. The arithmetic unit 61 functions as the edge computer 6 of this embodiment 1 by executing edge application programs (program products) stored in the storage unit 62 (described later). Furthermore, each functional unit of the edge computer 6 can be implemented in software or partially or entirely in hardware.

[0058] The storage unit 62 is a non-volatile memory such as a hard disk, EEPROM (Electrically Erasable Programmable ROM), or flash memory. The storage unit 62 stores edge programs, which are used to enable the arithmetic unit 61 to perform simple fault detection processing of the molding machine 1.

[0059] The communication unit 63 is a communication circuit that transmits and receives information according to communication protocols such as Ethernet (registered trademark). The communication unit 63 is connected to the control device 14 via a first communication network such as LAN, and the arithmetic unit 61 can transmit and receive various types of information between the communication unit 63 and the control device 14.

[0060] The first network is connected to router 4, and the communication unit 63 is connected to the information processing device 2 in the cloud, which serves as the second communication network, via router 7. The computing unit 61 can send and receive various types of information with the information processing device 2 via the communication unit 63 and router 7.

[0061] Input section 64 is an input interface for input signals. Sensor 5 is connected to input section 64, and sensor value data output from sensor 5 is input.

[0062] Figure 5 This is a block diagram illustrating an example configuration of the information processing apparatus 2 according to Embodiment 1. The information processing apparatus 2 is a computer, including a processing unit 21, a storage unit 22, and a communication unit 23. The storage unit 22 and the communication unit 23 are connected to the processing unit 21.

[0063] The processing unit 21 includes a CPU, multi-core CPU, GPU (Graphics Processing Unit), GPGPU (General-purpose computing on graphics processing units), TPU (Tensor Processing Unit), ASIC, FPGA, NPU (Neural Processing Unit), and other arithmetic processing circuits, as well as internal storage devices such as ROM and RAM, and I / O terminals. The processing unit 21 functions as the information processing apparatus 2 of this embodiment 1 by executing the computer program (program product) P stored in the storage unit 22 (described later). Furthermore, each functional unit of the information processing apparatus 2 can be implemented in software or partially or entirely in hardware.

[0064] Storage unit 22 is a non-volatile memory such as hard disk, EEPROM, or flash memory. Storage unit 22 stores computer program P for enabling the computer to perform processing to detect the state of molding machine 1, enterprise information DB22a, machine information DB22b, individual DB group 3, and AI model group 4.

[0065] The computer program P can be in a form that can be readable by a computer and recorded in the storage medium 20. The storage unit 22 stores the computer program P read from the storage medium 20 by a reading device (not shown). The storage medium 20 is a semiconductor memory such as flash memory. Alternatively, the storage medium 20 can also be an optical disc such as a CD (Compact Disc)-ROM, DVD (Digital Versatile Disc)-ROM, or BD (Blu-ray Disc). Furthermore, the storage medium 20 can also be a disk such as a floppy disk or a hard disk, or a magneto-optical disk. In addition, the computer program P can be downloaded from an external server (not shown) connected to a communication network (not shown) and stored in the storage unit 22.

[0066] The communication unit 23 is a communication circuit that transmits and receives information according to communication protocols such as Ethernet (registered trademark). The communication unit 23 is connected to the edge computer 6 and the terminal device 8 via a second communication network, and the processing unit 21 can transmit and receive various types of information between the communication unit 23 and the edge computer 6 and the terminal device 8.

[0067] Figure 6This is a conceptual diagram illustrating an example of the record structure of the enterprise information DB22a in Implementation 1. The enterprise information DB22a includes a hard disk and a DBMS (Database Management System) to store information about enterprise A, which is the user. For example, the table of the enterprise information DB22a has a "Enterprise ID" column, an "Authentication Information" column, an "Associated Enterprise ID" column, and a "Model Selection Information" column.

[0068] The "Enterprise ID" column stores the inherent identification information of each of the multiple Enterprise A's. The "Authentication Information" column stores the information used to authenticate each Enterprise A. The "Associated Enterprise ID" column stores the IDs of other Enterprise A's associated enterprises that share the information required for the status detection of molding machine 1. The "Model Selection Information" column stores the learning model selected by Enterprise A from among multiple learning models; that is, information indicating the learning model used by Enterprise A (hereinafter referred to as model selection information). The model selection information will be explained in detail later.

[0069] Figure 7 This is a conceptual diagram illustrating an example of the record structure of the machine information DB22b according to Embodiment 1. The machine information DB22b stores information about the molding machine 1. For example, the table of the machine information DB22b has a "Machine ID" column, a "Company ID" column, a "Factory ID" column, and an "Edge ID" column.

[0070] The "Machine ID" column stores the unique identification information for each of the multiple molding machines 1. The "Enterprise ID" column stores the ID of the enterprise A that uses the molding machine 1. The "Factory ID" column stores the ID representing the factory that owns the molding machine 1. The "Edge ID" column stores the ID used to identify the edge computer 6 connected to the molding machine 1.

[0071] Figure 8 This is a conceptual diagram illustrating an example of the record structure of an individual database 31 constituting an individual DB group 3 in Embodiment 1. The individual DBs store information detected by the sensor 5. For example, the tables of the individual DBs have columns for "Machine ID", "Date and Time", "Molding Machine Control Parameters", "Machine Configuration Data", "Sensor Data", "Basic Data", and "Status Data".

[0072] The "Machine ID" column stores the ID of the molding machine 1, which is the object being detected. The "Date and Time" column stores the date and time when the physical quantity of the molding machine 1 was detected. The "Molding Machine Control Parameters" column stores parameters used to control the molding machine 1, such as the screw rotation speed, resin feed rate or extrusion rate, cylinder temperature, etc. The "Machine Configuration Data" column contains data indicating the configuration of the molding machine 1, such as the model of the molding machine 1, the screw configuration, the cylinder 10 configuration, mold shape, etc. The "Sensor Data" column stores sensor values ​​obtained from the sensor 5 installed on the molding machine 1, such as the vibration of the reducer 13, shaft torque, motor torque, shaft vibration, resin temperature, resin pressure, outlet resin temperature, outlet resin pressure, and image data obtained from capturing the resin line, etc. The "Basic Data" column stores data indicating the resin material, such as the physical properties of the resin, etc.

[0073] The "Status Data" column stores data indicating the state of the molding machine 1 when the sensor value is obtained, provided that the state of the molding machine 1 can be confirmed. This data may include information such as whether various parts of the molding machine 1, such as the reducer 13 and the screw, are functioning correctly, and whether the molding machine 1 is operating normally. The "Status Data" column can also store specific abnormalities indicating the molding machine 1 and its operating state.

[0074] AI model group 4 includes multiple learning models for detecting various states of molding machine 1. The states of molding machine 1 detected by the molding machine system include, for example, abnormal states of various parts and operating states of molding machine 1. Abnormalities of molding machine 1 include, for example, abnormal vibration of reducer 13, overload, abnormal dimensions of molded products, scratches, cracks, wear, corrosion of screw shaft 11, performance degradation or abnormal oil due to wear of reducer 13 and its parts, inefficient operating states (increased energy consumption, etc.), operating states that lead to quality abnormalities (forced start-up, etc.), abnormalities of strands (dimensions, foreign objects, color, winding (twisting), etc.), and abnormal resin viscosity (quality), etc.

[0075] In this embodiment, the information processing device 2 uses AI model group 4 to detect the presence of abnormal vibrations in the reducer 13, scratches, cracks, wear, corrosion on the screw shaft 11, performance degradation of the reducer 13, abnormal oil conditions, inefficient operating conditions, and operating conditions that lead to quality abnormalities. The information processing device 2 then performs processing to determine the faulty part of the reducer 13 (e.g., bearings, gears, etc.).

[0076] Specifically, AI model group 4 includes a learning model for detecting abnormal vibration of reducer 13, a learning model for detecting scratches, cracks, wear, corrosion, etc. of screw, a learning model for detecting performance degradation or oil abnormality of reducer 13, a learning model for detecting faulty parts of reducer 13, a learning model for detecting inefficient operating conditions, and a learning model for detecting abnormal operating conditions that lead to quality abnormalities.

[0077] The learning models that make up AI model group 4 are models with feature extraction layers of CNN (Convolutional Neural Network), such as single classification models, U-Net, RNN (Recurrent Neural Network) and other neural networks, other machine learning models such as SVM (Support Vector Machine), Bayesian networks, or regression trees.

[0078] When the sensor value data is time series data, the information processing device 2 only needs to generate a time series data image that represents the sensor value data in an image.

[0079] The learning models for detecting screw wear, inefficient operating conditions, and abnormal operating conditions are convolutional neural network models, such as single-classification models, that extract features from time-series data images and output extracted feature values ​​when input. The processing unit 21 can detect the presence or absence of abnormalities, inefficient operating conditions, and abnormal operating conditions of the screw shaft 11 by comparing the feature values ​​under normal conditions with the feature values ​​of the state detection object. The feature values ​​under normal conditions are obtained by inputting sensor value data during normal operation of the molding machine 1 into the aforementioned learning model, which corresponds to or is similar to the molding machine control parameters, machine configuration data, and basic data of the state detection object obtained from the molding machine 1.

[0080] Furthermore, while a single-classification model was explained, for example, given the accumulation of sensor value data under normal conditions and sensor value data under various abnormal conditions, the state of molding machine 1 can be used as teacher data to create a learning dataset with labels added to the sensor value data or time series data images, which can then be used to train a learning model composed of CNNs or similar components.

[0081] The learning model has an input layer, intermediate layers, and an output layer. The intermediate layers contain multiple convolutional and pooling layers that extract image features. The output layer has multiple nodes corresponding to multiple states of the molding machine 1, outputting the confidence level of being in that state. When the learning model is given sensor value data or a time-series image of the sensor value data as input, the weight coefficients of the intermediate layers are optimized in a way that makes the state of the molding machine 1 output from the learning model approximate the state shown in the teacher data. These weight coefficients are, for example, the weights (association coefficients) between neurons. The method for parameter optimization is not particularly limited; various parameter optimization methods can be used, such as steepest descent, backpropagation, etc.

[0082] Based on the learning model generated in this way, the state of the molding machine 1 can be diagnosed by inputting a time-series data image of sensor value data obtained from the molding machine 1, the object of diagnosis, into the learning model. For example, the processing unit 21 determines that the molding machine 1 is in a state corresponding to the node with the highest output confidence.

[0083] Alternatively, the learning model can be configured to output more optimized forming conditions or adjustments to the forming conditions when a time-series data image is input to it. The information processing device 2 sends the forming conditions output from the learning model to the edge computer 6, and the edge computer 6 sends the received forming conditions to the control device 14.

[0084] The learning model for detecting abnormal vibrations in the reducer 13 includes an RNN that outputs data indicating whether the reducer 13 is abnormal, given sensor value data obtained from the vibration sensor as input. Alternatively, it can be configured to use the same learning model as the single-classification model described above, and employ the same method to detect abnormalities in the reducer 13, relative to sensor value data or time-series data images obtained from the vibration sensor.

[0085] As a learning model for detecting performance degradation or oil abnormalities in the reducer 13, SVM or similar methods can be used. Alternatively, it can be configured to use the same learning model as the single-classification model described above, and the same method can be used to detect the presence or absence of performance degradation or oil abnormalities in the reducer 13.

[0086] As a learning model for detecting fault locations in the reducer 13, a multi-class classification model employing a neural network is sufficient.

[0087] On the other hand, the edge computer 6 detects overload, abnormal dimensions of the molded product, abnormal strands, abnormal resin viscosity, etc., based on sensor value data. Specifically, the edge computer 6 calculates the average, variance, root mean square, and other statistical quantities of the sensor value, and detects overload, abnormal resin viscosity, etc. of the reducer 13 based on the processing of the load based on the calculated statistical quantities.

[0088] In addition, the edge computer 6 detects dimensional abnormalities and thread abnormalities in the molded product by calculating the difference between the image of the molded product and the image of a normal molded product.

[0089] Furthermore, the division of the anomaly detection processing performed by the edge computer 6 and the state detection processing performed by the information processing device 2 is an example. As long as it can be processed by the edge computer 6, it can also be configured to be executed on the edge computer 6 side.

[0090] Figure 9 This is a concept diagram illustrating an example of model selection information. Figure 9 This illustrates the concept of model selection information, not the database structure. Model selection information is stored by associating a company ID with information showing whether the company A represented by that company ID uses or does not use a particular learning model. Figure 9 In the diagram, a checkmark (√) indicates that the learning model is used, while an empty column indicates that the learning model is not used. Each company A, in order to detect the status of its molding machine 1, can freely choose which of the multiple learning models to use.

[0091] For example, the model selection information for Company A1 indicates that Company A1 utilizes all learning models. Furthermore, the model selection information for Company A2 indicates that Company A2 utilizes learning models for detecting screw wear, learning models for detecting deterioration of reducer 13, and learning models for detecting abnormal operating conditions of molding machine 1. Moreover, the model selection information for Company A3 indicates that it utilizes learning models for detecting abnormal vibration of reducer 13, learning models for detecting screw wear, learning models for detecting deterioration of reducer 13, and learning models for detecting faulty parts of reducer 13.

[0092] Figure 10 This is a conceptual diagram illustrating an example of a cloud platform provided by information processing device 2. Information processing device 2, as... Figure 10 The sensor value data of molding machine 1 collected from each enterprise A is stored in the individual database 31 of each enterprise A. Each enterprise A is configured to only be able to access its own individual database 31 and refer to the sensor value data, and cannot access the individual databases 31 of other enterprises A.

[0093] Information processing device 2 generates AI model group 4 using sensor value data accumulated in individual databases 31 of each company A. Furthermore, information processing device 2 updates AI model group 4 using sensor value data accumulated in individual databases 31 of each company A. By utilizing sensor value data collected from multiple companies A, multiple learning models with rich variations can be generated. Each company A can freely choose a desired learning model or the learning model optimized for that company A from the multiple learning models available.

[0094] Information processing device 2 sends sensor value data of molding machine 1 from enterprise A. Upon a state detection request, it inputs the sensor value data into one or more learning models to detect the state of molding machine 1 and sends the state detection result back to enterprise A. Enterprise A is restricted from accessing AI model group 4 in a manner that prevents it from referencing it, and can only obtain the state detection results obtained using AI model group 4.

[0095] Figure 11 This is a flowchart illustrating the processing sequence for selecting a learning model. The information processing device 2 performs authentication processing for the user of the control device 14 or terminal device 8. If authentication is successful, the information processing device 2 provides a model selection screen to the control device 14 or terminal device 8, and performs the selection process for the learning model used by the user, company A, via the control device 14 or terminal device 8 in the manner described below.

[0096] The control device 14 or terminal device 8, which is certified as normal, displays a model selection screen (step S1) and accepts the selection of the learning model to be used by enterprise A (step S2). For example, enterprise A1, as the user, can select one or more learning models to be used by enterprise A1 from a pool of available learning models. The information processing device 2 is processed by the one or more learning models selected by enterprise A1.

[0097] The control device 14 or the terminal device 8 sends the model selection information, indicating whether the learning model is used or not, received in step S2, along with the enterprise ID, to the information processing device 2 (step S3).

[0098] The information processing device 2 receives model selection information and enterprise ID sent from the control device 14 or the terminal device 8 (step S4), and stores the received model selection information and enterprise ID in the enterprise information DB22a (step S5).

[0099] Figure 12 as well as Figure 13This is a flowchart illustrating the processing sequence of the state detection of the molding machine 1 in Embodiment 1. The control device 14 of the molding machine 1 used by each company A sends the machine ID, molding machine control parameters indicating the state of operation control of the molding machine 1, machine configuration data, and basic data to the edge computer 6 (step S11).

[0100] The arithmetic unit 61 of the edge computer 6 receives the machine ID, molding machine control parameters, machine configuration data and basic data sent from the control device 14 via the communication unit 63 (step S12).

[0101] The calculation unit 61 acquires time-series sensor value data output from sensors 5 of molding machines 1 used by multiple companies A (step S13). Furthermore, the calculation unit 61 calculates statistical measures such as the mean, variance, and root mean square of the sensor values ​​based on the sensor value data (step S14), and performs simple anomaly detection processing on molding machine 1 based on the calculated statistical measures (step S15).

[0102] Next, the arithmetic unit 61 uses the communication unit 63 to send the judgment result indicating whether the molding machine 1 has any abnormality, as well as the statistical quantity calculated in step S14, to the control device 14 (step S16).

[0103] The control device 14 receives the judgment result and statistics sent from the edge computer 6 (step S17), and monitors the operation of the molding machine 1 based on the received judgment result or statistics (step S18). For example, if the judgment result indicates a specified abnormality, the control device 14 stops the operation control of the molding machine 1.

[0104] Next, the control device 14 displays the received judgment results and statistics on the display unit (step S19).

[0105] Next, the processing of the edge computer 6, which sent the judgment result and statistics, will be explained. After the processing in step S16 is completed, the arithmetic unit 61 of the edge computer 6 uses the communication unit 63 to send the machine ID, molding machine control parameters, machine configuration data and basic data received in step S12, and the sensor value data obtained in step S13 to the information processing device 2 (step S20).

[0106] Here, the processing unit 61 of the edge computer 6 is preferably configured to not only send the sensor value data required for the learning model used by company A of the molding machine 1 using the edge computer 6 to the information processing device 2, but also send other sensor value data that can be detected to the information processing device 2. This is because the other sensor value data can be used to generate or update the learning model used by other companies A. By sending and providing various sensor value data to the information processing device 2 by each company A, a more diverse AI model group 4 can be generated.

[0107] Information processing device 2 receives machine ID, molding machine control parameters, machine configuration data, basic data, and sensor value data sent from edge computer 6 (step S21). Processing unit 21 stores molding machine control parameters, machine configuration data, basic data, and sensor value data in individual database 31 of enterprise A represented by enterprise ID corresponding to the received machine ID, and stores machine configuration data, basic data, and sensor value data (step S22). Next, information processing device 2 selects the learning model used by enterprise A by referring to model selection information (step S23). Furthermore, information processing device 2 detects the status of molding machine 1 by inputting sensor value data, basic data, and sensor value data into the selected learning model (step S24). Next, processing unit 21 sends the detection result to edge computer 6 via communication unit 23 (step S25).

[0108] The arithmetic unit 61 of the edge computer 6 receives the detection result sent from the information processing device 2 (step S26), and sends the received detection result to the control device 14 using the communication unit 63 (step S27).

[0109] The control device 14 receives the detection results sent from the edge computer 6 (step S28) and displays the received detection results on the display unit (step S29).

[0110] Furthermore, the edge computer 6 can send the anomaly judgment result based on the calculation unit 61, the statistical values ​​of the calculated sensor data, and the detection results based on the processing unit 21 to the terminal device 8 upon request from the terminal device 8. The terminal device 8 displays the received anomaly judgment result, statistical values, and detection results.

[0111] On the other hand, when the specified conditions are met, such as accumulating a specified amount of sensor value data in individual DB groups 3, the processing unit 21 of the information processing device 2 reads the sensor value data stored in the individual DBs of multiple companies A, and uses the read sensor value data to generate or update the learning model (step S30).

[0112] When the learning model is a single-classification model, the processing unit 21 of the information processing device 2 inputs a time-series data image obtained by imagerizing the sensor value data accumulated in individual DB groups 3 during normal operation, and an arbitrary reference image into the learning model before learning or updating. The learning model can learn by focusing on features where the local density of the output time-series data image increases and the recognizability with the feature values ​​of the reference image increases. This learning process can generate or update the learning model.

[0113] When the learning model is a classification model, the processing unit 21 of the information processing device 2 inputs a time-series data image obtained by visualizing the sensor value data accumulated in individual DB groups 3 to the learning model before learning or updating. Various parameters of the learning model are optimized using methods such as steepest descent and backpropagation to minimize the difference between the output state detection result and the teacher data. The teacher data corresponds to the sensor value data and represents the state of the molding machine 1 when the sensor value data is obtained. The learning model can be generated or updated through this learning process.

[0114] As explained above, according to the information processing method of Embodiment 1, sensor value data of molding machine 1 can be collected from multiple companies A. By collecting sensor value data from a majority of companies A, multiple learning models can be generated or updated, which can use the sensor value data to detect various changes in the state of molding machine 1.

[0115] Moreover, it is possible to appropriately select and utilize the learning models or optimized learning models required by each enterprise A as the user of the information processing device 2.

[0116] (Variation Example 1)

[0117] The information processing device 2 can also be configured to perform billing processing on state detection processing that uses learning models. For example, the information processing device 2 only needs to calculate the state detection fee based on the number of molding machines 1, the number of sensors 5, the type of sensors 5, the amount of data of the state detection processing object, and the number of learning models used, and then store the calculated diagnostic fee, the user ID of the anomaly diagnosis system, and the identification ID of the molding machine 1 in the storage unit 22 in a corresponding association using date and time, data volume, and diagnostic items. The service provision and billing method for state detection processing can also be in the form of a subscription.

[0118] Alternatively, the cost of using the learning model can be adjusted to correspond to the amount of sensor data provided by Company A. Alternatively, a portion of the cost of using the learning model can be made free for Company A, which provides a specified amount of sensor data.

[0119] (Variation Example 2)

[0120] There is no particular time limit for updating the learning model, but it can be updated regularly, such as on a monthly basis. Alternatively, the learning model can be updated when new abnormal patterns are obtained, when the state detection accuracy deviates from the actual value, or when the state detection accuracy decreases.

[0121] The updating of the learning model can be configured to be executed automatically or to be initiated by user instruction.

[0122] (Implementation Method 2)

[0123] The differences between the information processing method, information processing device 2, molding machine system, and computer program P in Embodiment 2 and Embodiment 1 mainly lie in the composition of the AI ​​model group 4, the model selection information, and the processing order. Other components of the molding machine system are the same as those in Embodiment 1; therefore, the same reference numerals are used to label the same parts, and detailed descriptions are omitted.

[0124] Figure 14 This is a conceptual diagram illustrating the methods for generating, updating, and fine-tuning multiple learning models according to Embodiment 2. The individual database groups 3 of Embodiment 2 include multiple individual databases 31 prepared for each enterprise A, and a testing machine database 30. The testing machine database 30 stores sensor value data, molding machine control parameters, machine configuration data, basic data, etc., obtained by operating the molding machine 1, which serves as the testing machine. In the case of the testing machine, the molding machine 1, which simulates an abnormal state, is prepared to be used. The sensor value data obtained from the molding machine 1, along with information and state data representing the abnormal state, can be stored in the testing machine database 30 in a corresponding association. Alternatively, sensor value data can be calculated using a simulation that reproduces the operation of the molding machine 1, and the calculated sensor value data can be stored in the testing machine database 30.

[0125] The AI ​​model group 4 of implementation method 2 includes a first general AI group 41, a second general AI group 42, and an individual fine-tuning AI group 43.

[0126] In Embodiment 2, the information processing device 2 uses sensor value data stored in the testing machine database 30 to generate a first general AI group 41 for detecting the state of the molding machine 1. Similar to Embodiment 1, the first general AI group 41 includes a first general learning model for detecting abnormal vibration of the reducer 13, a first general learning model for detecting scratches, cracks, wear, corrosion, etc. on the screw, a first general learning model for detecting performance degradation or oil abnormalities in the reducer 13, a first general learning model for detecting faulty parts of the reducer 13, a first general learning model for detecting inefficient operating states, and a first general learning model for detecting abnormal operating states that lead to quality abnormalities.

[0127] Alternatively, it can be configured such that, in the event of insufficient sensor value data stored in the testing machine database 30, a second general learning model, as described later, is generated instead of a first general learning model.

[0128] In addition, the information processing device 2 uses the testing machine database 30 and sensor value data stored in the individual databases 31 of each company A to generate a second general AI group 42 for detecting the state of the molding machine 1. Similar to Embodiment 1, the second general AI group 42 includes a second general learning model for detecting abnormal vibration of the reducer 13, a second general learning model for detecting scratches, cracks, wear, corrosion, etc. of the screw, a second general learning model for detecting performance degradation or oil abnormalities of the reducer 13, a second general learning model for detecting faulty parts of the reducer 13, a second general learning model for detecting inefficient operating states, and a second general learning model for detecting abnormal operating states that cause quality abnormalities.

[0129] In addition, the second general AI group 42 is generally able to detect the state of the molding machine 1 with higher accuracy compared to the first general AI group 41.

[0130] Furthermore, the second general AI group 42 exhibits more variations compared to the first general AI group 41. The information processing device 2 can utilize a large amount of sensor value data collected from various companies A, thus enabling the generation or updating of more second general learning models and the detection of a wider variety of states. In other words, it can generate learning models that would be impossible to generate using only the sensor value data from the testing machine database 30.

[0131] Alternatively, the information processing device 2 can also generate a second general learning model by fine-tuning the first general learning model using sensor value data stored in an individual database 31.

[0132] Furthermore, by using sensor value data stored in individual databases 31 of each enterprise A, the information processing device 2 can individually fine-tune the second general learning model, thereby obtaining individually fine-tuned AI groups 43. The unit for fine-tuning is as follows: Figure 14 The diagram shows a hierarchical structure.

[0133] Sensor data is generally stored in individual databases 31 of each company A, and each company A makes its own fine-tuning. However, in the case of related companies, it is also possible to use sensor data from multiple companies A that constitute the related company to generate a learning model shared by the related companies.

[0134] Figure 15 This is a conceptual diagram illustrating the hierarchical structure of the fine-tuning of the learning model in Implementation Method 2. Furthermore, in Figure 15 In this context, the Edge Computer 6 is simply referred to as the Edge.

[0135] Information processing device 2 fine-tunes the second general learning model, for example, based on sensor value data stored in individual databases 31 of multiple companies A constituting related companies, thereby optimizing the second general learning model for molding machine 1 of related companies.

[0136] The information processing device 2 fine-tunes the second general learning model based on sensor value data and other data stored in the individual databases 31 of each company A, thereby optimizing the second general learning model for the molding machine 1 of each company A. If a learning model has been fine-tuned for related companies, it can also be fine-tuned based on sensor value data and other data stored in the individual databases 31 of each company A.

[0137] The information processing device 2 fine-tunes the second general learning model based on sensor value data obtained from the molding machines 1 in each factory, thereby optimizing the second general learning model for use with the molding machines 1 in each factory. Furthermore, it can also fine-tune a learning model that has been fine-tuned for companies such as A, based on sensor value data obtained from the molding machines 1 in each factory.

[0138] The information processing device 2 fine-tunes the second general learning model based on sensor value data obtained from the molding machine 1 connected to the specific edge computer 6, thereby optimizing the second general learning model for use with the molding machine 1 connected to the edge computer 6. Furthermore, it can also fine-tune a learning model that has been fine-tuned for use in a factory based on sensor value data obtained from the molding machine 1 connected to the edge computer 6.

[0139] The information processing device 2 fine-tunes the second general learning model based on sensor value data obtained from each molding machine 1, thereby optimizing the second general learning model for each molding machine 1. Furthermore, it can also fine-tune the learning model, which has been fine-tuned for use in enterprise A, for the factory, and for the molding machine 1 connected to the edge computer 6, based on sensor value data obtained from molding machines 1 in each factory.

[0140] The information processing device 2 fine-tunes the second general learning model based on sensor value data obtained from multiple molding machines 1 of the same model using the same raw materials and stored in individual databases 31 of each enterprise A. This allows the second general learning model to be optimized on a process-by-process basis. Furthermore, the aforementioned sensor value data for specific processes can also be used to fine-tune the learning model, which is then used for fine-tuning the model for the associated enterprise, for enterprise A, for the factory, and for the molding machine 1 connected to the edge computer 6.

[0141] Figure 16 This is a conceptual diagram illustrating model selection information in Implementation Method 2. Each company A can select the level of fine-tuning based on the type of state of the molding machine 1, which is the object of inspection. The information processing device 2 stores the level of fine-tuning selected by each company A as model selection information. Figure 16 This conceptually illustrates the model selection information for a specific company, A. Each column represents the type of learning model, and each row represents the level of fine-tuning.

[0142] exist Figure 16 In the diagram, the checkmark (√) indicates that the learning model has been fine-tuned using an appropriate hierarchy. For example, a first general learning model is shown as a learning model for detecting abnormal vibrations in the reducer 13, without fine-tuning. A second general learning model is also shown as a learning model for detecting wear on the screw, etc., without fine-tuning.

[0143] The learning model for detecting the degradation of reducer 13 shows fine-tuning on a per-related enterprise basis.

[0144] The learning model for detecting fault locations in reducer 13 shows fine-tuning at the enterprise, factory, edge, and machine levels. The learning model for detecting inefficient operating conditions shows fine-tuning at both the enterprise and edge levels. Furthermore, since sufficient sensor data may not be available for fine-tuning at all machines and factories, there are cases where learning models fine-tuned at multiple levels coexist. For example, factory 1 might use a learning model fine-tuned for that factory, while factory 2, due to insufficient sensor data, might use a learning model fine-tuned for the entire company.

[0145] The learning model for detecting abnormal operating conditions demonstrates fine-tuning on a process-by-process basis.

[0146] Figure 17 This is a flowchart illustrating the fine-tuning process. The processing unit 21 of the information processing device 2 performs fine-tuning of the learning models for each company A at predetermined time intervals. The processing unit 21 reads the model selection information for a company A and performs fine-tuning on each learning model of multiple types up to the appropriate level. The following describes the process of fine-tuning a learning model of one type for a company A. The processing content for fine-tuning other learning models is the same.

[0147] The processing unit 21 refers to the model selection information and determines whether to perform fine-tuning on a per-affiliated-enterprise basis (step S211). If it is determined that fine-tuning should be performed on a per-affiliated-enterprise basis (step S211: Yes), the processing unit 21 reads sensor value data, etc., from the individual database 31 of each enterprise A constituting the affiliated enterprises, and performs fine-tuning on the second general learning model based on the read sensor value data (step S212).

[0148] If the processing in step S212 has ended, or if it is determined that fine-tuning will not be performed on a company-by-company basis (step S211: No), the processing unit 21 determines whether to perform fine-tuning on a company-by-company basis by referring to the model selection information (step S213). If it is determined that fine-tuning will be performed on a company-by-company basis (step S213: Yes), the processing unit 21 reads sensor value data, etc., from the individual database 31 of company A, and fine-tunes the second general learning model based on the read sensor value data (step S214). Furthermore, the processing unit 21 may also be configured to fine-tune the learning model if a learning model fine-tuned in step S212 already exists.

[0149] If the processing in step S214 is completed, or if it is determined that no fine-tuning was performed on a company-by-company basis (step S213: No), the processing unit 21 determines whether to perform fine-tuning on a factory-by-factory basis, referring to the model selection information (step S215). If it is determined that fine-tuning was performed on a factory-by-factory basis (step S215: Yes), the processing unit 21 reads sensor value data, etc., from the individual database 31 of company A, and performs fine-tuning on the second general learning model on a factory-by-factory basis based on the read sensor value data (step S216). Furthermore, the processing unit 21 may also be configured to perform fine-tuning on the learning model if a learning model fine-tuned in step S212 or step S214 already exists.

[0150] If the processing in step S216 is completed, or if it is determined that no fine-tuning has been performed on a factory-by-factory basis (step S215: No), the processing unit 21 determines whether to perform fine-tuning on an edge-by-edge basis based on the model selection information (step S217). If it is determined that fine-tuning is performed on an edge-by-edge basis (step S217: Yes), the processing unit 21 reads sensor value data, etc., of the molding machine 1 connected to each edge computer 6 from the individual database 31 of the enterprise A, and performs fine-tuning on the second general learning model on an edge-by-edge basis based on the read sensor value data (step S218). Furthermore, the processing unit 21 may also be configured to perform fine-tuning on the learning model if a learning model fine-tuned in step S212, step S214, or step S216 already exists.

[0151] If the processing in step S218 has ended, or if it is determined that no fine-tuning was performed on an edge-by-edge basis (step S217: No), the processing unit 21 determines, based on the model selection information, whether to perform fine-tuning on a machine-by-machine basis (step S219). If it is determined that fine-tuning was performed on a machine-by-machine basis (step S219: Yes), the processing unit 21 reads sensor value data, etc., of each molding machine 1 from the individual database 31 of company A, and performs fine-tuning of the second general learning model on a molding machine-by-machine basis based on the read sensor value data (step S220). Furthermore, the processing unit 21 may also be configured to perform fine-tuning on the learning model if a learning model fine-tuned in steps S212, S214, S216, or S218 already exists.

[0152] If the processing in step S220 is completed, or if it is determined that no fine-tuning was performed on a machine-by-machine basis (step S219: No), the processing unit 21 determines whether to perform fine-tuning on a process-by-process basis (step S221) by referring to the model selection information. If it is determined that fine-tuning was performed on a process-by-process basis (step S221: Yes), the processing unit 21 reads sensor value data of molding machine 1 obtained from the individual database 31 of each company A, which is based on a specific machine configuration, the same process, that is, molding machine 1 using the same raw materials and of the same model, and fine-tunes the second general learning model on a process-by-process basis based on the read sensor value data (step S222). In addition, the processing unit 21 may also be configured to fine-tune the learning model if a learning model fine-tuned in step S212, step S214, step S216, step S218 or step S220 exists.

[0153] Figure 18 as well as Figure 19This is a flowchart illustrating the processing sequence of state detection in Embodiment 2. It explains how the processing unit 21 of the information processing device 2 uses a learning model to detect the state of a molding machine 1. The processing content is the same for other molding machines 1 and other types of state detection.

[0154] The processing unit 21 of the information processing device 2 refers to the model selection information and determines whether a learning model fine-tuned on a process-by-process basis exists (step S231). If it is determined that a learning model fine-tuned on a process-by-process basis exists (step S231: Yes), the processing unit 21 uses the learning model fine-tuned on a process-by-process basis to detect the state of the molding machine 1 (step S232) and ends the processing.

[0155] If it is determined that there is no learning model fine-tuned on a process-by-process basis (step S231: No), the processing unit 21 refers to the model selection information to determine whether there is a learning model fine-tuned on a machine-by-machine basis (step S233). If it is determined that there is a learning model fine-tuned on a machine-by-machine basis (step S233: Yes), the processing unit 21 uses the learning model fine-tuned on a machine-by-machine basis to detect the status of the molding machine 1 (step S234) and ends the processing.

[0156] If it is determined that there is no learning model fine-tuned on a machine-by-machine basis (step S233: No), the processing unit 21 refers to the model selection information to determine whether there is a learning model fine-tuned on an edge-by-edge basis (step S235). If it is determined that there is a learning model fine-tuned on an edge-by-edge basis (step S235: Yes), the processing unit 21 uses the learning model fine-tuned on an edge-by-edge basis to detect the state of the molding machine 1 (step S236) and ends the processing.

[0157] If it is determined that there is no learning model fine-tuned on an edge-by-edge basis (step S235: No), the processing unit 21 refers to the model selection information to determine whether there is a learning model fine-tuned on a factory-by-factory basis (step S237). If it is determined that there is a learning model fine-tuned on a factory-by-factory basis (step S237: Yes), the processing unit 21 uses the learning model fine-tuned on a factory-by-factory basis to detect the status of the molding machine 1 (step S238) and ends the processing.

[0158] If it is determined that there is no learning model fine-tuned on a factory basis (step S237: No), the processing unit 21 refers to the model selection information to determine whether there is a learning model fine-tuned on an enterprise basis (step S239). If it is determined that there is a learning model fine-tuned on an enterprise basis (step S239: Yes), the processing unit 21 uses the learning model fine-tuned on an enterprise basis to detect the status of the molding machine 1 (step S240) and ends the processing.

[0159] If it is determined that there is no learning model fine-tuned on a company-by-company basis (step S239: No), the processing unit 21 refers to the model selection information to determine whether there is a learning model fine-tuned on a company-by-company basis (step S241). If it is determined that there is a learning model fine-tuned on a company-by-company basis (step S241: Yes), the processing unit 21 uses the learning model fine-tuned on a company-by-company basis to detect the status of the molding machine 1 (step S242) and ends the processing.

[0160] If it is determined that there is no learning model fine-tuned on a per-related-enterprise basis (step S241: No), the processing unit 21 refers to the model selection information to determine whether a second general learning model exists (step S243). If it is determined that the second general learning model is used (step S243: Yes), the processing unit 21 uses the second general learning model to detect the status of the molding machine 1 (step S244) and ends the processing. If it is determined that the second general learning model is not used (step S243: No), the processing unit 21 uses the first general learning model to detect the status of the molding machine 1 (step S245) and ends the processing.

[0161] According to the information processing method of Embodiment 2, the learning model can be fine-tuned by unit of associated enterprise, unit of enterprise, unit of factory, unit of edge, unit of machine, and unit of process, thereby detecting the state of molding machine 1 with higher accuracy.

[0162] Each company A can select the level to be fine-tuned and can use the learning model fine-tuned with the desired level to detect the state of molding machine 1.

[0163] Furthermore, an example of fine-tuning the learning model performed by the information processing device 2 has been described. However, since the confidentiality of the first general learning model is very low, it can also be configured to provide the first general learning model to enterprise A, and have the edge computer 6 use the first general learning model to detect the state of the molding machine 1. Alternatively, it can be configured to have the edge computer 6 use sensor value data to fine-tune the first general learning model.

[0164] (Implementation Method 3)

[0165] The difference between the information processing method, information processing device 2, molding machine system, and computer program P in Embodiment 3 and Embodiment 2 is that information that will become the index for selecting the optimal learning model from among multiple learning models is provided to Company A. Other components of the molding machine system, etc., are the same as those in Embodiment 2; therefore, the same reference numerals are used to label the same parts, and detailed descriptions are omitted.

[0166] Figure 20 as well as Figure 21 This is a flowchart illustrating the processing sequence of the proposal for the optimized learning model in Embodiment 3. The processing unit 21 of the information processing device 2 receives sensor value data, etc., sent from an edge computer 6 of an enterprise A (step S311), and stores the received sensor value data, etc., in an individual database 31 (step S312). Next, the processing unit 21 selects a learning model to be used based on model selection information (step S313), and inputs sensor value data, etc., into the selected learning model, thereby detecting the state of the molding machine 1 (step S314).

[0167] Next, the processing unit 21 determines whether the first general learning model has been used (step S315). If it is determined that the first general learning model has been used (step S315: Yes), the processing unit 21 selects the second general learning model (step S316), inputs sensor value data, etc., into the second general learning model, thereby detecting the state of the molding machine 1 (step S317). Then, the processing unit 21 compares the state detection accuracy using the first general learning model with the state detection accuracy using the second general learning model (step S318). In addition, the processing unit 21 can calculate the state detection accuracy by referring to sensor value data, etc., from past anomalies.

[0168] If it is determined in step S315 that the first general learning model is not being used (step S315: No), the processing unit 21 determines whether the second general learning model is being used (step S319). If it is determined that the second general learning model is being used (step S319: Yes), the processing unit 21 selects the fine-tuned learning model (step S320) and inputs sensor value data, etc., into the fine-tuned learning model to detect the state of the molding machine 1 (step S321). Furthermore, in embodiment 3, even if company A only uses the second general learning model, the information processing device 2 anticipates future use and continuously fine-tunes and prepares the learning model on a company-by-company basis.

[0169] Then, the processing unit 21 compares the state detection accuracy using the second general learning model with the state detection accuracy using the fine-tuned learning model (step S322).

[0170] If it is determined in step S319 that the second general learning model was not used (step S319: No), then it is determined whether multiple learning models that detect the same state were used (step S323). That is, the processing unit 21 determines whether multiple learning models finely adjusted at different levels were used to detect the state of the molding machine 1. If it is determined that multiple learning models of the same type were used (step S323: Yes), the processing unit 21 compares the state detection accuracy of each of the multiple learning models (step S324). After completing the processing of steps S318, S322, or S324, the processing unit 21 sends the state detection accuracy of each learning model and the comparison result together with the state detection result to the control device 14 or the terminal device 8 (step S325), and ends the processing.

[0171] According to the information processing method of Embodiment 3, the control device 14 or the terminal device 8 receives and displays the state detection results, state detection accuracy, and comparison results sent from the information processing device 2. Users of each company A can select which learning model to use by referring to the state detection results and comparison results of various learning models.

[0172] For example, Company A, which uses the first general learning model, can confirm that the detection accuracy has been improved by referring to the comparison results of the state detection accuracy with the second general learning model, and can choose to use the second general learning model as needed.

[0173] Similarly, the comparison of the state detection accuracy of the enterprise A reference using the second general learning model and the fine-tuned learning model confirms that the detection accuracy has been improved, and the fine-tuning can be selected as needed.

[0174] In addition, Company A, which uses a learning model fine-tuned at multiple levels, can obtain the state detection results of the most accurate optimized learning model and can choose to use that learning model.

[0175] In addition, the information processing device 2 may also be configured to list the detection accuracy of each learning model along with multiple learning models and provide it to the control device 14 or the terminal device 8.

[0176] (Implementation Method 4)

[0177] The information processing method, information processing device 2, molding machine system, and computer program P in Embodiment 4 differ from those in Embodiments 2 and 3 in that they are stored in a corresponding association with environmental data and sensor value data, and the learning model is fine-tuned using the environment of the molding machine 1. Other components, such as the molding machine system, are the same as those in Embodiments 2 and 3; therefore, the same reference numerals are used for the same parts, and detailed descriptions are omitted.

[0178] Figure 22 This is a conceptual diagram showing the individual database 31 of Embodiment 4. The molding machine system of Embodiment 4 includes an environmental detection unit that detects physical quantities representing the environment of each molding machine 1, each edge computer 6, or each factory. The environmental detection unit detects, for example, ambient temperature and humidity and outputs them to the edge computer 6. The edge computer 6 sends the machine ID, sensor value data output from sensor 5, molding machine control parameters, and machine configuration data, along with the environmental data output from the environmental detection unit, to the information processing device 2.

[0179] The information processing device 2 stores sensor data and environmental data sent from the edge computer 6 in a corresponding database 31.

[0180] Information processing device 2 uses the same processing as described in embodiment 2 to fine-tune the learning model on a process-by-process basis. However, the information processing in embodiment 4 is based on sensor value data obtained from multiple molding machines 1 of the same model using the same raw materials in the same or similar environment and stored in individual databases 31 of each enterprise A to fine-tune the second general learning model. As a result, the second general learning model can be optimized on a process-by-process basis. In addition, the aforementioned sensor value data for a specific process can also be used to fine-tune the learning model that has been fine-tuned for use with associated enterprises, for enterprise A, for the factory, and for the molding machine 1 connected to the edge computer 6.

[0181] As the user, Company A can use sensor value data obtained from molding machine 1 of the same model and using the same materials in the same or similar environment to select a finely tuned learning model as the learning model for detecting the state of molding machine 1, thereby enabling more accurate detection of abnormalities in molding machine 1.

[0182] According to the information processing method of Implementation 4, the learning model can be fine-tuned taking into account the surrounding environment.

[0183] (Implementation Method 5)

[0184] The information processing method, information processing device 2, molding machine system, and computer program P in Embodiment 5 differ from those in Embodiments 2-4 in the display processing of status detection results and charts, etc. Other components of the molding machine system are the same as those in Embodiments 2-4; therefore, the same reference numerals are used for the same parts, and detailed descriptions are omitted.

[0185] Figure 23This is an example of a screen displaying the status detection results of the molding machine 1 according to Embodiment 5. The processing unit 21 of the information processing device 2 provides the status detection result display screen 9 to the control device 14 or the terminal device 8 for display. The status detection result display screen 9 includes a search condition input unit 91 for retrieving graphs of sensor value data, a graph display unit 92, and an analysis result display unit 92a.

[0186] Figure 24 This is a flowchart illustrating the classification of charts and the processing order of labeling. The processing unit 21 of the information processing device 2 classifies charts of various sensor value data (step S511). For example, company A, as a user, can select the charts displayed on the status detection result display screen 9 for classification. Furthermore, charts of important sensor value data that should be displayed on the top screen can be classified as important charts. The method for classifying the charts is not particularly limited.

[0187] Next, the processing unit 21 assigns label information to each chart (step S512). For example, the label information includes the name of the chart, the part of the molding machine 1 where the sensor value data is obtained, the urgency of the status detection result, the analysis result of the chart trend, the user-defined attention level, the user's viewing frequency and viewing history, the user's preferences or settings, etc.

[0188] Figure 25 This is a flowchart illustrating the processing sequence of the chart display. The processing unit 21 of the information processing device 2 provides a chart retrieval screen to the control device 14 or the terminal device 8 (step S531), and accepts the chart retrieval conditions (step S532). Then, based on the accepted retrieval conditions, the processing unit 21 retrieves the appropriate chart by referring to the tag information of multiple charts (step S533). The processing unit 21 provides the chart obtained through the retrieval process and the associated status detection results to the control device 14 or the terminal device 8 (step S534).

[0189] Next, the processing unit 21 accepts the selection of one or more charts for additional analysis (step S535) and performs additional analysis (step S536).

[0190] Figure 26 This is a flowchart illustrating the prediction processing sequence as an example of additional analysis. In the prediction processing, the processing unit 21 displays the selected chart on the control device 14 or the terminal device 8 (step S551), and accepts the selection of the prediction chart and the prediction target chart (step S552). The prediction target chart is a chart showing the sensor value to be predicted. The prediction chart is a chart including the predicted value used to predict that sensor value.

[0191] Figure 27This is a schematic diagram showing the selected chart. The top chart shows the time variation of physical quantity A, the middle chart shows the time variation of physical quantity B, and the bottom chart shows the time variation of physical quantity C.

[0192] Next, the processing unit 21 accepts the selection of analysis conditions (step S553) and predicts the change of the physical quantity C relative to the most recent value (step S554). That is, it predicts the value of the physical quantity C in the future. Then, the processing unit 21 displays a graph including the predicted physical quantity C to the control device 14 or the terminal device 8 (step S555).

[0193] Figure 28 This is a schematic diagram showing the prediction results. For example, if the user selects the upper and middle charts as the prediction charts and selects the lower chart as the prediction target chart, the processing unit 21 performs a prediction of the physical quantity C, such as... Figure 28 As shown, a graph containing the predicted physical quantity C is displayed to the control device 14 or the terminal device 8.

[0194] Figure 29 This is a flowchart illustrating the analysis process sequence as an example of additional analysis. Figure 30 This is an example of a screen displaying analysis results. During analysis processing, the processing unit 21 displays the selected chart on the control device 14 or the terminal device 8 (step S571), and accepts the selection of any evaluation period (step S572). The processing unit 21 calculates various statistics for the selected evaluation period (step S573). For example, it calculates the maximum, minimum, average, dispersion, and standard deviation of physical quantities A, B, and C during the evaluation period. Then, the processing unit 21 provides and displays the calculated statistics as analysis results on the control device 14 or the terminal device 8 (step S574).

[0195] Next, the processing unit 21 accepts the selection of the comparison period (step S575). The user can set the comparison period as needed. When the comparison period is set, the processing unit 21 calculates various statistics for the comparison period and compares the various statistics in the evaluation period with the various statistics in the comparison period (step S576). Then, the comparison analysis results are provided and displayed to the control device 14 or the terminal device 8 (step S577).

[0196] According to the information processing method of Embodiment 5, graphs of various sensor values ​​and associated state detection results can be displayed. Furthermore, the processing unit 21 can retrieve and display graphs desired by the user. Moreover, the processing unit 21 can predict and display specific physical quantities. Furthermore, the processing unit 21 can display various statistics within an arbitrarily set evaluation period. Furthermore, the processing unit 21 can display comparison results of various statistics calculated within an arbitrarily set evaluation period and comparison period.

[0197] Furthermore, in embodiments 1 to 5, an example of a company or other enterprise A being the user is given, but it could also be a joint venture, research institution, or other organization.

[0198] Explanation of reference numerals in the attached figures

[0199] 1 Molding machine

[0200] 2. Information processing device

[0201] 3 Individual DB groups

[0202] 4AI model group

[0203] 5 sensors

[0204] 6 Edge computing

[0205] 7 Routers

[0206] 8. Terminal devices

[0207] 9. Status detection result display screen

[0208] 10-cylinder block

[0209] 11 Screw Shaft

[0210] 12 motors

[0211] 13. Reducer

[0212] 14 Control device

[0213] 21 Processing Department

[0214] 22 Storage Department

[0215] 23 Ministry of Communications

[0216] 22a Enterprise Information DB

[0217] 22b Machine Information DB

[0218] 30 Testing Machine Database

[0219] 31 Individual Databases

[0220] 41 General AI Group 1

[0221] 42 General AI Group 2

[0222] 43 Individual AI Fine-tuning Groups

[0223] 61. Arithmetic Unit

[0224] 62 Storage Unit

[0225] 63 Ministry of Communications

[0226] 64 Input Section

[0227] P is a computer program.

Claims

1. An information processing method for detecting the status of multiple manufacturing devices used by multiple enterprises, characterized in that, Sensor value data are obtained by detecting the physical quantities of the multiple manufacturing devices used by the multiple enterprises respectively. The sensor value data obtained from the multiple manufacturing devices will be saved to multiple databases prepared for each enterprise using the manufacturing devices. Based on sensor value data stored in the multiple databases, multiple learning models for detecting the state of the manufacturing apparatus are generated or updated through machine learning. The identification information of the enterprise is stored in association with the model selection information representing one or more learning models selected by the enterprise. The state of the manufacturing equipment of an enterprise is calculated by inputting sensor value data obtained from the manufacturing equipment of the enterprise into one or more learning models represented by model selection information associated with the identification information of the enterprise. At least one of the plurality of learning models is fine-tuned based on sensor value data obtained from at least one of the plurality of manufacturing devices. The identification information of the plurality of enterprises is stored in association with information indicating which of the learning models the enterprise uses—either the untuned model or the model fine-tuned at different levels. Furthermore, by inputting sensor value data obtained from the manufacturing equipment of an enterprise into the learning model represented by the information associated with the identification information of that enterprise, the state of the manufacturing equipment of that enterprise is calculated.

2. The information processing method according to claim 1, characterized in that, The manufacturing device is a molding machine. The multiple learning models include a learning model for detecting abnormalities or deterioration of the reducer constituting the molding machine, a learning model for detecting abnormalities or deterioration of the rotating shaft constituting the molding machine, and a learning model for detecting abnormal or inefficient operating states of the molding machine.

3. The information processing method according to claim 1 or 2, characterized in that, The learning model is fine-tuned based on sensor data obtained from the manufacturing equipment of a company, and used as the learning model for the manufacturing equipment of the company. The state of the manufacturing equipment of the enterprise is calculated by inputting sensor value data obtained from the manufacturing equipment of the enterprise into the fine-tuned learning model.

4. The information processing method according to claim 1 or 2, characterized in that, In the case where multiple factories belonging to the enterprise each have the manufacturing unit, the learning model is fine-tuned based on sensor value data obtained from the manufacturing unit located in one of the factories, and used as the learning model for that one factory. The state of the manufacturing device located in the factory is calculated by inputting sensor value data obtained from the manufacturing device located in the factory into the fine-tuned learning model.

5. The information processing method according to claim 1 or 2, characterized in that, In the case where multiple factories belonging to the enterprise each have the manufacturing equipment, the learning model is fine-tuned based on sensor value data obtained from the manufacturing equipment located in the multiple factories, and used as a learning model shared by the multiple factories. The state of the manufacturing device is calculated by inputting sensor value data obtained from the manufacturing device located in one of the plurality of factories into the fine-tuned learning model.

6. The information processing method according to claim 1 or 2, characterized in that, When sensor value data is acquired from each of the manufacturing devices via an edge computer connected to the multiple manufacturing devices, the learning model is fine-tuned based on the sensor value data of each manufacturing device obtained via the edge computer, and used as a learning model shared by the multiple manufacturing devices connected to the edge computer. The state of the manufacturing device connected to the edge computer is calculated by inputting sensor value data obtained from the manufacturing device connected to the edge computer into the fine-tuned learning model.

7. The information processing method according to claim 1 or 2, characterized in that, In the case where there are multiple manufacturing devices used by a single enterprise, the learning model is fine-tuned based on sensor data obtained from each manufacturing device to serve as the learning model for each device. The state of a manufacturing device is calculated by inputting sensor value data obtained from a manufacturing device into a learning model that is finely tuned for use with the manufacturing device.

8. The information processing method according to claim 1 or 2, characterized in that, The learning model is fine-tuned based on sensor value data obtained from multiple manufacturing devices used by multiple related companies, to serve as a shared learning model for the multiple related companies. The state of the manufacturing device is calculated by inputting sensor value data obtained from the manufacturing device used by one of the multiple related companies into the fine-tuned learning model.

9. The information processing method according to claim 1 or 2, characterized in that, The manufacturing device is a molding machine. The learning model is fine-tuned based on sensor data obtained from multiple manufacturing devices of the same model that use the same raw materials. The state of the manufacturing device is calculated by inputting sensor value data obtained from the manufacturing device of the same model using the same raw materials into the fine-tuned learning model.

10. The information processing method according to claim 1 or 2, characterized in that, The manufacturing device is a molding machine. The learning model is fine-tuned based on sensor data obtained from multiple manufacturing devices of the same model located in the same or similar environment and using the same raw materials. The state of the manufacturing device is calculated by inputting sensor value data obtained from the same or similar manufacturing device that is located in the same or similar environment and uses the same raw materials into the fine-tuned learning model.

11. The information processing method according to claim 1, characterized in that, When multiple fine-tuned learning models exist for detecting the same type of state of the manufacturing apparatus, the state detection accuracy using each learning model is calculated. The system displays the calculated state detection accuracy of each learning model.

12. The information processing method according to claim 1 or 2, characterized in that, The identification information of the multiple enterprises is stored in a corresponding association with usage information, wherein the usage information indicates which of the following learning models the enterprise uses: one generated based on sensor value data stored in the multiple databases, and the other generated based on sensor value data obtained from a manufacturing device used as a testing machine or simulation, for detecting the state of the manufacturing device. Furthermore, by inputting sensor value data obtained from the manufacturing equipment of an enterprise into the learning model represented by the information associated with the identification information of that enterprise, the state of the manufacturing equipment of that enterprise is calculated.

13. The information processing method according to claim 12, characterized in that, The state detection accuracy of the learning model, generated using sensor value data stored in the multiple databases, was calculated. The state detection accuracy of the learning model used to detect the state of the manufacturing apparatus, generated based on sensor value data obtained from the manufacturing apparatus as a testing machine or simulation, is calculated. The system displays the calculated state detection accuracy of each learning model.

14. An information processing device for detecting the status of multiple manufacturing devices used by multiple enterprises, characterized in that the information processing device comprises: The acquisition unit acquires sensor value data obtained by detecting physical quantities of the multiple manufacturing devices used by the multiple enterprises respectively; Multiple databases, each storing sensor value data obtained from the manufacturing apparatus separately for each enterprise using the manufacturing apparatus; and Processing Department The processing unit generates or updates multiple learning models for detecting the state of the manufacturing apparatus based on sensor value data stored in the multiple databases through machine learning. The processing unit stores the enterprise's identification information in association with model selection information representing one or more learning models selected by the enterprise. Furthermore, by inputting sensor value data obtained from the manufacturing equipment of a company into one or more learning models represented by model selection information associated with the identification information of that company, the state of the manufacturing equipment of that company is calculated. The processing unit further fine-tunes at least one of the plurality of learning models based on sensor value data obtained from at least one of the plurality of manufacturing devices. The processing unit stores the identification information of the plurality of enterprises in association with information indicating which of the learning models the enterprise used—either the untuned model or the model fine-tuned at different levels. Furthermore, by inputting sensor value data obtained from the manufacturing equipment of an enterprise into the learning model represented by the information associated with the identification information of that enterprise, the state of the manufacturing equipment of that enterprise is calculated.

15. A molding machine system, characterized in that, The device comprises the information processing apparatus and the molding machine as described in claim 14. The information processing device detects the status of the molding machine.

16. A program product having a computer program for causing a computer to perform processing to detect the status of multiple manufacturing devices used by multiple enterprises, the program product being characterized in that, A computer program that causes the computer to perform the following processes: Sensor value data are obtained by detecting the physical quantities of the multiple manufacturing devices used by the multiple enterprises respectively. The sensor value data obtained from the multiple manufacturing devices will be saved to multiple databases prepared for each enterprise using the manufacturing devices. Based on sensor value data stored in the multiple databases, multiple learning models for detecting the state of the manufacturing apparatus are generated or updated through machine learning. The identification information of the enterprise is stored in association with the model selection information representing one or more learning models selected by the enterprise. The state of the manufacturing equipment of an enterprise is calculated by inputting sensor value data obtained from the manufacturing equipment of the enterprise into one or more learning models represented by model selection information associated with the identification information of the enterprise. At least one of the plurality of learning models is fine-tuned based on sensor value data obtained from at least one of the plurality of manufacturing devices. The identification information of the plurality of enterprises is stored in association with information indicating which of the learning models the enterprise uses—either the untuned model or the model fine-tuned at different levels. Furthermore, by inputting sensor value data obtained from the manufacturing equipment of an enterprise into the learning model represented by the information associated with the identification information of that enterprise, the state of the manufacturing equipment of that enterprise is calculated.

17. A computer-readable storage medium having a computer program recorded thereon for causing the computer to perform processing to detect the status of multiple manufacturing devices used by multiple enterprises, the storage medium being characterized in that... The computer program contains a record that causes the computer to perform the following processes: Sensor value data are obtained by detecting the physical quantities of the multiple manufacturing devices used by the multiple enterprises respectively. The sensor value data obtained from the multiple manufacturing devices will be saved to multiple databases prepared for each enterprise using the manufacturing devices. Based on sensor value data stored in the multiple databases, multiple learning models for detecting the state of the manufacturing apparatus are generated or updated through machine learning. The identification information of the enterprise is stored in association with the model selection information representing one or more learning models selected by the enterprise. The state of the manufacturing equipment of an enterprise is calculated by inputting sensor value data obtained from the manufacturing equipment of the enterprise into one or more learning models represented by model selection information associated with the identification information of the enterprise. At least one of the plurality of learning models is fine-tuned based on sensor value data obtained from at least one of the plurality of manufacturing devices. The identification information of the plurality of enterprises is stored in association with information indicating which of the learning models the enterprise uses—either the untuned model or the model fine-tuned at different levels. Furthermore, by inputting sensor value data obtained from the manufacturing equipment of an enterprise into the learning model represented by the information associated with the identification information of that enterprise, the state of the manufacturing equipment of that enterprise is calculated.

Images