Question answering device, question answering method, question answering program, question answering system, industrial machine

Abstract

A question answering device 10 for answering questions about an industrial machine is provided with: an information acquisition unit 101 that acquires question information received by a question reception unit; a determination processing unit 102 that determines whether or not the acquired question information includes unique information about the industrial machine; and an answer generation unit 103 that, if it is determined that the question information includes the unique information, generates an answer to the question information on the basis of a closed learning model that has been trained about information that relates to the industrial machine and the unique information and that can be provided only to users of the industrial machine.

Description

Technical Field

[0001] Embodiments of the present invention relate to a technique for answering questions from users, particularly users who handle industrial machines.

Background Art

[0002] User support is widely provided in which questions from users are received by email or phone and answered by an operator. Regarding such user support, an answer system that utilizes AI (Artificial Intelligence) instead of an operator has been rapidly spreading in recent years. As an answer system, there is an AI-type chatbot such as Chat GPT (Chat Generative Pre-trained Transformer) provided by OpenAI, in which a learned AI answers questions in an interactive format. Since such a chatbot can comprehensively use known information, it can generate reasonable answers to general questions from users.

[0003] In related technologies, there is a communication and open-loop / closed-loop control system for at least one filling system, wherein the filling system includes a machine, the machine is configured to recognize voice input and / or text input by an operator, and / or output or display information regarding the operating status of the machine, a software communication robot, particularly a chatbot, and an open-loop / closed-loop control device connected to the software communication robot for data communication and configured to control the machine of the filling system in an open-loop and / or closed-loop manner based on the voice input and / or text input recognized by the software communication robot. (See Patent Document 1 below).

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

[0005] The problem that embodiments of the present invention aim to solve is to provide a technology capable of generating good answers to a variety of questions, including specialized questions from users. Other problems and novel features will become apparent from the description herein and the accompanying drawings. [Means for solving the problem]

[0006] In one embodiment, the question answering device generates an answer to a question from a user of an industrial machine, based on a closed learning model that has learned about the industrial machine and the specific information, and which is only available to the user of the industrial machine. [Brief explanation of the drawing]

[0007] [Figure 1] This is a schematic block diagram showing the configuration of the question answering system according to the first embodiment. [Figure 2] This is a block diagram showing the hardware configuration of a question answering device according to the first embodiment. [Figure 3] This is a block diagram showing the hardware configuration of a control device included in an injection molding machine according to the first embodiment. [Figure 4] This is a block diagram showing the functional configuration of a question answering device according to the first embodiment. [Figure 5] This is a block diagram showing the functional configuration of the control device included in the injection molding machine according to the first embodiment. [Figure 6] This is a flowchart illustrating the question answering process according to the first embodiment. [Figure 7] This is a flowchart illustrating the question answering process according to the first embodiment. [Figure 8]This is a flowchart illustrating the question answering process according to the first embodiment. [Figure 9] This is a diagram illustrating the answer text to the question information according to the first embodiment. [Figure 10] This is a flowchart illustrating the question correction process according to the first embodiment. [Figure 11] This is a diagram illustrating the process of correcting the question information according to the first embodiment. [Figure 12] This is a block diagram showing an application example of the question answering device according to the first embodiment. [Figure 13] This is a block diagram showing the functional configuration of a question answering device according to the second embodiment. [Figure 14] This is a block diagram showing the functional configuration of the control device included in the injection molding machine according to the second embodiment. [Figure 15] This is a flowchart illustrating the question answering process according to the second embodiment. [Modes for carrying out the invention]

[0008] Embodiments of the present invention will be described below with reference to the drawings. However, the invention is not limited to the embodiments described below. For clarity, the following description and drawings have been simplified as appropriate. In each drawing, the same elements are denoted by the same reference numerals, and redundant explanations have been omitted where necessary. In addition, hatching has been omitted in some parts of the drawings to avoid clutter.

[0009] <First Embodiment> (Configuration of the question answering system) The configuration of the question answering system according to this embodiment will now be described. Figure 1 is a schematic block diagram showing the configuration of the question answering system according to this embodiment.

[0010] As shown in FIG. 1, the question-and-answer system 1 for industrial machines according to the present embodiment includes a question-and-answer device 10 and a plurality of injection molding machines 20A to 20N. When not distinguishing the injection molding machines 20A to 20N from each other, they will be described as the injection molding machine 20.

[0011] The question-and-answer device 10 for industrial machines is an information processing device such as a server that is communicably connected to a plurality of injection molding machines 20 via a network NW such as the Internet. The question-and-answer device 10 acquires questions from the injection molding machines 20. Specifically, it acquires questions input by the user of the injection molding machine 20 using a hardware I / F 240 (see FIG. 3) described later, generates answers to the questions, and provides them to the user. The question-and-answer device 10 has an automatic conversation program such as a chatbot that utilizes dialogical AI installed, and generates answers by means of the program. The answers by the question-and-answer device 10 are generated based on a closed learning model 11 and / or an open learning model 12. The closed learning model 11 is a machine learning model learned using the information stored in the closed database 160 as a dataset. Also, the open learning model 12 is a machine learning model learned using the information stored in the open database 170 as a dataset.

[0012] The injection molding machine 20 has a heating cylinder, a screw, a mold clamping device to which a mold is attached, etc. After melting and kneading the resin pellets charged into the hopper connected to the heating cylinder, it injects the molten resin into the mold to obtain a desired resin molded product, which is an industrial machine. The injection molding machine 20 includes a control device 200 that controls the above-described series of device operations. The control device 200 according to the present embodiment further receives questions from the user, appropriately corrects the questions, etc., and transmits them to the question-and-answer device 10. When receiving an answer to the question from the question-and-answer device 10, it notifies the user of the answer.

[0013] (Hardware Configuration) The hardware configurations of the above-described question-and-answer device 10 and the control device 200 will be described in detail. FIG. 2 is a block diagram showing the hardware configuration of the question-and-answer device according to the present embodiment. FIG. 3 is a block diagram showing the hardware configuration of the control device included in the injection molding machine according to the present embodiment.

[0014] As shown in FIG. 2, the question-and-answer device 10 includes a CPU (Central Processing Unit) 110, a ROM (Read Only Memory) 120, a RAM (Random Access Memory) 130, a communication I / F (InterFace) 140, a storage device 150, a closed database 160, and an open database 170.

[0015] The CPU 110 executes by expanding the BIOS (Basic Input / Output System) read from the ROM 120, which is a non-volatile storage area, the OS (Operating System) read from the storage device 150, general-purpose applications, and various programs onto the RAM 130, which is a volatile storage area. As programs, for example, in addition to the above-described automatic conversation program, a question-and-answer program that is part of the question-and-answer processing (question-and-answer method for industrial machines) described later is included. The communication I / F 140 has a gateway or the like (not shown) and is used for communication with the injection molding machine 20 via the network NW.

[0016] The storage device 150 is, for example, an HDD (Hard disk drive) or an SSD (Solid State Drive), and stores various information used for the question-and-answer processing. The information stored includes a list showing the association between the password and the closed learning model 11. The password will be described later. The closed database 160 and the open database 170 are storage areas that contain different information (which may partially overlap) used for answering questions. These databases are constructed by a storage device such as an HDD or an SSD.

[0017] The closed database 160 is a storage area containing information about industrial machinery, in this embodiment, the injection molding machine 20, and information about unique components, and is used as a dataset for the closed learning model 11. The information about the injection molding machine stored in the closed database 160 includes, for example, the contents of the instruction manual for the injection molding machine 20, and information that can only be provided to the user of the injection molding machine 20 (including the company to which the user belongs), in other words, highly confidential information. Therefore, the closed learning model 11, which has been trained based on the closed database 160, is linked to a password (identification information) that uniquely identifies the user and / or the injection molding machine 20 to which the question was asked.

[0018] The PIN code is information that uniquely identifies the injection molding machines 20A to 20N, or in other words, it is used to identify the user (individual or organization such as a company to which an individual belongs) who owns the injection molding machine 20. Therefore, if the same user owns multiple injection molding machines 20 of different types, the PIN codes for each of the multiple injection molding machines 20 may be the same, or the type of injection molding machine 20 may be associated with the PIN code and stored in the storage device 150. Furthermore, the PIN code also serves as information that indicates the user's unique closed learning model 11.

[0019] In this embodiment, a unique closed learning model 11 is provided for each user, and a unique closed database 160 is associated with each closed learning model 11. Therefore, the user indicated by the password will receive an answer based on the closed learning model 11 associated with that user. Consequently, only answers that can be disclosed to the user will be provided, and answers based on other closed learning models 11, i.e., answers that cannot be disclosed to the user, will not be provided.

[0020] Particularly confidential information includes methods for handling each error code, information on the molding conditions (operating conditions) of the injection molding machine 20 for obtaining the desired molded product, specifications of the injection molding machine 20, and actual measured values ​​during molding in the injection molding machine 20. It is preferable that multiple methods for handling each error code be provided, corresponding to the molding conditions of the injection molding machine 20.

[0021] Specifically, the closed-type learning model 11 has learned the relationship between molding condition information, specification information, error codes, information indicating abnormalities, information regarding molded products, and / or measured value information as questions, and molding condition information and / or answers in the instruction manual as answers. Information indicating abnormalities here mainly refers to abnormalities not indicated by error codes, such as questions about unusual noises that the user suspects to be abnormal. Information regarding molded products here mainly refers to questions about not being able to obtain the molded product the user desires. It is also possible to assume that the user prioritizes production, energy saving, or machine life. Therefore, if the question includes what the user prioritizes, it is preferable that the answer pattern (e.g., changes in molding conditions) is linked to the above-mentioned relationship so that molding condition information aligned with those priorities can be included in the answer document.

[0022] The specifications of the injection molding machine include the inner diameter of the heating cylinder, screw type (L / D), servo motor performance (rated output, rated torque, rated rotational speed, etc.), clamping mechanism performance, and the usage history of the injection molding machine. The molding condition information of the injection molding machine 20 includes the heating cylinder control temperature, injection speed during injection, screw rotational speed and back pressure during metering, clamping force and mold opening / closing speed of the clamping device, material type, and material supply amount. The measured value information includes the detection results of various sensors such as temperature sensors and timers on the injection molding machine 20, and includes, for example, the measured temperature of the heating cylinder, molding operation information, injection information such as injection speed, metering information such as metering completion position and screw rotational speed during metering, and clamping information such as clamping force and mold opening / closing speed. For example, if this specification information, molding condition information, and measured value information (operational information) are acquired as a question along with an error code, the question answering device 10 can accurately answer how to respond to the error code corresponding to the specification information, molding condition information, and measured value information by referring to the closed learning model 11. Preferably, the response method includes information such as how to change the values ​​of the molding conditions. By providing such a response method, it becomes possible to provide an accurate answer that is appropriate to the user's current situation.

[0023] Unique information is information used to determine whether to answer a question using a closed learning model 11. In this embodiment, this includes error codes, specifications of the injection molding machine 20, molding condition information, and measured value information. Since there are multiple molding condition information, it may be limited to specific molding conditions. However, it is not limited to error codes, specifications, molding condition information, and measured value information; any information suitable for generating an answer using the closed learning model 11 is acceptable. For example, the type (model) of the injection molding machine 20, wording related to the instruction manual, such as the word "instruction manual" or symbols described in the instruction manual may be considered unique information.

[0024] The closed learning model 11 learns to improve the accuracy of responses through machine learning methods such as reinforcement learning, supervised learning, deep learning, and additional learning, based on user evaluations obtained after a response. Preferably, the closed database 160 is also updated each time as a result of this learning. It is desirable that not only the evaluation but also the process leading to the question and answer be fed back into the learning process. Therefore, the closed learning model 11 and the closed database 160 are individually customized for each user. Furthermore, because the closed learning model 11 and the closed database 160 are individually associated with each user, the results of the different learning processes performed for each user are not shared among users. Therefore, the leakage of information within the closed database 160 can be prevented.

[0025] On the other hand, in this embodiment, the open database 170 is a storage area containing less confidential information related to the injection molding machine, in other words, publicly known information that can be provided not only to the user of the injection molding machine 20 but also to various other users. The open database 170 is used as the dataset for the open learning model 12. Therefore, the open learning model 12 has learned the relationships of publicly known information and is used when a publicly known question is asked. The open database 170 can be, for example, a database used in chat GPT provided by Open AI.

[0026] It is preferable that information be added to the closed database 160 and the open database 170 as appropriate. In the open database 170, the added information is publicly known information, whereas in the closed database 160, information related to industrial machinery (in this case, injection molding machines 20) is added. For example, the closed database 160 is automatically updated by the question answering device 10 periodically accessing external information and collecting equipment information from industrial machinery manufacturers via the network NW. This is preferably done not only for equipment manufacturers but also when new knowledge about parts manufacturers or industrial machinery is discovered. For example, when new knowledge is discovered, the maintenance and operation company of the question answering device 10 can add the new knowledge to the closed database 160 as appropriate. Alternatively, the maintenance and operation company may access a dedicated server for collecting new knowledge to automatically update the closed database 160. When information is added to the closed database 160 and the open database 170 in this way, the closed learning model 11 and the open learning model 12 are retrained.

[0027] As shown in Figure 3, the control device 200 includes a CPU 210, ROM 220, RAM 230, hardware I / F 240, a user interface consisting of a display 241, an input device 242, a microphone 243, and a speaker 244, a communication I / F 250, and a storage device 260.

[0028] The CPU 210 executes the BIOS read from the non-volatile memory area ROM 220, the OS read from the storage device 260, general-purpose applications, and various programs by loading them onto the volatile memory area RAM 230. These programs include those that handle part of the question answering process, which will be described later.

[0029] The hardware interface 240 is used for communication with user interfaces such as the display 241, input device 242, microphone 243, and speaker 244. Specifically, the hardware interface 240 is used to receive questions from the user and to notify the user of the answers. The display 241 displays the questions and their answers as text. The input device 242 includes a keyboard and mouse and accepts user input. The microphone 243 accepts questions from the user via voice. The speaker 244 provides answers to questions via voice.

[0030] The user interface of these injection molding machines 20 is a question receiving unit according to the present invention that receives questions from the user via voice or key input. The user interface is also a question answering unit according to the present invention that conveys the answers generated by the answer generation unit 103 of the question answering device 10 (described later) to the user as voice from the display 241 or speaker 244. Furthermore, the user interface of the injection molding machine 20 contributes to solving the problems of the invention. In this embodiment, the user interface, including the display 241, input device 242, microphone 243, and speaker 244 that constitute the question receiving unit and question answering unit of the question answering device according to the present invention, is provided in each injection molding machine 20. However, the user interface may be shared among multiple injection molding machines 20 as a portable device. A user interface dedicated to an injection molding machine 20, which is attached to the injection molding machine 20 and equipped with a portable question receiving unit and question answering unit, can be considered as part of the injection molding machine 20. Examples of such a portable user interface include tablet terminals such as smartphones and notebook PCs (Personal Computers). Furthermore, a desktop PC or similar device installed in the user's factory may be equipped with both a question reception unit and a question answering unit.

[0031] The communication interface 250 has a gateway (not shown) and is used for communication with the question answering device 10 via the network NW. The storage device 260 is, for example, an HDD (Hard disk drive) or SSD (Solid State Drive) and stores various information used in the question answering process. The information stored includes a password, additional stored information, and answer documents to questions sent from the question answering device 10. The additional stored information is information added in the question correction judgment process, which will be described later, and its details will be described later.

[0032] (Functional Configuration) The functional configurations of the question answering device 10 and the control device 200 will be described in detail. Figure 4 is a block diagram showing the functional configuration of the question answering device according to this embodiment. Figure 5 is a block diagram showing the functional configuration of the control device provided in the injection molding machine according to this embodiment.

[0033] As shown in Figure 4, the question answering device 10 includes an information acquisition unit 101, a judgment processing unit 102, an answer generation unit 103, a verification processing unit 104, an answer transmission unit 105, and a learning update unit 106. These functions are realized through the cooperation of the hardware of the question answering device 10 described above.

[0034] The information acquisition unit 101 acquires various information transmitted from the injection molding machine 20. The information acquired includes question information indicating the user's question, and evaluation information indicating the user's evaluation of the answer to the question. The judgment processing unit 102 performs various judgments in the answer generation process. These judgments include, for example, whether or not there is unique information in the question information, and whether or not it is necessary to correct the question information.

[0035] The answer generation unit 103 is a so-called AI chatbot that generates answers to question information based on a closed learning model 11 and / or an open learning model 12. Note that it is not limited to an AI chatbot; any automated response (automated conversation) program capable of generating answers to questions using each learning model is acceptable. The verification processing unit 104 verifies the generated answers to determine whether or not they should be modified.

[0036] The response transmission unit 105 transmits the generated and, if modified, response to the user interface installed on the injection molding machine 20 that sent the question information. The learning update unit 106, when it acquires information for machine learning, performs machine learning based on that information and updates the closed learning model 11 and the closed database 160. Information for machine learning includes evaluation information and condition change information. Condition change information includes measured values ​​obtained when settings are changed from the settings included in the response in the injection molding machine 20 that received the response to the question, and the changed settings themselves.

[0037] As shown in Figure 5, the control device 200 includes the following functions: an information acquisition unit 201, a conversion processing unit 202, an information transmission unit 203, a determination processing unit 204, a notification unit 205, a monitoring processing unit 206, and a question correction unit 207. These functions are realized through the cooperation of the aforementioned hardware components of the control device 200.

[0038] The information acquisition unit 201 acquires various information from user input and from the question answering device 10. User input includes questions and evaluations. Information to be acquired includes PIN codes and answers generated by the question answering device 10. The conversion processing unit 202 converts the voice-inputted questions into text.

[0039] The information transmission unit 203 transmits question information and evaluation information to the question answering device 10. The judgment processing unit 204 performs various judgments in the answer generation process. These judgments include, for example, whether evaluation information has been input or whether the monitoring process is ON or OFF. The monitoring process monitors the behavior of the injection molding machine 20 after an answer to the question has been obtained, and collects information such as various settings and detection results obtained from various sensors. The notification unit 205 notifies the user of the answer to the question by voice or text. The monitoring processing unit 206 executes the monitoring process. The question correction unit 207 corrects the question information in the question correction judgment process described later.

[0040] (Question and answer processing) The question answering process by the question answering system according to this embodiment will be described in detail. Figures 6 to 8 are flowcharts of the question answering process according to this embodiment. This process is triggered when a question button is displayed on the display 241 (question receiving unit) of the injection molding machine 20, on various setting screens for injection molding, etc., and the button is selected. However, this is not the only way; for example, the control device 200 may constantly collect sound using the microphone 243 (question receiving unit), and this process may be executed when a specific voice command is received.

[0041] First, as shown in Figure 6, the information acquisition unit 201 acquires the user's voice question as voice information (S101). Specifically, the information acquisition unit 201 collects the voice emitted by the person using the microphone 243, and when a certain period of time has elapsed since the start of voice collection and the voice has stopped, it acquires the collected voice as voice information indicating the question. If the voice does not stop after a predetermined collection time has elapsed since the start of voice collection, or if the voice cannot be picked up, the notification unit 205 may use the speaker 244 to announce the error audibly or use the display 241 to display text, etc., to inform the user. In this case, the question answering process is terminated.

[0042] After acquiring the audio information, the conversion processing unit 202 converts the acquired audio information into text and generates question information (S102). The process of converting audio to text can be performed using existing speech-to-text conversion programs, such as AI-based speech recognition. It is also advisable to modify the question information into a sentence-like structure using existing natural language processing techniques. The question may also be acquired as text input by the user using the input device 242. In this case, the input text can be used directly as the question information. After generating the question information, the information acquisition unit 201 acquires the PIN code stored in the storage device 260 (S103).

[0043] After obtaining the PIN code, a determination is made as to whether or not the generated question information needs to be corrected, and if necessary, a question correction determination process is executed to perform the correction (S104). Details of the question correction determination process will be described later. After the question correction determination process, the information transmission unit 203 transmits the question information along with the PIN code to the question answering device 10 (S105), and the control device 200 enters a standby state during the question answering process.

[0044] The question answering device 10 is always in a standby state where it can acquire various information. When question information and a password are transmitted in this state, the information acquisition unit 101 of the question answering device 10 acquires (receives) them (S106). After acquisition, the determination processing unit 102 determines whether or not unique information is included in the question information (S107).

[0045] If it is determined that unique information is included in the question information (S107, YES), the information acquisition unit 101 acquires the password received along with the question information (S108). After acquisition, the answer generation unit 103 reads the list stored in the storage device 150 and selects a closed learning model 11 corresponding to the acquired password (S109). After selection, the answer generation unit 103 generates an answer document corresponding to the question information based on the selected closed learning model 11 (S110).

[0046] The response document generated using the closed learning model 11 preferably includes, for example, a more detailed explanation of countermeasures for error codes, particularly those described in the instruction manual, if the question information includes an error code. Furthermore, if the question information includes molding condition information, measured values, or specifications, the response generated will be derived from those conditions and measured values. Such a response may include appropriate changes to the molding conditions.

[0047] In particular, if an error code is included in the question information, it is preferable to provide multiple values ​​for modifying the molding conditions that can avoid the abnormal condition indicated by the error code. Furthermore, if the above-mentioned priorities are included in the question information, molding conditions that align with those priorities should be included in the response document. Even if such priorities are not included in the question document, multiple response patterns may be provided for each. Moreover, it is preferable to provide candidate molding conditions in order of likelihood of not causing problems with the operation of the injection molding machine.

[0048] On the other hand, if it is determined that no unique information is included in the question information (S107, NO), the answer generation unit 103 selects the open learning model 12 (S111), and in step S110, an answer document corresponding to the question information is generated based on the open learning model 12. If no unique information is included, it can be determined that a general question with low confidentiality has been asked, and a sufficient answer can be obtained from the open learning model 12. Even if there are unclear parts in the question that are expected to provide an answer that fully satisfies the user, the question information is corrected so that a highly accurate answer that satisfies the user can be obtained through the question correction judgment process, which will be described in detail later.

[0049] As shown in Figure 7, after the question document is generated, the determination processing unit 102 determines whether the generated answer document is verifiable (S112). This determination is made based on whether the answer document contains enough molding conditions to enable verification processing. Verification processing is a so-called simulation process using the molding conditions. If the molding conditions necessary for the simulation process are included in the answer document, it is determined that the generated answer document is verifiable. In addition to determining whether the answer document contains enough molding conditions to enable verification processing, it is also possible to determine whether the question information corresponding to the answer document contains specification information that enables verification processing. In this case, if both determination results are positive, it is determined that verification processing is possible, and simulation processing using the molding conditions and specification information is performed.

[0050] If the generated response document is verifiable (S112, YES), the verification processing unit 104 executes the verification process (S113) and obtains the verification result. After the verification result, the judgment processing unit 102 determines whether or not the response document needs to be corrected (S114). This determination is made based on whether or not there is a problem with the verification result, such as the molded product not being of generally good quality, or an error occurring (e.g., inconsistency in the changed values ​​of the molding conditions, physical damage to the equipment, etc.). Information on the quality of the molded product may be stored in the storage device 150 in advance as data used in the verification process, or it may be obtained by a closed-type learning model 11 or an open-type learning model 12. The response document may also contain information on the quality of the molded product. Furthermore, it is preferable that the obtained verification result is used as training data for the closed-type learning model 11 at a predetermined timing. At this time, the weights of the neural network of the closed-type learning model 11 may be adjusted by a method such as backpropagation.

[0051] If it is determined that the response document needs to be corrected (S114, YES), the response generation unit 103 corrects the response document (S115), and the verification process in step S113 is performed again. Preferably, the corrections here are determined according to the verification results, i.e., the changes to the molding conditions. The correspondence between the verification results and the changes may be stored in the storage device 150 in advance. In addition, the priority order of the types of molding conditions to be changed may be determined in advance. This priority order may be changed according to the type of resin and the type of injection molding machine 20. Furthermore, based on the question information including the verification results, the answer may be regenerated again using the closed learning model 11 or the open learning model 12.

[0052] On the other hand, if it is determined that the response document does not require any correction (S114, NO), or if it is determined that the generated response document is not verifiable (S112, NO), the response transmission unit 105 transmits the response document to the injection molding machine 20 that sent the question information (S116). After transmission, the question answering device 10 enters a standby state. If the transmitted response document was generated using the closed learning model 11, it is preferable that it be stored in the storage device 150 along with the password. This is because the response document will be used for training the closed learning model 11 later.

[0053] After transmission, the information acquisition unit 201 of the injection molding machine 20 acquires (receives) the response document (S117). After acquisition, the notification unit 205 notifies the user of the response document (S118). This notification may include displaying the response document using the display 241 or broadcasting the response document as an audio using the speaker 244. Therefore, user interfaces such as the display 241 and speaker 244 contribute to solving the problems of the present invention.

[0054] After notification, the determination processing unit 204 determines whether or not the user has a follow-up question (S119). This determination is made, for example, by displaying a window on the display 241 as a GUI (Graphical User Interface) that shows whether or not there is a follow-up question in text and has selectable "YES" and "NO" buttons, and determining whether or not each button is selected. Alternatively, the speaker 244 may be used to broadcast an audio message indicating whether or not there is a follow-up question, and the determination may be made by using the microphone 243 to determine whether or not the user's voice response (for example, an audio message similar to "yes" or "no," or a follow-up question) is obtained within a predetermined period.

[0055] If it is determined that the user has a follow-up question (S119, YES), the voice question input processing in step S101 shown in Figure 6 is performed. On the other hand, if it is determined that the user has no follow-up questions (S119, NO), the determination processing unit 204 determines whether or not an evaluation has been made by the user (S120), as shown in Figure 8. This determination is made based on whether or not an evaluation has been acquired by the information acquisition unit 201 within a predetermined time after the determination in step S119. For example, a window with an "Evaluate" button is displayed on the display 241, and depending on the selection of this button, a window with "Resolved" (good evaluation) buttons and "Not Resolved" (bad evaluation) buttons is displayed on the display 241. Depending on the selection of these "Resolved" or "Not Resolved" buttons, an evaluation is acquired, and it is determined that an evaluation has been made.

[0056] Furthermore, after selecting the "Evaluate" button, the system may acquire the evaluation from the text or voice input. It may also be possible to provide a multi-level evaluation, such as a 5-point scale, rather than simply "Resolved" or "Not Resolved." The predetermined time for this evaluation is preferably set to allow sufficient time for the user to recognize the evaluation result by operating the injection molding machine 20 according to the response document. Such a time could be, for example, several minutes to several hours. The time available for evaluation varies depending on the content of the question. Therefore, it is preferable to store the acquired response document in the storage device 260 so that the evaluation can be performed at any time. In this case, it is preferable to separate steps S120 and S121, and the subsequent steps S126 to S129 (described later), from the question answering process and make them executable at any time as evaluation processing.

[0057] If it is determined that an evaluation has been made by the user (S120, YES), the information transmission unit 203 transmits evaluation information indicating the evaluation result along with a password to the question answering device 10 (S121), and proceeds to step S122. On the other hand, if it is determined that an evaluation has not been made by the user (S120, NO), the processing in step S121 is not performed and the system proceeds to step S122.

[0058] In step S122, the determination processing unit 204 determines whether the monitoring process is set to ON (S122). The monitoring process is a process performed by the monitoring processing unit 206 to monitor the operation of the injection molding machine 20, and collects molding conditions and measured values ​​during operation (injection molding). The monitoring process can be switched ON / OFF using the selectable "ON" and "OFF" buttons included in the screen displayed on the display 241 while the injection molding machine 20 is in operation.

[0059] If it is determined that the monitoring process is set to ON (S122, YES), the determination processing unit 204 determines whether the molding conditions have been changed by the user (S123). This determination is made based on whether the molding conditions were changed within a predetermined change determination time after obtaining the response document. If it is determined that the molding conditions have been changed by the user (S123, YES), the determination processing unit 204 determines whether the changed molding conditions are OK, that is, whether there were any problems with the molding conditions (S124). This determination is made based on whether molding was performed a predetermined number of times consecutively with the changed molding conditions. Alternatively, the user may input that the molding conditions are OK by selecting a button or the like, and the determination processing unit 204 may make a determination based on whether such input has been made.

[0060] If the modified molding conditions are determined to be NG, i.e., there was a problem with the molding conditions (S124, NO), the user is expected to change the molding conditions again independently. Therefore, the determination processing unit 204 performs the determination process in step S124 again, and this determination is repeated until it is determined that there was no problem with the molding conditions. Since the monitoring process is continuing, the determination process in step S124 may be performed again triggered by a further change in the molding conditions. Also, at this stage, the user may ask another question. If a further question is received, this process should be interrupted or terminated and the question information should be generated and transmitted. In this case, it is preferable that the question answering device 10 manages the received question by linking it to the previous question.

[0061] If the modified molding conditions are determined to be OK (S124, YES), the information transmission unit 203 transmits the modified condition information, including the modified molding conditions and the password, to the question answering device 10 (S125), and the question answering process on the injection molding machine 20 side is terminated. Preferably, the modified condition information includes all information that can be collected in the monitoring process along with the modified conditions. On the other hand, if it is determined that the monitoring process is set to OFF (S122, NO), or if it is determined that the molding conditions have not been changed by the user (S123, NO), the question answering process on the injection molding machine 20 side is terminated. Note that if it is determined that the molding conditions have not been changed by the user, it means that the elapsed time since the acquisition of the answer document has exceeded the modification determination time described above.

[0062] When evaluation information and a password are transmitted to the question answering device 10, the information acquisition unit 101 of the question answering device 10 acquires (receives) the evaluation information and also acquires the password (S126, S127). After acquisition, the determination processing unit 102 determines whether or not the answer document corresponding to the evaluation information was generated using the closed learning model 11 (S128). The storage device 150 stores the password together with the answer document. Therefore, it is possible to determine from the password acquired along with the evaluation information whether or not the answer document corresponding to the evaluation information was generated using the closed learning model 11.

[0063] If it is determined that the answer document corresponding to the evaluation information is based on the closed learning model 11 (S128, YES), the learning update unit 106 updates the closed learning model 11 based on the evaluation information (S129). The update here may include, for example, additional learning for the closed learning model 11 based on the evaluation information and the answer document. In this case, it is preferable that the closed database 160 is also updated. After the update, the question answering device 10 enters a standby state (S130). On the other hand, if it is determined that the answer document corresponding to the evaluation information is not based on the closed database 160, such as when the answer document is not stored in the storage device 150 (S128, NO), the device proceeds to the standby state of step S130.

[0064] When the condition change information and the password are transmitted to the question answering device 10, the information acquisition unit 101 of the question answering device 10 acquires (receives) the condition change information and the password (S131, S132). After acquisition, the determination processing unit 102 determines whether or not the answer document corresponding to the condition change information was generated using the closed learning model 11 (S133). The storage device 150 stores the password together with the answer document. Therefore, it is possible to determine from the password acquired together with the condition change information whether or not the answer document corresponding to the condition change information was generated using the closed learning model 11.

[0065] If it is determined that the response document corresponding to the condition change information uses the closed learning model 11 (S133, YES), the learning update unit 106 updates the closed learning model 11 based on the condition change information (S134). The update here may include, for example, additional learning for the closed learning model 11 based on the condition change information and the response document. In this case, it is preferable that the closed database 160 is also updated. After the update, the question answering process on the question answering device 10 side is terminated. On the other hand, if it is determined that the response document corresponding to the condition change information does not use the closed learning model 11, for example, if the response document is not stored in the storage device 150 (S133, NO), the question answering process on the question answering device 10 side is terminated.

[0066] An example of a response document to question information is described below. Figure 9 is a diagram illustrating a response document to question information according to this embodiment. In Figure 9, the code Qu represents the question information, and the codes An1 to An3 represent different response documents to the question information Qu. The question information Qu includes unique information such as error code A1, specification information, molding condition information, and measured value information. Therefore, for such question information Qu, a closed learning model 11 is used to generate the response document. In this example, response documents An1 to An3 are generated because the information that is withheld in the question information Qu is different.

[0067] According to response document An1, an appropriate solution, including molding conditions, is provided for the servo motor overload anomaly, which corresponds to error code A1. According to response document An2, it is indicated that the user is forcing the injection molding machine 20, which is incompatible with the overload anomaly in use, as shown in the question document, and an appropriate response is provided. Furthermore, according to response document An2, the user's PIN code allows the system to identify the injection molding machine 20 owned by the user. Therefore, the response includes the fact that the problem would not occur with a different type of injection molding machine 20 owned by the user. In addition, response document An3 indicates that the injection molding machine 20 requires inspection. These responses can be said to be accurate and appropriate because each specific piece of information, especially the specifications, molding conditions, and measured values, is included in the question information Qu. However, even if the question information Qu does not include specifications, molding conditions, or measured values, the question answering device 10 can still provide an answer by referring to at least the contents described in the instruction manual. Therefore, an accurate answer can be obtained, and the answer can be obtained much faster than the time it would take the user to refer to the instruction manual themselves.

[0068] The question correction and judgment process will be explained in detail. User questions may be subjective, and the relationship between subjects and predicates may be unclear. In such cases, even if the user's question is directly fed into the closed learning model 11 of the question answering device 10, an appropriate answer may not be obtained. In such cases, the question is reorganized using an open learning model with high processing power for language models, such as ChatGPT, or it is processed by a dedicated conversion processing unit 202 before being fed into the closed learning model 11. These processes are also called prompt engineering. Figure 10 is a flowchart of the question correction and judgment process according to this embodiment.

[0069] As shown in Figure 10, first, the judgment processing unit 204 of the injection molding machine 20 determines whether additional storage information is required (S201). Additional storage information here includes explanatory text that explains various error codes in detail, specification information of the injection molding machine 20 that is the source of the inquiry, such as the type, model, and product name of the injection molding machine 20, explanations of technical terms related to the injection molding machine 20, and explanations of the instruction manual for the error codes. Therefore, if the inquiry information includes an error code, technical terms related to the injection molding machine 20, or technical terms and symbols listed in the instruction manual, or if basic information about the injection molding machine 20 is not included, it is determined that additional storage information is required.

[0070] If it is determined that additional storage information is required (S201, YES), the question correction unit 207 obtains the necessary information from the storage device 260 (S202) and corrects the question information (S203). For example, the part "The model number of this injection molding machine is ○○..." in the third paragraph from the top of the question information Qu shown in Figure 9 is the corrected specification information as additional storage information. In other words, this part is a sentence that was automatically added in response to the user's verbal question, and it is desirable that it be replaced with linguistic information like this sentence and fed into a conversational closed learning model. After correction, the determination processing unit 204 determines whether molding conditions and / or measured values ​​are required in the question information (S204). This determination is made by determining whether a specific keyword is included in the question information. Examples of keywords include error codes. It is preferable to include all error codes, but it is also acceptable to include only some error codes. Examples of some error codes include those that result in a very wide range of answer choices when molding conditions or measured values ​​are not available, or those for which it is difficult to obtain a highly accurate answer.

[0071] If it is determined that molding conditions and / or measured values ​​are required in the question information (S204, YES), the determination processing unit 204 determines whether the setting allows the molding conditions and / or measured values ​​to be transmitted to the question answering device 10 (S205). The transmission of molding conditions and / or measured values ​​to an external device often involves confidential company information. Therefore, even when transmitting information to the question answering device 10, where the information is managed confidentially, it is preferable that the transmission setting can be switched ON / OFF. For example, in order to change the settings of the injection molding machine 20, it is preferable to display a selectable toggle button on the setting screen displayed on the display 241. The toggle button may be provided to be switchable individually for each molding condition and for each measured value. Depending on the selection of these buttons, the transmission setting can be switched ON / OFF. Therefore, the determination in step S205 is made by determining whether the transmission setting is set to ON or OFF.

[0072] If it is determined that the molding conditions and / or measured values ​​are set to be transmittable to the question answering device 10 (S205, YES), the question correction unit 207 acquires the current molding conditions and measured values ​​that are set to be transmittable (S206). After acquisition, the question correction unit 207 adds the acquired molding conditions and / or measured values ​​to the question information (S207), and the question correction determination process ends. For example, the part "The operating status of the injection molding machine before the question was asked..." in the fourth paragraph from the top of the question information Qu shown in Figure 9 is the added and corrected molding conditions and measured values. It is desirable that this part be replaced with linguistic information as shown in Figure 9 and fed into a conversational closed learning model.

[0073] On the other hand, if it is determined that the molding conditions and / or measured values ​​are not set to be transmittable to the question answering device 10 (S205, NO), or if it is determined that the molding conditions and / or measured values ​​are not required in the question information (S204, NO), the question correction determination process ends. Also, if it is determined that additional saved information is not required (S201, NO), the process proceeds to the determination process in step S204.

[0074] Figure 11 is a diagram illustrating the process of correcting question information according to this embodiment. The symbols Qu1 to Qu3 shown in Figure 11 each represent question information, and the correction progresses as the last number increases. Question information Qu1 is the uncorrected initial question information, and although the error code is known here, the specific error cannot be determined. Therefore, if the error code is not set in the unique information, the answer will be obtained using the open database 170, but there is a high possibility that no answer will be obtained or an incorrect answer will be obtained.

[0075] On the other hand, question information Qu2 has the content of the error code added as saved additional information through the question correction judgment process. In this state, even if the answer is provided using the open database 170, a reasonable answer can be obtained, but the likelihood of obtaining an answer that satisfies the user is low. In contrast, question information Qu3 has the content of the instruction manual corresponding to the error code added as saved additional information through the question correction judgment process. In this state, even if the answer is provided using the open learning model 12, it can be said that an answer that satisfies the user to some extent can be obtained. Furthermore, when molding conditions and measured values ​​are added to the question information through the question correction judgment process, it becomes the question information Qu shown in Figure 9. In this state, a highly accurate answer that satisfies the user can be obtained.

[0076] Furthermore, if the error code is already included in the question information and the error code is set in the unique information, the closed learning model 11 has already learned the error code information and the instruction manual information (stored in the closed database 160), so even if the question information Qu1 is received, it can generate an answer equivalent to that of receiving question information Qu3 without correction. Therefore, the ability to correct molding conditions and / or measured values ​​is particularly useful in the question correction judgment process.

[0077] According to the embodiment described above, it is possible to provide good answers to the user's specialized questions. Furthermore, compared to when the user refers to the instruction manual themselves, answers can be obtained extremely quickly. In addition, for questions including molding conditions and measured values, answers are generated using a closed learning model 11 that has already learned specialized knowledge, resulting in more specific and accurate answers. Moreover, since the closed learning model 11 is prepared specifically for each user, it is possible to prevent the leakage of confidential information to external parties, such as the disclosure of know-how to other users. Similarly, since feedback on questions, answers, and evaluations of answers is provided only to the closed learning model 11, the accuracy of the learning model's answers can be constantly improved while preventing the leakage of confidential information to external parties.

[0078] Furthermore, the generated answers undergo verification processing whenever possible, thus preventing the sending of incorrect answers to users. Questions are also subject to appropriate question correction and evaluation processing, allowing for enrichment of question content to obtain highly accurate answers.

[0079] Furthermore, users can ask questions verbally and receive answers in audio or text format, making it extremely convenient and easy to obtain answers as if a skilled technician were right beside them.

[0080] In this embodiment, a question-answering system 1 that provides answers to questions from users of the injection molding machine 20 has been described. However, the user is not limited to users of the injection molding machine 20. For example, it can target users of any industrial machinery, such as mixing equipment, extruders, presses, and semiconductor manufacturing equipment. Furthermore, the aforementioned industrial machinery such as the injection molding machine 20 also includes its peripheral equipment. Plant equipment including at least one or more devices such as a mixing equipment, an extruder, and a press is also included in the industrial machinery according to the present invention. In that case, naturally, the closed database 160 of the question-answering device 10 of the question-answering system 1 stores information on various industrial machines, and the closed learning model 11 is trained based on the information on various industrial machines.

[0081] Furthermore, in this embodiment, the question correction determination process is described as being performed in the control device 200 of the injection molding machine 20. However, the question correction determination process may also be performed in the question answering device 10. In this case, the question answering device 10 is provided with a question correction unit 207 as a function, and additional storage information is pre-stored in the storage device 150. Also, in step S205, the question answering device 10 confirms whether it is possible to transmit molding conditions and / or measured values ​​to the injection molding machine 20 that is the source of the question. After that, if it is possible to transmit molding conditions and / or measured values, the molding conditions and / or measured values ​​are transmitted to the question answering device 10 as a response from the injection molding machine 20, and the question answering device 10 acquires them. The question correction unit 207 provided in such a question answering device 10 constitutes a part of the question answering device according to the present invention. Furthermore, the question correction unit 207 may be provided in the intermediate part connecting the information acquisition unit 101 and the determination processing unit 102 of the question answering device 10 in Figure 5.

[0082] Furthermore, in this embodiment, it has been explained that once a question is received, the question information is generated, corrected, and then transmitted to the question answering device 10. However, depending on the content of the question, the user may be asked a question back to enrich the question information. For example, if the question "An abnormality has occurred during weighing, what should I do?" is received, the determination processing unit 204 will determine from the word "abnormality" whether or not an error code is included. If it is determined that no error code is included, the notification unit 205 may notify the user with a message such as, "Is an abnormality number displayed on the screen? If an abnormality number is displayed, please tell us the number (X digits)." This makes it possible to include an error code in the question. This may also be used to prompt the user to answer questions about other conditions of the injection molding machine 20. Therefore, the part of the question correction unit 207 of the question answering device that returns a question to the user to enrich the question information and sends it to the answer generation unit of the question answering device corresponds to the prompt engineering unit.

[0083] Furthermore, although this embodiment describes the question answering device 10 as having an open learning model 12 and an open database 170, it may also have only a closed learning model 11 and a closed database 160 without these. In other words, the closed database 160 may contain various known information so that general questions can also be answered by the closed learning model 11; in other words, the open database 170 may be incorporated into the closed database 160.

[0084] In this embodiment, the question correction judgment process requires that the molding conditions and measured values ​​be set to be transmittable to the question answering device 10 in order to acquire them. However, the molding conditions and measured values ​​may be acquired automatically at the same time as the operator's verbal questions. Also, as described above, if the question answering device 10 has a question correction unit 207 and the question correction judgment process is executed on the question answering device 10 side, it is preferable that a process be performed to confirm the molding conditions and measured values ​​of the injection molding machine 20 after acquiring the question information. For example, the question correction unit 207 of the question answering device 10 requests the injection molding machine 20 for molding conditions and measured values, and the injection molding machine 20, upon receiving the request, transmits the molding conditions and measured values ​​set at that time to the question answering device 10.

[0085] Furthermore, as shown in Figure 12, the question answering device 10 may consist of a first question answering device 10-1 comprising a closed database 160 and hardware such as a CPU 110, and a second question answering device 10-2 comprising an open database 170 and hardware such as a CPU 110, both installed in separate locations and connected to each other via a communication interface such as the Internet. For example, a chatbot such as ChatGPT, which is the open database 170, may be stored on a server of an operating company that manages such conversational AI, and this server may function as the second question answering device 10-2. On the other hand, the first question answering device 10-1, which comprises a server comprising a closed database 160 and hardware such as a CPU 110, may be located at the manufacturer that produces the injection molding machine 20. Alternatively, the first question answering device 10-1, which comprises a server comprising a closed database 160 and hardware such as a CPU 110, may be located at the factory of a user that operates the injection molding machine 20.

[0086] Furthermore, the functions of the information acquisition unit 101, judgment processing unit 102, answer generation unit 103, verification processing unit 104, answer transmission unit 105, and learning update unit 106 of the question answering device 10, as well as each learning model, may be incorporated into the injection molding machine 20. In this case, the various functions of the question answering device 10 described above will be realized through the cooperation of the hardware resources of the injection molding machine 20, and the injection molding machine 20 will be constructed as a question answering device according to the present invention. Of course, only some functions of the question answering device 10, such as only the first question answering device 10-1, which constitutes the closed database 160 and hardware such as the CPU 110, may be incorporated into the injection molding machine 20.

[0087] <Second Embodiment> Figure 13 is a block diagram showing the functional configuration of the question answering device according to this embodiment. Figure 14 is a block diagram showing the functional configuration of the control device provided in the injection molding machine according to this embodiment.

[0088] This embodiment differs from the first embodiment in that the question answering system 1 has a question answering device 10A instead of the question answering device 10. Furthermore, it differs from the first embodiment in that the injection molding machine 20 is equipped with a control device 200A instead of the control device 200.

[0089] As shown in Figure 13, the question answering device 10A differs from the question answering device 10 according to the first embodiment in that it includes an open-type answer generation unit 103A and a closed-type answer generation unit 103B instead of the answer generation unit 103. As shown in Figure 14, the control device 200A differs from the control device 200 according to the first embodiment in that it further includes an information extraction unit 208.

[0090] In the question answering device 10A, the open-type answer generation unit 103A is an AI chatbot that generates answers to publicly known question information, described later, based on the open-type learning model 12. On the other hand, the closed-type answer generation unit 103B is an AI chatbot that generates answers to question information based on the closed-type learning model 11. The closed-type answer generation unit 103B also integrates the two obtained answers and generates an answer document to be sent to the injection molding machine 20.

[0091] The information extraction unit 208 in the control device 200A extracts publicly known information from the question information and generates publicly known question information.

[0092] (Question and answer processing) The question answering process by the question answering system according to this embodiment will be described in detail. Figure 15 is a flowchart of the question answering process according to this embodiment. The question answering process according to this embodiment has some additions and modifications from the question answering process according to the first embodiment. Here, only the added and modified processes will be highlighted and explained.

[0093] As shown in Figure 15, after the question correction judgment process, the information extraction unit 208 of the control device 200A generates publicly known question information by extracting publicly known information from the question information (S301). Here, publicly known information is information obtained by subtracting unique information from the question information. In other words, the part of the question information from which highly confidential information such as error codes, molding condition information, measured value information, and specification information has been removed is extracted as publicly known information. In addition, according to the question correction judgment process, a detailed explanation is added to the error code in the question information. It is preferable that this detailed explanation is treated as publicly known information, and that "(error code A1)" etc. included in the question information Qu2 shown in Figure 11 are not treated as publicly known information and are not extracted. After the generation of publicly known question information, the information transmission unit 203 transmits the question information and the publicly known question information together with the password to the question answering device 10A (S302).

[0094] The question answering device 10A is always in a standby state where it can acquire various information. When question information, publicly known question information, and a password code are transmitted in this state, the information acquisition unit 101 of the question answering device 10 acquires (receives) them (S303). After acquisition, the open-type answer generation unit 103A generates an open-type answer document (first answer) corresponding to the publicly known question information based on the open-type learning model 12 (S304). After selecting the closed-type learning model 11, the closed-type answer generation unit 103B generates a closed-type answer document (second answer) corresponding to the question information based on the closed-type learning model 11 (S305). After generating the closed-type answer document, the closed-type answer generation unit 103B integrates the obtained open-type answer document and closed-type answer document to generate a single answer document (S306). The question information from the closed-type learning model 11 may be in the form of closed-type answer data for question data, not only in the form of a closed-type answer document (second answer) corresponding to conversational question information. In the above, "question data" refers to measured value information (including historical information) and questions that are not in a conversational format. "Answer data" refers to numerical values ​​such as molding conditions and answers that are not in a conversational format. In the above case as well, the closed-type answer generation unit 103B integrates the obtained open-type answer document and closed-type answer data to generate a single answer document (S306).

[0095] Integration here may involve adding a closed response document (including closed response data) to an open response document, and may also involve organizing the information as appropriate. Information organization may include modifying the format of the response document, deleting duplicate content, or deleting conflicting answers from the open response document. In particular, the format of the open response document is likely to be easy for users to understand. Therefore, it is preferable to maintain the format of the open response document while organizing the information to supplement the content of the closed response document.

[0096] The subsequent processing is the same as the question answering process according to the first embodiment, so the explanation is omitted here. Also, if it is determined in step S107 that the unique information is not included in the question information (S107, NO), the process proceeds to step S306. In this case, since there is no closed answer document to integrate, the open answer document becomes the answer document to be sent to the injection molding machine 20.

[0097] According to this embodiment, since both open-type and closed-type response documents can be generated, it becomes possible to make the response documents even more accurate.

[0098] Furthermore, it is conceivable that a confidentiality agreement may be concluded with the operating company of the question answering device 10A, stipulating that highly confidential information will not be leaked to external parties. In such cases, answers may be generated based on the open learning model 12 for the question information, even if it contains unique information, without using publicly known question information. Even when using an external company's open learning model 12, if a confidentiality agreement has been concluded with that company, answers may be generated based on the open learning model 12 for question information containing unique information. In the above case, highly confidential question information regarding the injection molding machine 20, etc., will not be incorporated into the open database 170, and the open database 170 can be used only for answers to the question information.

[0099] In each of the embodiments described above, the program (a question-answering program related to industrial machinery) that implements the various functions of the question-answering device 10 and the control device 200 was described as being pre-installed on the question-answering device 10 and the control device 200. However, the program may be provided by recording it on a computer-readable recording medium, such as a file in an installable or executable format. Here, the storage medium includes all media that can be read and executed by the computer as the question-answering device 10 and the control device 200 described above, such as magnetic tape, magnetic disks (hard disk drives, etc.), optical disks (CD-ROMs, DVDs, etc.), magneto-optical disks (MOs, etc.), flash memory, etc., which are removable from the control device 7, and media that can be transmitted via a network. Examples of media that can be transmitted via a network include those stored on a computer such as an external server connected to a network such as the Internet, and which are provided via the network. At least some of the functions of the question-answering device 10 and the control device 200 may be constructed by an external server, etc., of the question-answering system 1, and data may be provided from the server to the question-answering device 10 and the control device 200 via communication over the network.

[0100] While several embodiments of the invention have been described, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications are permitted without departing from the spirit of the invention. These embodiments and their variations are included within the scope and spirit of the invention, as well as within the scope of the invention and its equivalents as described in the claims. [Explanation of Symbols]

[0101] 1. Question Answering System 10 Question answering device 11. Closed learning models 12 Open-Type Learning Models 16 Closed Databases 17 Open Databases 101 Information Acquisition Unit (Question Information Acquisition Unit, Identification Information Acquisition Unit, Evaluation Acquisition Unit, Operation Information Acquisition Unit) 102 Judgment Processing Unit (Unique Information Judgment Unit, Correction Judgment Unit, Verification Judgment Unit) 103 Answer generation part 103A Open answer generator 103B Closed answer generator 104 Verification Processing Unit (Verification Section) 105 Learning Update Department (Update Department) 20 injection molding machine 200 Control device 207 Question Correction Section 241 Display (Question and Answer Section) 242 Input device (question reception section) 243 Mike (Question Reception) 244 Speakers (Question and Answer Section)

Claims

1. A question answering device that answers questions about industrial machinery, A question information acquisition unit that acquires question information received by the question reception unit, A unique information determination unit that determines whether the acquired question information includes unique information relating to the industrial machine, The answer generation unit generates answers to acquired question information based on an open learning model that has learned about publicly known information that can be provided not only to the user of the industrial machine but also to other users, and a closed learning model that has learned about information related to the industrial machine and information related to the unique information that can be provided only to the user of the industrial machine, and if it is determined that the question information includes the unique information, the answer generation unit generates an answer to the question information based on the closed learning model. Equipped with, The question information acquisition unit acquires publicly known question information, which consists of publicly known information extracted from the question, along with the question information. The response generation unit generates a first response to the publicly known question information based on the open learning model, generates a second response to the question information based on the closed learning model, and integrates the first and second responses to generate a response to the question information. Question answering device.

2. The system further includes an identification information acquisition unit that acquires identification information relating to the user of the industrial machine, The response generation unit generates responses to the extent that they can be disclosed to the user indicated in the acquired identification information. The question answering device according to claim 1.

3. The system has multiple closed learning models, Each of the multiple closed learning models is individually associated with each user's identification information. The response generation unit selects a closed learning model associated with the acquired identification information and generates a response based on the selected closed learning model. The question answering device according to claim 2.

4. Used in accordance with a confidentiality agreement concluded between the company operating the question answering device and the user handling the industrial machinery. The question answering device according to claim 1.

Images