Information processing method, program, information processing device, and information processing system

Abstract

To provide an information processing method and the like that can determine a control amount for a plurality of control items before a step according to state data after each step, and can adjust the control amount according to a state after a step executed by the control amount after determination.SOLUTION: An information processing method according to one aspect is characterized by executing processing for: acquiring state data via a sensor that detects a state after a step in a plurality of steps; determining a control amount for a plurality of control items before the step according to the acquired state data; and adjusting the control amount according to a state after a step executed by the control amount after determination.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to an information processing method, a program, an information processing apparatus, and an information processing system.

Background Art

[0002] In recent years, the development of technologies for optimizing manufacturing or production has been actively promoted. For example, Patent Document 1 discloses a production management system that determines control information for controlling production elements in a target production process based on analysis results obtained by analyzing operation data (data related to production elements) and quality data (feedback data for evaluating production elements in the target production process based on the quality of downstream products) in the target production process among a plurality of production processes for producing products from raw materials.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the invention according to Patent Document 1, in order to determine control information (control amount) for controlling production elements, it is necessary to collect quality data (feedback data) for evaluating production elements in the target production process from the quality of downstream products. Therefore, there is a problem that it is not possible to determine the control amount for a plurality of control items before the process according to the state data after each process (immediately after the process), and adjust the control amount according to the state after the process executed with the determined control amount.

[0005] In one aspect, it is to provide an information processing method or the like capable of determining the control amount for a plurality of control items before the process according to the state data after each process, and adjusting the control amount according to the state after the process executed with the determined control amount.

Means for Solving the Problem

[0006] An information processing method according to one aspect acquires state data via a sensor that detects the state after a process among a plurality of processes, determines a control amount for a plurality of control items before the process according to the acquired state data, and executes a process of adjusting the control amount according to the state after the process executed by the determined control amount.

Advantages of the Invention

[0007] In one aspect, it becomes possible to determine a control amount for a plurality of control items before a process according to the state data after each process, and to adjust the control amount according to the state after the process executed by the determined control amount.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Figure 19

Figure 20

Figure 21

Figure 22

Figure 23

Best Mode for Carrying Out the Invention

[0009] Hereinafter, the present invention will be described in detail based on the drawings showing its embodiments.

[0010] (Embodiment 1) Embodiment 1 relates to a form in which, according to state data after a process (after the process has been executed) during a plurality of processes, a control amount for a plurality of control items before the process (before the process is executed) is determined, and the control amount is adjusted according to the state after the process executed by the determined control amount.

[0011] There is a problem that no problem is detected until entering an inspection process for inspecting the quality of a target product, and a large number of defective products are discharged. The target object includes food products or industrial products, etc. Food products include, for example, confectionery (such as chocolate, ice cream, biscuits, rice crackers or candies), processed meat products (such as processed meat, minced meat, hamburgers or sausages), kamaboko products (fish paste products formed by shaping and heating fish paste), omelets or rice balls, etc. Industrial products are molded products made of metal or resin (such as cast molded products, forged molded products, injection molded products or extrusion molded products).

[0012] In this embodiment, by performing inspections for each process of manufacturing the target object, the state after each process in the target object is detected. According to the state data immediately after the detected process, a control amount for a plurality of control items before the process is determined. Then, when defective products of the target object are discharged according to the state data after the process executed by the determined control amount, the control amount for the plurality of control items before the process is adjusted. In this way, by automatically adjusting the process, it is possible to minimize defects and improve the yield (the ratio of non-defective products among the target objects produced in the manufacturing industry).

[0013] FIG. 1 is an explanatory diagram showing an overview of an automated operation system. The system of this embodiment includes an information processing device 1, an information processing terminal 2, a manufacturing device 3, and a plurality of sensors, and each device transmits and receives information via a network N such as the Internet.

[0014] The information processing apparatus 1 is an information processing apparatus that performs processing, storage, and transmission / reception of various information. Specifically, the information processing apparatus 1 is an apparatus that acquires state data from sensors and performs operation control on the manufacturing apparatus 3, etc. The information processing apparatus 1 is, for example, a server apparatus, a personal computer, or a general-purpose tablet PC (personal computer), etc. In the present embodiment, hereinafter, for simplicity, the information processing apparatus 1 is read as the control apparatus 1.

[0015] The information processing terminal 2 is an administrator's terminal device that receives and displays notifications including state data, the process corresponding to the state data, and the control amounts for a plurality of control items before the process, etc. The information processing terminal 2 is, for example, an information processing device such as a smartphone, a mobile phone, a wearable device such as an Apple Watch (registered trademark), a tablet, or a personal computer terminal. Hereinafter, for simplicity, the information processing terminal 2 is read as the administrator terminal 2.

[0016] The manufacturing apparatus 3 is an apparatus that manufactures an object. For example, it may be a molding apparatus for manufacturing a molded object including a casting molded product, a forging molded product, an injection molded product, or an extrusion molded product, etc., or a blast furnace for manufacturing a mold (a general term for a mold made of metal). The molding apparatus is an apparatus that pours a raw material fabric and processes it into a certain shape using a mold. The molding apparatus (molding machine) is, for example, an injection molding machine, a hollow molding machine, a film molding machine, an extruder, a twin-screw extruder, a spinning extruder, a granulator, or a magnesium injection molding machine, etc.

[0017] The sensor is a sensor for detecting the state after a plurality of processes. The sensor includes an imaging device 41, a weight measurement sensor 42, a temperature sensor 43, an acceleration sensor 44, a pressure sensor 45, a gyro sensor, an X-ray line sensor, or a humidity sensor, etc.

[0018] The imaging device is an imaging device such as a CCD (Charge Coupled Device) camera or a CMOS (Complementary Metal Oxide Semiconductor) camera. The imaging device may be, for example, an X-ray camera, a thermal camera, a near-infrared camera, a 3D camera, a line scan camera, or an area scan camera. The X-ray line sensor is a sensor that converts received X-rays into an electrical signal.

[0019] The control device 1 according to the present embodiment acquires state data via a sensor that detects the state after a process during a plurality of processes. The control device 1 determines the control amount for a plurality of control items before the process according to the acquired state data. The control device 1 adjusts the control amount for a plurality of control items before the process according to the state after the process executed by the determined control amount. Note that the control items and the control amount will be described later.

[0020] FIG. 2 is a block diagram showing a configuration example of the control device 1. The control device 1 includes a control unit 11, a storage unit 12, a communication unit 13, an input unit 14, a display unit 15, a reading unit 16, and a mass storage unit 17. Each configuration is connected by a bus B.

[0021] The control unit 11 includes an arithmetic processing device such as a CPU (Central Processing Unit), an MPU (Micro-Processing Unit), a GPU (Graphics Processing Unit), an FPGA (Field Programmable Gate Array), a DSP (Digital Signal Processor), or a quantum processor. The control unit 11 reads and executes a control program 1P (program product) stored in the storage unit 12 to perform various information processes or control processes related to the control device 1.

[0022] Note that the control program 1P can be deployed to be executed on a single computer, or placed at one site, or distributed across multiple computers interconnected by a communication network and executed on these multiple computers.

[0023] In FIG. 2, the control unit 11 is described as a single processor, but it may also be a multi-processor. Note that the control unit 11 may execute various information processing or control processing, etc. by the same processor in the control device 1, or may be executed by different processors in the control device 1.

[0024] The storage unit 12 includes memory elements such as RAM (Random Access Memory) and ROM (Read Only Memory), and stores the control program 1P or data, etc. necessary for the control unit 11 to execute processing. Also, the storage unit 12 temporarily stores data, etc. necessary for the control unit 11 to execute arithmetic processing. The communication unit 13 is a communication module for performing communication-related processing, and transmits and receives information to and from the administrator terminal 2, etc. via the network N.

[0025] The input unit 14 is an input device such as a mouse, keyboard, touch panel, or button, and outputs the received operation information to the control unit 11. The display unit 15 is a liquid crystal display or an organic EL (electroluminescence) display, etc., and displays various information according to the instructions of the control unit 11. Note that the input unit 14 may also be a touch panel integrated with a keyboard, mouse, or the display unit 15.

[0026] The reading unit 16 reads a portable storage medium 1a including a CD (Compact Disc)-ROM or a DVD (Digital Versatile Disc)-ROM. The control unit 11 may read the control program 1P from the portable storage medium 1a via the reading unit 16 and store it in the large-capacity storage unit 17. Also, the control unit 11 may download the control program 1P from another computer via a network N or the like and store it in the large-capacity storage unit 17. Furthermore, the control unit 11 may read the control program 1P from the semiconductor memory 1b.

[0027] The large-capacity storage unit 17 includes a recording medium such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive). The large-capacity storage unit 17 includes a control amount determination model 171, a state data DB (database) 172, a control amount reference DB 173, and a threshold value DB 174.

[0028] The control amount determination model 171 is a causal relationship model in which the relationship between the cause (state data) and the result (control item), that is, the causal relationship is clearly defined. The control amount determination model 171 will be described later. The state data DB 172 stores the state data of the object for each process. The control amount reference DB 173 stores reference data for determining the control amounts for a plurality of control items for each process. The threshold value DB 174 stores the threshold values of various state data.

[0029] In this embodiment, the storage unit 12 and the large-capacity storage unit 17 may be configured as an integrated storage device. Also, the large-capacity storage unit 17 may be configured by a plurality of storage devices. Furthermore, the large-capacity storage unit 17 may be an external storage device connected to the control device 1.

[0030] The control device 1 may execute various information processing and control processing, etc., on a single computer, or may execute the processing in a distributed manner on multiple computers. The control device 1 may also be realized by multiple virtual machines provided in one server, or may be realized using a cloud server.

[0031] In this embodiment, an example will be described in which the object is chocolate, but the present invention can be similarly applied to other objects.

[0032] 3 is an explanatory diagram showing an example of a record layout of the status data DB 172. The status data DB 172 includes a production line ID column, a process ID column, a process name column, an image data column, a status column, a status quantity column, a control item column, a control quantity column, and a decision date and time column.

[0033] The production line ID column stores a unique production line ID to identify a production line that includes multiple processes. The process ID column stores a unique process ID to identify each process. The process name column stores the name of the process. For example, a chocolate production line has multiple processes including "insertion of raw materials," "mixing," "refining," "temperature adjustment," and "molding."

[0034] The image data column stores image data of the chocolate after the process. The status column stores the status of the chocolate after the process. For example, in the "ingredients addition" process, the status may be "normal", "insufficient sugar", or "insufficient flavoring". In the "mixing" process, the status may be "normal", "bias (degree of mixing)", or "foreign matter mixed in". In the "refining" process, the status may be "normal", "excessive particle size", or "uneven particle size".

[0035] In the "refining" process, the state includes "normal", "high acidity", "excessive moisture content", etc. In the "temperature adjustment" process, the state includes "normal", "not solidifying", "too solid", etc. In the "molding" process, the state includes "normal", "chipped", "dented", "cracked", "protruding", "rough surface", etc.

[0036] The state quantity column stores state quantities corresponding to the states. Note that the state quantities will be described later. The control item column stores control items corresponding to each process. For example, the control items corresponding to the "raw material input" process include "cocoa bean weighing" or "sugar weighing", etc. The control items corresponding to the "mixing" process include "mixing amount", "stirring speed", "stirring time", or "temperature", etc.

[0037] The control items corresponding to the "refinement" process include "rotation speed", etc. The control items corresponding to the "refining" process include "refining temperature" or "refining time", etc. The control items corresponding to the "tempering" process include "temperature", etc. The control items corresponding to the "molding" process include "injection speed", "injection time", "amount in the kettle", or "temperature", etc.

[0038] The control quantity column stores control quantities for the control items. For example, the control quantity for the control item "cocoa bean weighing" may be the increase or decrease in the weight of the cocoa beans, and the control quantity for the control item "temperature" may be the increase or decrease value of the temperature. The determination date and time column stores the date and time information when the control quantity was determined.

[0039] FIG. 4 is an explanatory diagram showing an example of the record layout of the control quantity reference DB 173. The control quantity reference DB 173 includes a process ID column, a process name column, a state column, a state quantity column, a control item column, and a control quantity column. Note that since each column of the control quantity reference DB 173 is the same as that of the state data DB 172, the description is omitted.

[0040] FIG. 5 is an explanatory diagram showing an example of the record layout of the threshold value DB 174. The threshold value DB 174 includes a process ID column, a process name column, an item column, and a threshold value column. The process ID column stores the process ID for identifying the process. The process name column stores the name of the process. The item column stores the items for state determination. The threshold value column stores the threshold values of the states.

[0041] Note that the storage forms of the above-described each DB are just examples, and other storage forms may be used as long as the relationships between the data are maintained.

[0042] FIG. 6 is a block diagram showing a configuration example of the administrator terminal 2. The administrator terminal 2 includes a control unit 21, a storage unit 22, a communication unit 23, an input unit 24, and a display unit 25.

[0043] The control unit 21 includes an arithmetic processing device such as a CPU or an MPU, and reads and executes a control program 2P (program product) stored in the storage unit 22 to perform various information processing and control processing related to the administrator terminal 2. In FIG. 6, the control unit 21 is described as a single processor, but it may be a multi-processor.

[0044] The storage unit 22 includes a memory element such as a RAM or a ROM, and stores a control program 2P or data necessary for the control unit 21 to execute processing. The storage unit 22 also temporarily stores data necessary for the control unit 21 to execute arithmetic processing.

[0045] The communication unit 23 is a communication module for performing communication-related processing, and transmits and receives information to and from the control device 1 etc. via the network N. The input unit 24 may be a keyboard, a mouse, or a touch panel integrated with the display unit 25. The display unit 25 is a liquid crystal display or an organic EL display etc., and displays various information according to the instructions of the control unit 21.

[0046] FIG. 7 is an explanatory diagram showing an example of a chocolate manufacturing line (manufacturing process). The chocolate manufacturing line has a plurality of steps such as "raw material input", "mixing", "refinement", "refining", "tempering", and "molding". Note that the chocolate manufacturing line is not limited to the above-described steps, and may include steps such as "roasting", "separation (removing the skin)", or "packaging".

[0047] The "raw material input" process is a process of inputting raw materials such as cacao components (cacao mass, cocoa powder, cocoa butter), edible oils and fats, sugars, dairy products (skim milk powder), flavors, or emulsifiers into the manufacturing apparatus 3. The "mixing" process is a process of mixing the various input raw materials homogeneously while heating them to an appropriate temperature with a mixer mounted on the manufacturing apparatus 3 to obtain chocolate dough.

[0048] The "refinement" process is a process of grinding the chocolate dough and atomizing it to a predetermined particle size (for example, 20 microns (μm)). The "refining" process is a process of kneading the chocolate dough over a long period of time to obtain a chocolate base solution. The "tempering" process is a process of adjusting the temperature of the chocolate dough to make the crystals of the fats and oils in the chocolate into stable crystals. The "shaping" process is a process of cooling and solidifying the shaped chocolate base solution after shaping the chocolate base solution.

[0049] As shown in the figure, the manufacturing apparatus 3 is a manufacturing apparatus that takes chocolate raw materials as input and outputs chocolate manufactured using the raw materials through a plurality of processes including "raw material input", "mixing", "refinement", "refining", "tempering", and "shaping". An imaging device 41, a weighing sensor 42, a temperature sensor 43, an acceleration sensor 44, or a pressure sensor 45, etc. are installed in the manufacturing apparatus 3. Through these sensors, the state data after each process can be acquired.

[0050] In each process, when the state of the result after the process (immediately after the process) is abnormal, the control amount (increase amount or decrease amount, etc.) for a plurality of control items before the process can be determined. Also, when the abnormal state of the result after the process executed with the determined control amount continues, the control amount for a plurality of control items before the process can be further adjusted.

[0051] For example, in the "raw material input" process, when the input amount of various raw materials (such as sugar, flavor, or cocoa beans) obtained by the weighing sensor 42 is insufficient or excessive, the control amount (such as the increase amount or decrease amount) for various raw materials is determined. Then, in the "raw material input" process executed according to the determined control amount, if the situation where the input amount of various raw materials is insufficient or excessive continues, adjustments are made to increase or decrease the various raw materials.

[0052] In the "mixing" process, according to the mixing condition of the raw materials (for example, unevenness) or the mixing amount, etc. obtained by the imaging device 41 and the weighing sensor 42, etc., the control amount for the stirring speed, stirring time, or temperature is determined. Then, in the "mixing" process executed according to the determined control amount, adjustments are made to increase or decrease the stirring speed, stirring time, or temperature according to the mixing condition or mixing amount of the raw materials obtained again.

[0053] In the "refinement" process, according to the size of the particle size obtained from the imaging device 41, etc., the control amount for the rotation speed is determined. Then, in the "refinement" process executed according to the determined control amount, adjustments are made to increase or decrease the rotation speed according to the size of the particle size obtained again.

[0054] In the "refining" process, according to the state of the chocolate base solution (for example, color) or the water content ratio, etc. obtained by the imaging device 41, etc., the control amount for the refining temperature or refining time is determined. Then, in the "refining" process executed according to the determined control amount, adjustments are made to increase or decrease the refining temperature or refining time according to the state of the chocolate base solution or the water content ratio, etc. obtained again.

[0055] In the "tempering" process, according to the firmness of the chocolate or the like obtained by the imaging device 41, etc., the control amount for the temperature is determined. Then, in the "tempering" process executed according to the determined control amount, adjustments are made to increase or decrease the temperature according to the firmness of the chocolate or the like obtained again.

[0056] In the "molding" process, for the chocolate to be output, according to the mold, size, or state (e.g., overhang, crack, chip, surface roughness, or dent) obtained by the imaging device 41 or the like, the injection speed, injection time, the amount in the pot, or the control amount for the temperature is determined. Then, in the "molding" process executed with the determined control amount, the chocolate obtained again is adjusted to increase or decrease with respect to the injection speed, injection time, the amount in the pot, or the temperature according to the mold, size, or state, etc.

[0057] In this way, after each process, by performing the determination and adjustment processing of the control amount for a plurality of control items before the process, defects can be minimized and the yield can be improved.

[0058] Hereinafter, as an example of the "molding" process included in the chocolate production line, the process of determining the control amount for a plurality of control items before the process according to the state data after the process will be described. Although an example of the "molding" process is described, it can be similarly applied to other processes.

[0059] The control device 1 acquires the state data after the "molding" process via a sensor that detects the state after the process. A plurality of sensors may be installed according to actual needs. The sensors include, for example, the imaging device 41, the weight measurement sensor 42, the temperature sensor 43, the acceleration sensor 44, the pressure sensor 45, the X-ray line sensor, or the humidity sensor, etc.

[0060] From these sensors, it is possible to acquire data including the image data, length, shape, color tone or area of the chocolate, or the temperature, humidity, pressure, speed, weight, current or voltage, etc. set for each process. The control device 1 acquires the state data based on the data obtained by these sensors.

[0061] The state data includes the state and the state quantity corresponding to the state. For example, as shown in FIG. 7, the states of the chocolate in the "forming" process include normal, overflow, crack, chip, surface roughness, or dent, etc. Also, each of the state quantities corresponding to overflow, crack, chip, surface roughness, and dent states, etc. is indicated by an area. For example, when the state is a chip, the state quantity is the area of the chip. When the state is a dent, the state quantity is the area of the dent. When the state is a crack, the state quantity is the area of the crack. When the state is an overflow, the state quantity is the area of the overflow. When the state is surface roughness, the state quantity is the area of the location with surface roughness.

[0062] When the control device 1 acquires the image data of the chocolate after the "forming" process from the imaging device 41, based on the acquired image data, it identifies (discriminates) the state of the chocolate and the state quantity corresponding to the state from the above-mentioned multiple types of states.

[0063] For example, when the control device 1 inputs the image data of the chocolate, it inputs the acquired image data of the chocolate into a learned state classification model that outputs the state and state quantity of the chocolate, and outputs (identifies) the state and state quantity of the chocolate in the "forming" process.

[0064] The state classification model is used as a program module that is part of the artificial intelligence software. The state classification model is a classifier that has constructed a neural network with the image data of the chocolate as the input and the classification result of classifying the state of the chocolate as the output.

[0065] The state classification model is realized, for example, using a segmentation network such as U-Net (Convolutional Networks for Biomedical Image Segmentation). The state classification model includes a state extraction unit and an output unit.

[0066] The state extraction unit includes multiple encoder layers and multiple decoder layers connected behind them. Each encoder layer includes a pooling layer and a convolutional layer. Through semantic segmentation, a label is assigned to each pixel constituting the input chocolate image data. That is, when image data is input to the state extraction unit, the state extraction unit assigns a label to each pixel, classifying them into states such as normal, protruding, cracked, chipped, rough surface, or dented. The output unit outputs a classification result obtained by classifying the state in which a label is assigned to each pixel of the image data.

[0067] The classification result includes a state and a state quantity corresponding to the state according to the segmentation result of the U-Net. The state quantity (area) is calculated based on the number of pixels of the state. As shown in Figure 7, for example, the state of the chocolate is classified as "protruding 4 mm2", "cracked 3 mm2", "chipped 1 mm2", "protruding 4 mm2, and rough surface 15 mm2", or "dented 2 mm2", and the classification result is displayed on the screen. Also, when there are no segments of defects such as protruding, cracked, chipped, rough surface, and dented, the state of the chocolate is classified as normal. Note that when the number of pixels of an abnormal state (for example, 2) is less than or equal to a predetermined threshold (for example, 5), that is, when the area corresponding to the state (for example, 0.3 mm2) is less than or equal to a predetermined area threshold (for example, 0.5 mm2), the state of the chocolate may be classified as normal.

[0068] Note that the state classification model may be composed of a SegNet model, an FCN (Fully Convolutional Network) model, or the like.

[0069] Note that the state classification model is not limited to the above-described segmentation network, and may be implemented by other models such as CNN (Convolutional Neural Network), RCNN (Regions with Convolutional Neural Network), Fast RCNN, Faster RCNN, SSD (Single Shot Multibook Detector), YOLO (You Only Look Once), SVM (Support Vector Machine), Bayesian network, Transformer network, regression tree, or random forest.

[0070] Note that the specific process of identifying the state and state quantity of chocolate based on image data is not limited to the above-described specific process by machine learning. For example, the state of chocolate may be identified using a local feature extraction method such as A-KAZE (Accelerated KAZE), SIFT (Scale Invariant Feature Transform), SURF (Speeded-Up Robust Features), ORB (Oriented FAST and Rotated BRIEF), or HOG (Histograms of Oriented Gradients). Alternatively, light with a specific pattern may be projected onto the chocolate by a projector, and the state of the surface of the chocolate, such as unevenness, may be identified by analyzing the change in the pattern of the reflected light.

[0071] Alternatively, when the control device 1 acquires the weight data of the chocolate from the weight measurement sensor 42, the control device 1 identifies the state of the chocolate based on the acquired weight data. Specifically, the control device 1 compares the weight data acquired from the weight measurement sensor 42 with a weight threshold value. The control device 1 identifies the state based on the result of comparing the weight data with the weight threshold value.

[0072] For example, when the weight data of the chocolate is less than or equal to the weight threshold, the control device 1 determines that the state is "insufficient weight". The control device 1 calculates the difference between the weight data and the weight threshold (weight threshold - weight data) as the state quantity. Alternatively, when the weight data of the chocolate exceeds the weight threshold, the control device 1 determines that the state is "excessive weight". The control device 1 calculates the difference between the weight data and the weight threshold (weight data - weight threshold) as the state quantity.

[0073] The control device 1 specifies a plurality of control items before the process using the control quantity determination model 171 according to the specified state and state quantity of the chocolate. The control quantity determination model 171 is a model constructed using an existing causal search method such as, for example, the PC (Peter and Clark) algorithm, the GES (Greedy Equivalence Search) algorithm, the LiNGAM (Linear Non-Gaussian Acyclic Model), or the SAM (Structural Agnostic Modelling) which is a causal search method using Deep Learning.

[0074] By using these causal search methods, the control quantity determination model 171 may be represented as a directed acyclic graph (DAG) that shows the causal relationship between these nodes by connecting the nodes (variables) with arrows, so-called a causal diagram.

[0075] FIG. 8 is a schematic diagram for explaining a method of creating a causal diagram. FIG. 8 is an example of creating a causal diagram 11a when seven nodes from "A" to "G" are included. As shown in the figure, the causal diagram 11a has a plurality of nodes (variables) including "A (crack)", "B (protrusion)", "C (chip)", "D (dent)", "E (quantity in the pot)", "F (discharge rate)", and "G (temperature)", and connects two nodes with a causal relationship with an edge line. As shown in the figure, the left nodes (A, B, C, and D) are nodes indicating states (factors), and the right nodes (E, F, G) are nodes indicating control items (results).

[0076] Regarding the causal relationship, for example, when the amount in the pot is small, the discharge is temporarily interrupted, the area becomes small, or the thickness disappears and the unevenness becomes large. Or, when the discharge speed is fast, the discharge increases and overflows, and when the discharge speed is slow, there are gaps on both sides. Or, when the temperature is high, it leads to dripping and overflowing, and when the temperature is low, the viscosity increases, resulting in gaps or unevenness.

[0077] The control device 1, for example, uses the LiNGAM algorithm. With seven nodes from "A" to "G" as inputs, based on the generation process of linear data and the non-cyclic causal relationship, etc., the causal structure is uniquely determined from the data, and finally a linear structural equation is output. The control device 1 creates a causal diagram 11a showing the causal relationship of a plurality of nodes by exploring the causal relationship between the state and the control items based on the output linear structural equation. Since the LiNGAM algorithm is a known technology, the details of the method for creating the causal diagram 11a are omitted in this embodiment.

[0078] Note that regarding the causal relationship between the state and the control items, an operator or the like may determine it manually. The control device 1 associates the determined state and control items and stores them in the control amount reference DB 173.

[0079] The control device 1 uses the control amount determination model 171 represented by the above-described causal diagram 11a to identify a plurality of control items before the process according to the acquired state. For example, when the state of the chocolate is a dent, the control device 1 inputs the "D (dent)" node indicating the state as input data into the control amount determination model 171. The control device 1 outputs (identifies) the "E (amount in the pot)" node and the "G (temperature)" node indicating the control items that have a causal relationship with the "D" node from the control amount determination model 171.

[0080] The control device 1 determines the control amount for each control item based on the specified state and state quantity, and the single or multiple control items output from the control amount determination model 171. Specifically, the control device 1 refers to the control amount reference DB 173 based on the process ID, the specified state and state quantity, and the control items (for example, the amount and temperature in the kettle) output from the control amount determination model 171, and acquires the corresponding control amount.

[0081] In the above example, for example, the state of the chocolate is a dent, and the area of the dent is medium (for example, 3mm 2 ~5mm 2 ), and when the control items output from the control amount determination model 171 are "the amount in the kettle" and "temperature", the control device 1 refers to the control amount reference DB 173 based on the process ID, the state of "dent", the area of the dent, the control item of "the amount in the kettle", and the control item of "temperature", and acquires the control amounts for "the amount in the kettle" and "temperature". For example, the control amount for "the amount in the kettle" may be the increased amount in the kettle (for example, 50 mg), and the control amount for "temperature" may be the temperature increase value (for example, 4 °C).

[0082] In addition, when the control device 1 specifies a plurality of states (for example, "overflow 4 mm2 and crack 3.5 mm2"), each state is input to the control amount determination model 171, and a plurality of control items before the process are specified. Specifically, first, the control device 1 inputs the "B (overflow)" node indicating the overflow state as input data to the control amount determination model 171. The control device 1 specifies the "F (discharge rate)" node and the "G (temperature)" node indicating the control items that have a causal relationship with the "B" node from the control amount determination model 171.

[0083] Next, the control device 1 inputs the "A (crack)" node indicating the crack state as input data to the control amount determination model 171. The control device 1 specifies the "E (the amount in the kettle)" node indicating the control item that has a causal relationship with the "A" node from the control amount determination model 171.

[0084] Then, based on the identified state and state quantity, and the control items (e.g., discharge rate, temperature, and quantity in the kettle) output from the control quantity determination model 171, the control device 1 determines the control quantity for each control item by referring to the control quantity reference DB 173.

[0085] Note that the determination process of the control quantity for the control item is not limited to the above-described causal search method, and for example, the reinforcement learning model in Embodiment 2 described later may be used.

[0086] The control device 1 stores the identified state and state quantity, and the control quantity for the plurality of control items before the determined process, in association with the production line ID and the process ID in the state data DB 172. The control device 1 transmits the production line ID, the process ID, and the control quantity for the plurality of control items before the determined process to the administrator terminal 2. The administrator terminal 2 receives the production line ID, the process ID, and the control quantity transmitted from the control device 1 and displays them on the screen.

[0087] Next, a process of adjusting the control quantity according to the state after the process executed by the determined control quantity will be described.

[0088] FIG. 9 is an explanatory diagram showing an example of an adjustment screen for the control quantity for the control item. FIG. 9A is an explanatory diagram showing a first example of the adjustment screen for the control quantity. FIG. 9B is an explanatory diagram showing a second example of the adjustment screen for the control quantity. The screen includes a process display column 12a, an image display column 12b, a state display column 12c, a previous control quantity display column 12d, a proposed control quantity display column 12e, a change button 12f, a cancel button 12g, and a manual input button 12h.

[0089] The process display column 12a is a display column for displaying the name of the process and the like. The image display column 12b is a display column for displaying an image of chocolate. The state display column 12c is a display column for displaying the state of the chocolate (e.g., dents) and the state quantity corresponding to the state. The previous control quantity display column 12d is a display column for displaying the control quantity for the previous control item. The proposed control quantity display column 12e is a display column for displaying the control quantity for the control item proposed to be adjusted.

[0090] The change button 12f is a button for changing the control amount for the proposed control item. The cancel button 12g is a button for retaining the previous control amount without adopting the control amount for the proposed control item. The manual input button 12h is a button for accepting the input of the control amount for the control item to be adjusted.

[0091] The control device 1 acquires, via the imaging device 41, the image data after the process (for example, "molding") executed with the determined control amount. The control device 1 specifies the state of the chocolate (normal, overflow, crack, chip, surface roughness, or dent, etc.) and the state amount (area, etc.) corresponding to the state according to the acquired image data. Note that the state and the state amount may be specified using the above-described machine learning or local feature amount extraction method, etc.

[0092] The control device 1 displays the process name in the process display column 12a, displays the acquired chocolate image data in the image display column 12b, and displays the specified chocolate state and state amount in the state display column 12c. As shown in FIG. 9A, "A dent has been detected. Dent area: 2.6 mm 2 " is displayed in the state display column 12c. As shown in FIG. 9B, "A dent has been detected. Dent area: 5.5 mm 2 " is displayed in the state display column 12c.

[0093] The control device 1 acquires the control amount for each previous control item from the state data DB 172 based on the production line ID, process ID, specified state, and determination date and time. The control device 1 displays the control amount for each acquired control item in the previous control amount display column 12d. The control device 1 displays the control amount for each proposed control item to be adjusted in the proposed control amount display column 12e based on the control amount adjustment process (FIG. 10) described later according to the state after the process executed with the determined control amount in the present process.

[0094] FIG. 10 is an explanatory diagram for adjusting a control amount with respect to a control item. For example, when the control amount for the determined "amount in the kettle" is "+50 mg" and the control amount for "temperature" is "+4 °C", the state of the chocolate after the process executed with the adjusted control amount is acquired a predetermined number of times (for example, 3 times).

[0095] On the other hand, when the acquired state of the chocolate is continuously normal a predetermined number of times within a certain monitoring period (for example, 3 days), the control device 1 does not require readjustment for the determined control amount and does not update the control amount reference DB 173.

[0096] On the other hand, when there is an abnormality, for example, once, within a predetermined number of times, the control device 1 adjusts the control amount stored in the control amount reference DB 173 and updates the adjusted control amount to the control amount column of the control amount reference DB 173. When there are multiple abnormalities, the control device 1 may adjust the control amount corresponding to one randomly selected abnormality among the multiple abnormalities. Alternatively, the control device 1 may adjust the control amount corresponding to the abnormality with the largest state amount (for example, the area of the dent).

[0097] The control device 1 determines whether the abnormality has been improved according to the state after the process executed with the determined control amount. Specifically, the control device 1 determines whether the abnormality has been improved by comparing the state amount after the process executed with the determined control amount with the previous state amount. Based on the determination result of the improvement of the abnormality, the control device 1 adjusts the control amounts for a plurality of control items before the process.

[0098] Hereinafter, an example will be described in which the previous state is "dent" and the area of the dent, which is the state amount, is medium (for example, 3 mm 2 ~5 mm 2 ).

[0099] First, an example where the abnormality is improved will be described. For example, if the indentation of the chocolate after the process executed with the determined control amount still exists in the control device 1, but the area of the current indentation is smaller than the area of the previous indentation, it is determined that the abnormality is improved. In this case, based on the control amounts for each previous control item, the control amounts for each control item are adjusted to be further increased or decreased by the first adjustment amount. The control device 1 stores (updates) the control amount (first control amount) by the first adjustment amount in the control amount reference DB173 in association with the process ID, state, and control item.

[0100] For example, in the control device 1, when the area of the indentation of the chocolate after the process executed with the determined control amount becomes smaller and the area of the indentation is small (for example, 3 mm 2 or less), as the first adjustment amount, it may be adjusted to decrease a predetermined reduction amount (for example, 0.5 mg) with respect to the "amount in the pot". The control device 1 updates the control amount column of the control amount reference DB173 with the adjusted control amount of "amount in the pot: 49.5 mg" in association with the process ID, state, and control item.

[0101] Note that the adjustment process described above is not limited thereto. For example, the control device 1 may be adjusted to decrease a predetermined reduction amount (for example, 0.2 mg) with respect to the "amount in the pot" as the first adjustment amount, and to increase a predetermined temperature value (for example, 0.2 °C) with respect to the "temperature". The control device 1 updates the control amount column of the control amount reference DB173 with the adjusted control amount of "amount in the pot: +49.8 mg temperature: +4.2 °C" in association with the process ID, state, and control item.

[0102] Next, an example where the abnormality has not improved will be described. For example, if the area of the dent in the chocolate after the process executed with the controlled amount after determination is larger than the area of the previous dent, the control device 1 determines that the abnormality has not improved, that is, the abnormality has further deteriorated. In this case, the control device 1 adjusts the controlled amount for each control item to further increase or decrease it by a second adjustment amount different from the above-described first adjustment amount, or adds a new control item. The control device 1 stores (updates) the controlled amount (second controlled amount) based on the second adjustment amount in the controlled amount reference DB 173 in association with the process ID, state, and control item.

[0103] For example, if the area of the dent in the chocolate after the process executed with the controlled amount after determination becomes larger and the area of the dent is large (for example, 5 mm 2 or more), as the second adjustment amount, it may be adjusted to increase a predetermined increase amount (for example, 2 mg) with respect to the "amount in the kettle". The control device 1 updates the controlled amount column of the controlled amount reference DB 173 with the adjusted controlled amount of "amount in the kettle: 52 mg" in association with the process ID, state, and control item.

[0104] Note that the adjustment process described above is not limited thereto. For example, the control device 1 may be adjusted to increase a predetermined increase amount (for example, 1 mg) with respect to the "amount in the kettle" as the second adjustment amount and increase a predetermined temperature value (for example, 0.1 ° C) with respect to the "temperature". The control device 1 updates the controlled amount column of the controlled amount reference DB 173 with the adjusted controlled amount of "amount in the kettle: +51 mg temperature: +4.1 ° C" in association with the process ID, state, and control item.

[0105] Alternatively, the control device 1 may refer to the control quantity reference DB 173 and specify the "injection speed" other than the "quantity in the pot" and the "temperature" from the control quantity reference DB 173 based on the area of the dent (state quantity) which is large. The control device 1 acquires the control quantity (+5 mm / s) for the specified "injection speed" from the control quantity reference DB 173. In this case, the control device 1 may adjust to further increase a predetermined increase amount (for example, 1 mg) with respect to the "quantity in the pot" as the second adjustment amount. The control device 1 updates the adjusted control quantity "quantity in the pot: +51 mg injection speed: +5 mm / s" to the control quantity column of the control quantity reference DB 173 in association with the process ID, the state, and the control item.

[0106] Subsequently, returning to FIGS. 9A and 9B, FIG. 9A shows an example in which the abnormality is improved. As shown in FIG. 9A, when the control device 1 determines that the dent of the chocolate after the process executed with the determined control quantity still exists, but the area of the current dent (for example, 2.6 mm2) is smaller than the area of the previous dent (for example, 3.8 mm2), the control quantity for the control item is adjusted by the control quantity adjustment process shown in FIG. 10.

[0107] The control device 1 displays the adjusted control quantity in the proposed control quantity display column 12e. As shown in the drawing, the adjusted control quantity "quantity in the pot: +51 mg temperature: +4.1 °C" is displayed in the proposed control quantity display column 12e.

[0108] FIG. 9B shows an example in which the abnormality further deteriorates. Different proposals from those in FIG. 9A are shown in FIG. 9B. As shown in FIG. 9B, when the control device 1 determines that the area of the dent of the chocolate after the process executed with the determined control quantity (for example, 5.5 mm2) is larger than the area of the previous dent (for example, 3.8 mm2), the control quantity for the control item is adjusted by the control quantity adjustment process shown in FIG. 10.

[0109] The control device 1 displays the adjusted control amount in the proposed control amount display column 12e. As shown in the figure, the control device 1 adds the "injection speed", which is a new control item, to the proposed control amount display column 12e without the need for readjustment of "temperature", and displays it. Specifically, the adjusted control amount of "Amount in the kettle: +52 mg Injection speed: +5 mm / s" is displayed in the proposed control amount display column 12e.

[0110] When the control device 1 receives a touch (click) operation of the change button 12f, it outputs a change instruction including the control amount for each proposed control item to the manufacturing device 3 and the corresponding sensor (for example, the weight measurement sensor 42 or the temperature sensor 43).

[0111] In addition, when the state of the chocolate (for example, cracking) after the process executed with the determined control amount is different from the previously specified state (for example, denting), the control device 1 uses the control amount determination model 171 to specify a plurality of control items before the process according to the new state and state amount of the chocolate. The control device 1 determines the control amount for each control item by referring to the control amount reference DB 173 based on the new state and state amount of the chocolate and the plurality of control items specified by the control amount determination model 171.

[0112] When the control device 1 receives a touch operation of the cancel button 12g, it does not output a change instruction to the manufacturing device 3 and the corresponding sensor in order to hold the previous control amount. When the control device 1 receives a touch operation of the manual input button 12h, for example, it displays a control amount input screen. The control device 1 receives the input of the adjusted control amount by the administrator, operator, etc. through the control amount input screen. The control device 1 outputs a change instruction including the received adjustment amount to the manufacturing device 3 and the corresponding sensor.

[0113] Note that the above is not limited to manual input of the control amount. For example, the control device 1 transmits the specified state data (state and state quantity) of the chocolate to the administrator terminal 2. The administrator terminal 2 displays the state data of the chocolate transmitted from the control device 1 on the screen. The administrator terminal 2 may receive the input of the adjusted control amount by the administrator according to the state data of the chocolate, and transmit the received adjusted control amount to the control device 1.

[0114] Note that the control amount determination model 171 in this embodiment may be constructed by performing pre-training using a large-scale text data (dataset) with a large language model (LLM) such as ALBERT (A Lite BERT), GPT (Generative Pre-trained Transformer)-2, GPT-3, GPT-4, LLaVA (Large Language and Vision Assistant), MiniGPT-4, or BERT (Bidirectional Encoder Representations from Transformers).

[0115] In addition, when the state data exceeds the threshold value, the control device 1 notifies the administrator. Specifically, the control device 1 compares the acquired state data with the threshold value of the state data stored in the threshold value DB 174. When the acquired state data exceeds the threshold value, the control device 1 transmits (notifies) the state data, the process corresponding to the state data, and the determined control amount to the administrator terminal 2.

[0116] For example, the control device 1 acquires state data including the state of "dent (area)" and the state quantity of the dent area. According to the acquired state data, the control device 1 uses the control amount determination model 171 to determine the control amounts for a plurality of control items before the process. When the acquired dent area exceeds the threshold value of the dent area stored in the threshold value DB 174, the control device 1 transmits the state data, the process corresponding to the state data (for example, "molding"), and the determined control amount to the administrator terminal 2.

[0117] The manager terminal 2 receives the state data transmitted from the control device 1, the process corresponding to the state data, and the determined control amount, and displays them on the screen.

[0118] FIG. 11 is a flowchart showing a processing procedure for determining control amounts for a plurality of control items before a process. The control unit 11 of the control device 1 executes a subroutine for acquiring state data (step S101). The subroutine for acquiring state data will be described later. The control unit 11 inputs the acquired state data as input data to the control amount determination model 171, and specifies a plurality of control items before the process having a causal relationship (step S102).

[0119] Based on the process ID, the acquired state and state amount, and each specified control item, the control unit 11 refers to the control amount reference DB 173 to determine the control amount for each control item (step S103). The control unit 11 transmits the specified plurality of control items and the control amount for each determined control item to the manager terminal 2 by the communication unit 13 (step S104).

[0120] The control unit 21 of the manager terminal 2 receives the control items and the control amount transmitted from the control device 1 by the communication unit 23 (step S201). The control unit 21 displays the received control items and the control amount on the display unit 25 (step S202).

[0121] The control unit 11 stores the acquired state data and the control amounts for the plurality of control items before the determined process in the state data DB 172 of the mass storage unit 17 (step S105). Specifically, the control unit 11 stores the process name, image data, state, state amount, control item, the control amount for the control item, and the determination date and time as one record in the state data DB 172 in association with the production line ID and the process ID.

[0122] The control unit 11 outputs an execution instruction including a plurality of control items before the determined process and the control amount for each control item to the manufacturing apparatus 3 and the corresponding sensors via the communication unit 13 (step S106). The control unit 11 acquires a threshold value corresponding to the state from the threshold value DB174 of the mass storage unit 17 based on the process ID (for example, "molding") of the process and the specified state (step S107).

[0123] The control unit 11 determines whether or not the specified state quantity exceeds the threshold value (step S108). If the specified state quantity does not exceed the threshold value (NO in step S108), the control unit 11 ends the process. If the specified state quantity exceeds the threshold value (YES in step S108), the control unit 11 transmits the state data (state and state quantity), the process corresponding to the state data (for example, "molding"), and the determined control amount to the administrator terminal 2 via the communication unit 13 (step S109).

[0124] The control unit 21 of the administrator terminal 2 receives the state data, process, and control amount transmitted from the control device 1 via the communication unit 23 (step S203). The control unit 21 displays the received state data, process, and control amount on the display unit 25 (step S204) and ends the process.

[0125] FIG. 12 is a flowchart showing the processing procedure of a subroutine for acquiring state data. The control unit 11 of the control device 1 acquires image data of the chocolate after the process from the imaging device 41 via the communication unit 13 (step S01).

[0126] The control unit 11 specifies the state of the chocolate by performing image recognition processing on the acquired image data (step S02). For example, when the control unit 11 inputs the image data of the chocolate, it uses a learned state classification model that outputs the state of the chocolate to specify the state of the chocolate (normal, overflow, crack, chip, surface roughness, or dent, etc.) and the state quantity corresponding to the state (for example, the area of the dent or the area of the surface roughness).

[0127] The control unit 11 ends the subroutine for acquiring the state data and returns. In FIG. 12, an example of image data was described, but the same can be similarly applied to other types of data (such as weight data, temperature data, pressure data, or humidity data). That is, the state and state quantity of the chocolate can be specified using one or a plurality of sensors.

[0128] FIG. 13 is a flowchart showing the processing procedure for adjusting the control amount according to the state after the process. FIG. 13 is a flowchart showing the processing procedure for adjusting the control amount according to the state after the process executed by the control amount determined in the process of step S103 (FIG. 11).

[0129] The control unit 11 of the control device 1 acquires, via the imaging device 41, the image data after the process executed by the determined control amount by the communication unit 13 (step S111). The control unit 11 uses the learned state classification model described above to specify the state of the chocolate (normal, overflow, crack, chip, surface roughness, or dent, etc.) and the state quantity corresponding to the state according to the acquired image data (step S112).

[0130] The control unit 11 determines whether the number of executions of the acquisition process of the image data after the process executed by the determined control amount has reached a predetermined number (for example, 3 times) (step S113). If the number of executions has not reached the predetermined number (NO in step S113), the control unit 11 returns to the process of step S111. If the number of executions has reached the predetermined number (YES in step S113), the control unit 11 determines whether it is continuously normal for the predetermined number of times (step S114).

[0131] If it is continuously normal for the predetermined number of times (YES in step S114), the control unit 11 ends the process. If it is not continuously normal for the predetermined number of times (NO in step S114), the control unit 11 determines whether it has been improved compared to the previous abnormality (step S115).

[0132] For example, as an example where the previous state is "dent" and the area of the dent is medium, when the area of the dent after the process executed with the determined control amount is small, the control unit 11 determines that it has been improved from the previous abnormality. Or, when the area of the dent after the process executed with the determined control amount is large, the control unit 11 determines that it has not been improved from the previous abnormality.

[0133] When the control unit 11 determines that it has not been improved from the previous abnormality, that is, when it is determined that it has deteriorated further from the previous abnormality (NO in step S115), the control unit 11 acquires a second adjustment amount corresponding to the current state, state quantity, and control item from the storage unit 12 or the mass storage unit 17 (step S193). Note that the second adjustment amount is stored in advance in the storage unit 12 or the mass storage unit 17 according to the state, state quantity, and control item. The control unit 11 determines a second control amount for each control item based on the acquired second adjustment amount (step S194). For example, the control unit 11 adjusts to increase a predetermined increase amount (for example, 1 mg) for the "amount in the kettle" and adjusts to increase a predetermined temperature value (for example, 0.2 °C) for the "temperature", and determines the second control amounts for the "amount in the kettle" and the "temperature" according to the second adjustment amount. The control unit 11 transitions to the process of step S118 described later.

[0134] When the control unit 11 determines that it has been improved from the previous abnormality (YES in step S115), the control unit 11 acquires a first adjustment amount corresponding to the current state, state quantity, and control item from the storage unit 12 or the mass storage unit 17 (step S116). Note that the first adjustment amount is stored in advance in the storage unit 12 or the mass storage unit 17 according to the state, state quantity, and control item. The control unit 11 determines a first control amount for each control item based on the acquired first adjustment amount (step S117). For example, the control unit 11 adjusts to decrease a predetermined decrease amount (for example, 0.2 mg) for the "amount in the kettle" and adjusts to increase a predetermined temperature value (for example, 0.1 °C) for the "temperature", and determines the control amounts for the "amount in the kettle" and the "temperature" according to the first adjustment amount.

[0135] The control unit 11 displays a proposal including the adjusted control amount on the screen (FIG. 9A or FIG. 9B) via the display unit 15 (step S118). Note that the above-described proposal may be displayed on the screen in the form of a chatbot using a language generation model constructed using a large language model (LLM).

[0136] The control unit 11 receives an instruction from an operator or the like via the input unit 14 (step S119). The control unit 11 determines whether the received instruction is an instruction to change the adjusted control amount (step S120). If the instruction is not a change instruction (NO in step S120), the control unit 11 ends the process.

[0137] If the instruction is a change instruction (YES in step S120), the control unit 11 updates the adjusted control amount in the control amount column of the control amount reference DB173 in the mass storage unit 17 in association with the process ID, status, and control item (step S191). The control unit 11 outputs a change instruction including the adjusted control amount to the manufacturing apparatus 3 and the corresponding sensor via the communication unit 13 (step S192). The control unit 11 ends the process.

[0138] Note that if the abnormality is not improved and does not deteriorate, that is, if it is the same as the previous abnormality, the control unit 11 may not require readjustment of the previous control amount and may output an instruction to execute the process with the previous adjustment amount to the manufacturing apparatus 3 and the corresponding sensor.

[0139] According to the present embodiment, it is possible to determine control amounts for a plurality of control items before the process according to the post-process state data acquired via the sensor.

[0140] According to the present embodiment, it is possible to adjust the control amounts for a plurality of control items before the process according to the state after the process executed with the determined control amount.

[0141] According to the present embodiment, when the state data exceeds the threshold value, it is possible to notify the administrator terminal 2 of the state data, the process corresponding to the state data, and the determined control amount.

[0142] <Modification Example 1> A process of determining control amounts for a plurality of control items before a process will be described according to a plurality of state data in time series.

[0143] The plurality of state data in time series are, for example, data obtained according to a time series such as the state data of the time before last, the state data of the last time, and the state data of this time. The state data includes a state and a state quantity corresponding to the state. By using the plurality of state data in time series, control amounts for a plurality of control items before a process (for example, "molding") can be determined.

[0144] Specifically, the control device 1 acquires the state data of this time via an imaging device 41, a weight measurement sensor 42, a temperature sensor 43, an acceleration sensor 44, a pressure sensor 45, a humidity sensor, or the like. Since the acquisition process of the state data is the same as that in the first embodiment, the description thereof is omitted.

[0145] The control device 1 acquires the state data of the time before last and the state data of the last time from the state data DB 172 based on the production line ID, the process ID of the target process (for example, "molding"), and the determination date and time of the control amount.

[0146] The control device 1 determines control amounts for a plurality of control items before a process by using a control amount determination model 171 according to a plurality of state data in time series including the acquired state data of the time before last, the state data of the last time, and the state data of this time.

[0147] Specifically, the control device 1 inputs a plurality of state data in time series including the state data of the time before last, the state data of the last time, and the state data of this time as input data to the control amount determination model 171. The control device 1 outputs a control item having a causal relationship with the input data from the control amount determination model 171. The control device 1 acquires the corresponding control amount by referring to the control amount reference DB 173 based on the process ID, the plurality of state data in time series, and the control item output from the control amount determination model 171.

[0148] After that, the control device 1 adjusts the control amount according to the state after the process executed by the determined control amount. The control device 1 updates the adjusted control amount in the control amount reference DB 173. Note that since the adjustment and update processing of the control amount are the same as those in the first embodiment, the description thereof is omitted.

[0149] Also, when a plurality of time-series state data exceed the threshold value, the control device 1 transmits (notifies) the plurality of time-series state data, the processes corresponding to the plurality of time-series state data, and the determined control amount to the administrator terminal 2.

[0150] Specifically, the control device 1 compares the average value of the plurality of acquired time-series state data with the threshold value stored in the threshold value DB 174. When the average value exceeds the threshold value, the control device 1 transmits the plurality of time-series state data and processes, and the determined control amount to the administrator terminal 2.

[0151] For example, when the control device 1 acquires a plurality of time-series weight data in chocolate, it calculates the average value of the plurality of acquired time-series weight data. The control device 1 compares the calculated average value with the weight threshold value. When the calculated average value exceeds the threshold value, the control device 1 transmits the plurality of time-series weight data and corresponding processes, and the determined control amount to the administrator terminal 2. Note that not limited to the average value, for example, the median value, or the change rate of the plurality of time-series weight data may be used.

[0152] Note that it is not limited to the above-described processing. For example, the control device 1 compares each state data with the threshold value among the plurality of time-series state data. When any one of the plurality of time-series state data exceeds the threshold value, the control device 1 may transmit the plurality of time-series state data and corresponding processes, and the determined control amount to the administrator terminal 2.

[0153] Alternatively, the change rate or the like of a plurality of state data in time series may be used. For example, when the control device 1 acquires a plurality of temperature data in time series, it calculates the amount of change over time such as the temperature change rate. The temperature change rate (the amount of temperature change per unit time) indicates, for example, the change rate (T2 - T1 / T1) of the current temperature (T2) with respect to the previous temperature (T1). The control device 1 compares the calculated temperature change rate with a threshold value. When the calculated temperature change rate exceeds the threshold value, the control device 1 notifies the administrator terminal 2.

[0154] The administrator terminal 2 receives the plurality of state data in time series transmitted from the control device 1, the process corresponding to the plurality of state data in time series, and the determined control amount, and displays them on the screen.

[0155] FIG. 14 is a flowchart showing a processing procedure for determining control amounts for a plurality of control items before a process in Modification 1. Note that the same reference numerals are given to the contents overlapping with those in FIG. 11, and the description thereof is omitted.

[0156] The control unit 11 of the control device 1 acquires the current state data by executing a subroutine of the process of acquiring state data (step S121). The control unit 11 acquires past state data including the state data of the time before last and the state data of the last time from the state data DB172 in the mass storage unit 17 based on the production line ID, the process ID of the target process, and the date and time of determination of the control amount (step S122).

[0157] The control unit 11 specifies a plurality of control items before the process using the control amount determination model 171 according to the plurality of state data in time series including the acquired current state data and past state data (step S123). The control unit 11 determines the control amount for each control item by referring to the control amount reference DB173 based on the process ID, the plurality of state data in time series acquired, and the plurality of specified control items (step S124).

[0158] The control unit 11 executes the processes of steps S104 to S107. The control unit 21 of the administrator terminal 2 executes the processes of steps S201 to S202. The control unit 11 of the control device 1 calculates the average value of a plurality of state data in time series (step S125). Note that, not limited to the average value, for example, the median value, or the change rate of the plurality of state data in the time series may be used. The control unit 11 determines whether the calculated average value exceeds a threshold value (step S126).

[0159] When the calculated average value does not exceed the threshold value (NO in step S126), the control unit 11 ends the process. When the calculated average value exceeds the threshold value (YES in step S126), the control unit 11 transmits the plurality of state data in time series, the processes corresponding to the plurality of state data in the time series, and the determined control amount to the administrator terminal 2 via the communication unit 13 (step S127).

[0160] The control unit 21 of the administrator terminal 2 receives the plurality of state data in time series, the processes, and the control amount transmitted from the control device 1 via the communication unit 23 (step S221). The control unit 21 displays the received plurality of state data in time series, the processes, and the control amount on the display unit 25 (step S222) and ends the process.

[0161] According to this modification example, it becomes possible to determine the control amount for a plurality of control items before the process according to a plurality of state data in time series.

[0162] According to this modification example, when a plurality of state data in time series exceed the threshold value, it becomes possible to notify the administrator terminal 2 of the plurality of state data in time series, the processes corresponding to the plurality of state data in the time series, and the determined control amount.

[0163] (Embodiment 2) Embodiment 2 relates to a form in which a reinforcement learning model is used to determine and adjust the control amount for a plurality of control items before the process. Note that, descriptions of the contents overlapping with Embodiment 1 are omitted.

[0164] The control quantity determination model 171 in this embodiment is a learned model generated by reinforcement learning in each process of chocolate. In the following, although an example of the "molding" process will be described, it can be similarly applied to other processes.

[0165] FIG. 15 is a block diagram showing a configuration example of the control unit 11 in Embodiment 2. The control unit 11 includes a reward calculation unit 111, an action selection unit 112, and an action evaluation unit 113.

[0166] For example, when using reinforcement learning, the state data after the process is regarded as the "state", the control quantities for a plurality of control items before the process are regarded as the "actions", and the "reward" is calculated based on the state data after the process executed by the control quantity, and the value of the Q-value or Q-function (action value function) may be learned.

[0167] That is, the action selection unit 112 has a function as an action output unit, and outputs control quantities for a plurality of control items before the process based on the state data after the process and the Q-value or Q-function value (action evaluation information) of the action evaluation unit 113. The action evaluation unit 113 includes an evaluation value of the action in reinforcement learning, specifically, includes the Q-value or Q-function value. That is, based on the evaluation value of the action in the state data after the process, an action is selected and output from the actions that can be taken in the obtained state.

[0168] The reward calculation unit 111 calculates a reward based on the state data after the process executed by the control quantities for a plurality of control items before the process. The calculation of the reward can be made positive (with a reward) when the state data is within the required value or range, and 0 (without a reward) or negative (penalty) when the state data does not reach the required value or is not within the range. For example, when the state is normal, it can be made positive (with a reward), and when the state is abnormal, it can be made 0 or negative.

[0169] The action selection unit 112 has a function as an update unit, and updates the Q-value or Q-function value of the action evaluation unit 113 so that the reward calculated by the reward calculation unit 111 increases.

[0170] The control unit 11 can store the updated Q value or the value of the Q function of the action evaluation unit 113 in the storage unit 12 or the mass storage unit 17. By reading out the Q value or the value of the Q function stored in the storage unit 12 or the mass storage unit 17, the learned control amount determination model 171 can be reproduced.

[0171] FIG. 16 is a schematic diagram showing an example of the configuration of the control amount determination model 171. The control amount determination model 171 represents the control unit 11 (specifically, the action selection unit 112 and the action evaluation unit 113) in the present embodiment. The control amount determination model 171 has an input layer, an intermediate layer, and an output layer.

[0172] The number of input neurons in the input layer can be the number of states and state quantities of the chocolate, and the state of the chocolate (for example, a dent) and the state quantity corresponding to the state (for example, the area of the dent) are input to the input neurons of the input layer. Note that the state and state quantity of the chocolate may be obtained using the state classification model in Embodiment 1.

[0173] Note that the above-described input data is an example and is not limited thereto. For example, image data may be input to the input neurons. Alternatively, the state and state quantity of the chocolate, the weight data obtained by the weight measurement sensor 42, the temperature data obtained by the temperature sensor 43, or the pressure data obtained by the pressure sensor 45 may be input to the input neurons.

[0174] Note that the state and state quantity of each chocolate included in a lot grouped in a certain quantity (for example, 12) may be input to the input neurons. As shown in the figure, the lot 13a manufactured in one process includes 12 chocolates. The state of each chocolate is shown as s1 (normal), s2 (cracked), s3 (dented), ···, s12 (overflowed), and the state quantity is shown as the area corresponding to each state. The state and state quantity of each chocolate included in the lot 13a may be input to the input neurons simultaneously.

[0175] The number of output neurons in the output layer can be the number of action options. In the example of FIG. 16, the output neurons include output neurons for a single type of action or output neurons that combine different types of actions.

[0176] For example, as shown in FIG. 16, the output neurons can be the "value of the Q function when doing nothing", the "value of the Q function when increasing the injection speed", the "value of the Q function when decreasing the injection speed",..., the "value of the Q function when increasing the injection speed, shortening the injection time, and raising the temperature", the "value of the Q function when decreasing the injection speed and raising the temperature", etc. Note that the number of output neurons and the types of outputs are not limited to the example of FIG. 16.

[0177] Machine learning (deep reinforcement learning) using the control amount determination model 171 can be performed as follows. That is, when the state st is input to the input neurons of the control amount determination model 171, the output neurons output Q(st, at). Here, Q is a function that stores the evaluation of action a in state s. The update of the Q function can be performed by Equation (1).

[0178] Q(st, at) ← Q(st, at) + α{rt+1 + γ·maxQ(st+1, at+1) - Q(st, at)} ··· (1) Q(st, at) ← Q(st, at) + α{rt+1 - Q(st, at)} ··· (2) Q(st, at) ← Q(st, at) + α{γ·maxQ(st+1, at+1) - Q(st, at)} ··· (3)

[0179] In Equation (1), st represents the state at time t, at represents the action that can be taken in state st, α represents the learning rate (where 0 < α < 1), and γ represents the discount rate (where 0 < γ < 1). The learning rate α is also called the learning coefficient and is a parameter that determines the learning speed (step size). That is, the learning rate α is a parameter that adjusts the update amount of the Q value or the value of the Q function.

[0180] The discount rate γ is a parameter that determines how much to discount and consider the evaluation of future states (rewards or penalties) when updating the Q-function. That is, it is a parameter that determines how much to discount rewards and penalties when the evaluation in a certain state is related to the evaluation in past states.

[0181] In Equation (1), rt+1 is the reward obtained as a result of the action. When no reward is obtained, it becomes 0, and in the case of a penalty, it becomes a negative value. In Q-learning, the second term of Equation (1), {rt+1 + γ·maxQ(st+1,at+1) - Q(st,at)}, is made to be 0. That is, the Q(st,at) of the Q-function is controlled so that it becomes the sum of the reward (rt+1) and the maximum value (γ·maxQ(st+1,at+1)) among the actions possible in the next state st+1, and the parameters of the control amount determination model 171 are learned.

[0182] The parameters of the control amount determination model 171 are updated so as to bring the error between the expected value of the reward and the current action evaluation closer to 0. In other words, the value of (γ·maxQ(st+1,at+1)) is corrected based on the value of the current Q(st,at) and the maximum evaluation value obtained among the actions executable in the state st+1 after executing the action at.

[0183] When an action is executed in a certain state, a reward is not always obtained. For example, a reward may be obtained after repeating the action several times. Equation (2) represents the update formula of the Q-function when a reward is obtained in Equation (1), avoiding the problem of divergence. Equation (3) represents the update formula of the Q-function when no reward is obtained in Equation (1).

[0184] Figure 17 is an explanatory diagram showing an example of the action at. As shown in Figure 17, when the action at is the control of the injection speed, specifically, actions such as increasing the injection speed, decreasing the injection speed, or not changing the injection speed can be used. Here, how much to increase or decrease the injection speed can be set as appropriate.

[0185] When the action at is the control of the injection time, specifically, actions such as lengthening the injection time, shortening the injection time, or not changing the injection time can be used. Here, how much to lengthen or shorten the injection time can be set as appropriate.

[0186] When the action at is the control of the temperature, specifically, actions such as raising the temperature, lowering the temperature, or not changing the temperature can be used. Here, how much to raise or lower the temperature can be set as appropriate.

[0187] The output neuron can be configured to output the Q function by combining all or part of the actions illustrated in FIG. 17. Although not shown, in the "molding" process, the action at may include control items such as the amount in the kettle. That is, the action at includes control items provided for each process. For example, in the "refining" process, the action at may include the refining temperature, the refining time, and the like.

[0188] Note that it is not limited to the above-described reinforcement learning process. For example, the control amount determination model 171 may be learned using Decision Transformer, which is one of the reinforcement learning methods based on sequence modeling.

[0189] FIG. 18 is a flowchart showing the processing procedure when performing reinforcement learning of the control amount determination model 171. The control unit 11 of the control device 1 executes a subroutine of the process of acquiring state data to acquire the state data (state and state quantity) after the process (for example, "molding") (step S131). The control unit 11 inputs the acquired state and state quantity after the process to the learned control amount determination model 171 in FIG. 16 (step S132).

[0190] The control unit 11 determines the control amounts for a plurality of control items before the process using the control amount determination model 171 (step S133). The control unit 11 executes a subroutine of the process of acquiring state data to acquire the state data after the process executed by the determined control amounts (step S134).

[0191] The control unit 11 determines whether or not a predetermined number of times (for example, three times) has been reached (step S135). If the predetermined number of times has not been reached (NO in step S135), the control unit 11 returns to the process of step S134. If the predetermined number of times has been reached (YES in step S135), the control unit 11 determines whether or not it has been continuously normal for the predetermined number of times (step S136). Note that since the determination process of continuous normality is the same as the process of step S114, the description thereof is omitted.

[0192] If it has been continuously normal for the predetermined number of times (YES in step S136), the control unit 11 ends the process. If it has not been continuously normal for the predetermined number of times (NO in step S136), the control unit 11 performs reinforcement learning of the control amount determination model 171 based on the reward corresponding to the state data after the process executed by the acquired control amount after determination (step S137). The control unit 11 ends the process.

[0193] That is, the control unit 11 stores the training data of the chocolate controlled according to the result of the learned control amount determination model (reinforcement learning model) 171 in FIG. 16 in the normal or abnormal state, and performs the reinforcement learning training process of S137 at an appropriate timing.

[0194] Specifically, the reward calculation unit 111 of the control unit 11 calculates a reward based on the state after the process executed by the control amount after determination. For example, when the state is normal, it may be set to positive (with a reward), and when the state is abnormal, it may be set to 0 or negative. The action selection unit 112 updates the Q value or the value of the Q function of the action evaluation unit 113 so that the reward calculated by the reward calculation unit 111 becomes larger. That is, the control unit 11 performs reinforcement learning based on the reward calculated by the reward calculation unit 111, with the state data acquired by the process of step S131 as the state st and the action information for the control amount determined by the process of step S133 as the action at.

[0195] According to this embodiment, it is possible to determine control amounts for a plurality of control items before a process by using the control amount determination model 171 generated by a reinforcement learning method.

[0196] According to this embodiment, it is possible to adjust a highly accurate control amount by training (learning) the control amount determination model 171 based on a reward corresponding to state data after a process executed with the determined control amount.

[0197] <Modification Example 2> A process of determining and adjusting control amounts for a plurality of control items before a process using a reinforcement learning method based on a plurality of state data in time series will be described.

[0198] The control amount determination model 171 in this modification example is a learned model generated by reinforcement learning so as to output control amounts for a plurality of control items before a process when a plurality of state data in time series are input for each process of chocolate.

[0199] FIG. 19 is a schematic diagram showing an example of the configuration of the control amount determination model 171 in Modification Example 2. Note that the same reference numerals are given to the contents overlapping with FIG. 16, and the description thereof is omitted.

[0200] A plurality of states in time series and state amounts corresponding to the plurality of states in time series are input to the input neurons of the input layer. The plurality of states and state amounts in time series are data obtained according to a time series such as the states and state amounts (t1) of the time before last, the states and state amounts (t2) of the previous time, and the states and state amounts (t3) of the current time.

[0201] As shown in the figure, a plurality of states in time series (state t1, state t2, state t3, ···) and state amounts corresponding to the plurality of states in time series (state amount t1, state amount t2, state amount t3, ···) are input to the input neurons.

[0202] Note that the above input data is just an example and is not limited thereto. For example, only a plurality of time-series image data may be input to the input neuron. Alternatively, a plurality of time-series states and state quantities, and a plurality of time-series weight data, temperature data, pressure data, etc. may be input to the input neuron.

[0203] Furthermore, time-series data of lot 13a manufactured in one step of FIG. 16 may be input to the input neuron. For example, the state and state quantity of each chocolate included in the previous lot 13a, the state and state quantity of each chocolate included in the previous lot 13a, and the state and state quantity of each chocolate included in the current lot 13a are input to the input neuron.

[0204] Note that since the output neuron of the output layer and the reinforcement learning of the control quantity determination model 171 are the same as those in FIGS. 16 and 18, the description thereof is omitted.

[0205] FIG. 20 is a flowchart showing a processing procedure when performing reinforcement learning of the control quantity determination model 171 in Modification 2. Note that the same reference numerals are given to the contents overlapping with those in FIG. 18, and the description thereof is omitted.

[0206] The control unit 11 of the control device 1 acquires the current state data (state and state quantity) by executing a subroutine of a process for acquiring state data (step S171). The control unit 11 acquires past state data (state and state quantity) including the previous state data and the previous state data, etc. from the state data DB172 of the mass storage unit 17 based on the production line ID, the process ID of the target process, and the determination date and time of the control quantity (step S172).

[0207] The control unit 11 inputs a plurality of time-series states and state quantities including the acquired current state and state quantity and the past state and state quantity to the control quantity determination model 171 in FIG. 19 (step S173), and determines the control quantity for a plurality of control items before the process (step S174). The control unit 11 executes the processes of steps S134 to 137.

[0208] According to this modification example, based on a plurality of state data in time series, it is possible to determine the control amounts for a plurality of control items before the process using the control amount determination model 171.

[0209] According to this embodiment, by training the control amount determination model 171 based on the reward corresponding to the state data after the process executed by the determined control amount, it is possible to adjust the control amount with high accuracy.

[0210] <Modification Example 3> The process of adjusting the control amounts for a plurality of control items before the process will be described for each process.

[0211] FIG. 21 is an explanatory diagram for explaining the process of adjusting the control amount for each process. In this modification example, examples of the first process (for example, "tempering") and the second process (for example, "molding") in the chocolate production line will be described, but it can be similarly applied to three or more processes.

[0212] As shown in the figure, the control device 1 acquires the first state data after the "tempering" process, which is the first process. The first state data includes, for example, the firmness or sugar content of the chocolate. The control device 1 inputs the acquired first state data after the "tempering" process into the first control amount determination model 171 in FIG. 16 or FIG. 19, and determines the first control amounts for a plurality of control items (for example, temperature and speed) before the "tempering" process.

[0213] The control device 1 acquires the second state data after the "molding" process, which is the second process. The second state data includes, for example, the cracking state and the area of the cracks. The control device 1 inputs the acquired second state data after the "molding" process into the second control amount determination model 171 in FIG. 16 or FIG. 19, and determines the second control amounts for a plurality of control items (for example, injection speed, injection temperature, and the amount in the kettle) before the "molding" process.

[0214] Then, the control device 1 performs reinforcement learning (training) on the first control amount determination model 171 in FIG. 16 or FIG. 19 based on the reward corresponding to the first state data after the "temperature adjustment" process executed by the first control amount output from the first control amount determination model 171 and the second state data after the "molding" process executed by the second control amount output from the second control amount determination model 171.

[0215] As shown in the figure, the reward calculation unit 111 of the control device 1 is based on the state after the process executed by the first control amount for a plurality of control items before the "temperature adjustment" process and the state after the process executed by the second control amount for a plurality of control items before the "molding" process, and calculates the total reward (r total ).

[0216] For example, when the state after the "temperature adjustment" process is normal and the state after the "molding" process is normal, it may be set to positive (with a reward). Or, when the state after the "temperature adjustment" process is abnormal and the state after the "molding" process is normal, it may be set to 0. Or, when the state after the "temperature adjustment" process is abnormal and the state after the "molding" process is abnormal, it may be set to negative. The action selection unit 112 of the control device 1 updates the Q value or the value of the Q function of the action evaluation unit 113 so that the reward calculated by the reward calculation unit 111 becomes larger.

[0217] That is, the control device 1 performs reinforcement learning based on the total reward calculated by the reward calculation unit 111, taking the first state data after the "temperature adjustment" process and the second state data after the "molding" process as the state st and the action information for the first control amount and the second control amount as the action at.

[0218] FIG. 22 is a flowchart showing the processing procedure for determining the control amounts in the first process and the second process. The control unit 11 of the control device 1 acquires the first state data after the first process by executing a subroutine for acquiring state data (step S141). The control unit 11 inputs the acquired first state data after the first process into the first control amount determination model 171 in FIG. 16 or FIG. 19 (step S142).

[0219] The control unit 11 uses the first control quantity determination model 171 that has been reinforced based on the reward calculated based on the state after the process executed by the first control quantity, with the first state data after the first process as the state st and the action information for the first control quantity as the action at, to determine the first control quantity for a plurality of control items before the first process (step S143).

[0220] The control unit 11 acquires the second state data after the second process by executing a subroutine for the process of acquiring state data (step S144). The control unit 11 inputs the acquired second state data after the second process into the second control quantity determination model 171 in FIG. 16 or FIG. 19 (step S145).

[0221] The control unit 11 uses the second control quantity determination model 171 that has been reinforced based on the reward calculated based on the state after the process executed by the second control quantity, with the second state data after the second process as the state st and the action information for the second control quantity as the action at, to determine the second control quantity for a plurality of control items before the second process (step S146). The control unit 11 ends the process.

[0222] FIG. 23 is a flowchart showing the processing procedure when performing reinforcement learning of the first control quantity determination model 171. The control unit 11 of the control device 1 acquires the first state data after the first process executed by the first control quantity output from the first control quantity determination model 171 in FIG. 16 or FIG. 19 by executing a subroutine for the process of acquiring state data (step S151).

[0223] The control unit 11 acquires the second state data after the second process executed by the second control quantity output from the second control quantity determination model 171 in FIG. 16 or FIG. 19 by executing a subroutine for the process of acquiring state data (step S152).

[0224] Based on the rewards corresponding to the acquired first state data and second state data, the control unit 11 performs reinforcement learning on the first control amount determination model 171 in FIG. 16 or FIG. 19 (step S153). Specifically, the reward calculation unit 111 of the control unit 11 calculates the total reward based on the state after the process executed by the control amounts for a plurality of control items before the "temperature adjustment" process and the state after the process executed by the control amounts for a plurality of control items before the "molding" process.

[0225] The control unit 11 sets the first state data acquired by the process of step S141 in FIG. 22 and the second state data acquired by the process of step S144 as the state st, and sets the action information for the first control amount determined by the process of step S143 and the second control amount determined by the process of step S146 as the action at, and performs reinforcement learning based on the reward calculated by the reward calculation unit 111. The control unit 11 ends the process.

[0226] Note that this modified example describes an example of the control amount determination model 171 constructed by reinforcement learning in each process of chocolate production, but it is not limited thereto. For example, it can be similarly applied to the control amount determination model 171 constructed using the causal relationship method in Embodiment 1.

[0227] According to this modified example, it is possible to determine the control amounts for a plurality of control items before the process for each process.

[0228] According to this modified example, it is possible to perform reinforcement learning on the first control amount determination model 171 based on the first state data after the first process executed by the first control amount and the second state data after the second process executed by the second control amount.

[0229] It should be considered that all the embodiments disclosed this time are illustrative in all respects and not restrictive. The scope of the present invention is shown not by the above meaning but by the claims, and it is intended that all modifications within the meaning and scope equivalent to the claims are included.

[0230] The matters described in each embodiment can be combined with each other. Further, the independent claims and dependent claims described in the claims can be combined with each other in all possible combinations regardless of the citation format. Furthermore, although the claims use a format (multi-claim format) in which claims that cite two or more other claims are described, it is not limited to this. A format in which a multi-claim (multi-multi-claim) that cites at least one multi-claim may be described may also be used.

Explanation of Signs

[0231] 1 Information processing apparatus (control apparatus) 11 Control unit 111 Reward calculation unit 112 Action selection unit 113 Action evaluation unit 12 Memory unit 13 Communication unit 14 Input unit 15 Display unit 16 Reading unit 17 Mass storage unit 171 Control amount determination model (first control amount determination model; second control amount determination model) 172 State data DB 173 Control amount reference DB173 174 Threshold value DB174 1a Portable storage medium 1b Semiconductor memory 1P Control program 2 Information processing terminal (administrator terminal) 21 Control unit 22 Memory unit 23 Communication unit 24 Input unit 25 Display unit 2P Control program 3 Manufacturing apparatus 41 Imaging apparatus 42 Weight measurement sensor 43 Temperature sensor 44 Acceleration sensor 45 Pressure sensor

Claims

1. Obtain status data via a sensor that detects the status after a process among a plurality of processes, Determine the control amount for a plurality of control items before the process according to the obtained status data, Adjust the control amount according to the status after the process executed by the determined control amount Information processing method.

2. Obtain image data of an object after the process by an imaging device, Based on the obtained image data, identify the status of the object from a plurality of types of statuses, Identify the status amount corresponding to the identified status, Determine the control amount for a plurality of control items before the process according to the status data including the identified status and status amount The information processing method according to Claim 1.

3. When the status data exceeds a threshold, notify the administrator of the status data, the process corresponding to the status data, and the determined control amount The information processing method according to Claim 1 or 2.

4. Obtain a plurality of the status data in time series, Determine the control amount for a plurality of control items before the process according to the obtained plurality of status data in time series The information processing method according to Claim 1 or 2.

5. When the plurality of status data in time series exceeds a threshold, notify the administrator of the plurality of status data in time series, the processes corresponding to the plurality of status data in time series, and the determined control amount The information processing method according to Claim 4.

6. Input the obtained status data into a reinforcement learning model trained to output the control amount for a plurality of control items before the process, and by inputting the obtained status data, output the control amount for a plurality of control items before the process, Train the reinforcement learning model based on the reward according to the status data after the process executed by the output control amount The information processing method according to Claim 1 or 2.

7. Input the obtained plurality of status data in time series into a reinforcement learning model trained to output the control amount for a plurality of control items before the process, and by inputting the obtained plurality of status data in time series, output the control amount for a plurality of control items before the process, Train the reinforcement learning model based on the reward according to the status data after the process executed by the output control amount The information processing method according to Claim 1 or 2.

8. Obtain first status data via a sensor that detects the first status after the first process, Obtain second status data via a sensor that detects the second status after the second process, By inputting the acquired first state data into a first reinforcement learning model trained to output a first control amount for a plurality of control items before the first step, the first control amount for the plurality of control items before the first step is output. By inputting the acquired second state data into a second reinforcement learning model trained to output a second control amount for a plurality of control items before the second step, the second control amount for the plurality of control items before the second step is output. The information processing method according to claim 1 or 2.

9. Based on the reward corresponding to the first state data after the first step executed by the output first control amount and the second state data after the second step executed by the output second control amount, the first reinforcement learning model is trained. The information processing method according to claim 8.

10. Acquire state data via a sensor that detects the state after the process during a plurality of processes. According to the acquired state data, determine a control amount for a plurality of control items before the process. Adjust the control amount according to the state after the process executed by the determined control amount. A program for causing a computer to execute the process.

11. An information processing apparatus including a control unit, The control unit, Acquire state data via a sensor that detects the state after the process during a plurality of processes. According to the acquired state data, determine a control amount for a plurality of control items before the process. Adjust the control amount according to the state after the process executed by the determined control amount. Information processing apparatus.

12. An information processing system including a sensor that detects the state after the process during a plurality of processes and an information processing apparatus including a control unit, The control unit, Acquire state data via the sensor. According to the acquired state data, determine a control amount for a plurality of control items before the process. Adjust the control amount according to the state after the process executed by the determined control amount. Information processing system.

Images