Maintenance support automatic answering system and maintenance support automatic answering method

Abstract

This maintenance support automatic answering system (1) is characterized by comprising: a maintenance domain knowledge database (23) in which knowledge pertaining to machine maintenance is accumulated; a general knowledge database (22) in which knowledge not belonging to a specific domain is accumulated; a customer machine-specific knowledge database (24) in which the operation history of a machine of a customer is accumulated; and an answer generation unit (12) that, in a natural language, generates an answer to a question pertaining to machine maintenance from the customer on the basis of information accumulated in the maintenance domain knowledge database, the general knowledge database, and the customer machine-specific knowledge database.

Description

Maintenance Support Automatic Answer System and Maintenance Support Automatic Answering Method

[0001] The present invention relates to a maintenance support automatic answer system and a maintenance support automatic answering method.

[0002] When a malfunction occurs in equipment, a customer who maintains and manages the equipment refers to a maintenance manual or the like. However, when the malfunction is very serious and there is no time to refer to the maintenance manual or the like, or when no appropriate countermeasures can be found even after referring to the maintenance manual or the like, there is a known technique in which a computer installed in a maintenance center or equipment manufacturer automatically answers questions from customers.

[0003] The question-and-answer system of Patent Document 1 utilizes natural language information and non-language information (e.g., images, CAD data) and generates an answer based on a question by deep learning. That is, the question-and-answer system of Patent Document 1 receives a user's question as an input of voice data or text data, extracts information related to the question using natural language processing from information sources such as design specifications, maintenance reports, part information, CAD data, etc., generates an answer based on the extracted information, and presents the generated answer to the user through text display, voice output, and further display of the inference process.

[0004] Japanese Patent Application Laid-Open No. 2022-112541

[0005] In an actual maintenance site, the operation environment and operation method of equipment strongly reflect the individual circumstances of the customer (user). Therefore, even when a malfunction occurs in the equipment, the malfunction strongly reflects the individual circumstances of the customer. Thus, even if the equipment manufacturer and model are the same, the answers to questions regarding the malfunction differ for each customer and also differ over time. [[ID=十六]] [[ID=十七]]

[0006] However, Patent Document 1 lacks the idea that malfunctions occurring in equipment strongly reflect the individual circumstances of the customer. Therefore, the question-answering system in Patent Document 1 generates answers to customer questions using a single knowledge source. As a result, it takes a considerable amount of time to obtain the specific answers that the customer truly needs. Thus, the present invention aims to provide highly accurate maintenance support by utilizing information specific to each customer's machine.

[0007] The present invention provides an automated maintenance support response system comprising: a maintenance domain knowledge database that stores knowledge related to machine maintenance; a general-purpose knowledge database that stores knowledge not belonging to a specific domain; a customer machine-specific knowledge database that stores the operating history of the customer's machine; and a response generation unit that generates answers in natural language to customer questions regarding machine maintenance based on the information stored in the maintenance domain knowledge database, the general-purpose knowledge database, and the customer machine-specific knowledge database. Other means will be described in the section on embodiments for carrying out the invention.

[0008] According to the present invention, highly accurate maintenance support can be provided by utilizing information specific to each customer's machine. More specifically, this is as follows:

[0009] This invention enables the provision of personalized answers to each customer by utilizing data sheets that show the operating history, maintenance history, and machine specifications unique to each customer's machine, as well as troubleshooting information. This allows for the rapid provision of specific and precise maintenance information and diagnostic results based on the current state of the customer's machine, which was not possible with conventional systems, thereby improving customer satisfaction.

[0010] Furthermore, the present invention improves the accuracy of failure prevention and anomaly diagnosis by combining machine specifications, troubleshooting information, and operating history, thereby streamlining maintenance work. This minimizes machine downtime, enables early detection of anomalies and proactive maintenance planning, and reduces operating costs and optimizes maintenance schedules.

[0011] Furthermore, the present invention can provide real-time responses to customer inquiries, and through a combination of natural language processing (NLP), large language models (LLM), and retrieve augmented generation (RAG), it can deliver expert, personalized, and rapid answers. This enables quick and accurate responses, especially to urgent troubleshooting and specific questions regarding parts, thereby reducing the time to problem resolution.

[0012] In addition, by switching and using customer-specific search extension generation modules, the present invention provides accurate answers based on individual information, including machine specifications, without confusing customer-specific data. This improves data security and reliability, and enables customized services tailored to the needs of each customer.

[0013] Finally, the present invention improves customer satisfaction and strengthens recurring business by streamlining maintenance and troubleshooting for customers' machines. In particular, by providing high value-added support based on customer-specific data, it facilitates the building of long-term relationships with customers and promotes the renewal of maintenance and service contracts.

[0014] This is a diagram showing the overall configuration of the automated maintenance support response system of Embodiment 1. This is a diagram illustrating the functions of the main components of the automated maintenance support response system of Embodiment 1 in a step-by-step manner. This is a diagram showing the processing flow of Embodiment 1. This is a diagram illustrating the functions of the main components of the automated maintenance support response system of Embodiment 2 in a step-by-step manner. This is a diagram showing the processing flow of Embodiment 2. This is a diagram illustrating the functions of the main components of the automated maintenance support response system of Embodiment 3 in a step-by-step manner. This is a diagram showing the processing flow of Embodiment 4. This is a diagram illustrating the functions of the main components of the automated maintenance support response system of Embodiment 5 in a step-by-step manner. This is a diagram illustrating the chat screen. This is a hardware configuration diagram of the computer applied to the automated maintenance support response system.

[0015] (Introduction) Embodiments of the present invention provide expert and personalized answers by utilizing customer machine-specific information, including maintenance history, operating history, data sheets showing machine specifications, and troubleshooting information for industrial equipment such as centrifugal compressors.

[0016] Centrifugal compressors are used in a wide range of industries, including oil refineries, petrochemical and synthetic chemical plants, and natural gas plants, supporting stable production processes. In particular, proper maintenance and failure prevention are essential to maintain the performance and reliability of compressors, and after-sales services are widely provided to support this.

[0017] However, conventional after-sales services often only provide general technical information and standard maintenance procedures. Customized support that takes into account customer-specific operational history, maintenance history, machine specifications, data sheets, and troubleshooting information is often insufficient. As a result, customers are unable to adequately address the specific problems they face, leading to decreased customer satisfaction and inefficient maintenance work. Embodiments of the present invention aim to improve upon this issue.

[0018] Hereafter, embodiments 1 to 5 of the present invention will be described in detail with reference to the figures. For ease of understanding, a brief overview of these embodiments will be given first. Embodiment 1 is, so to speak, the basic model, which generates answers to questions using information from three different knowledge databases (Figures 1 to 3). Embodiments 2 to 5 are advanced versions of Embodiment 1, and hereafter, additional configurations compared to Embodiment 1 will be mainly described. Of these, Embodiment 2 acquires additional machine-specific knowledge other than natural language depending on the content of the question (Figures 4 and 5). Embodiment 3 manages machine-specific knowledge separately for each customer (Figure 6). Embodiment 4 updates machine-specific knowledge (Figure 7). Embodiment 5 uses external knowledge (Figure 8).

[0019] <Embodiment 1> Figure 1 shows the overall configuration of the automated maintenance support response system of Embodiment 1. The automated maintenance support response system 1 provides specialized and personalized maintenance information in response to maintenance-related questions from customers by referring to multiple knowledge databases (DBs).

[0020] The maintenance support automated response system 1 includes an application server 3, a web server 4, a customer information DB 21, a general-purpose knowledge DB 22, a maintenance domain knowledge DB 23, a customer machine-specific knowledge DB 24, an asset knowledge DB 25, and a machine specification DB 26. The maintenance support automated response system 1 is connected to the customer device 2 and an external portal site 5 via a wired or wireless network (not shown), such as the Internet.

[0021] The application server 3 has a chatbot application 11. The chatbot application 11 has an answer generation unit 12, a customer question acquisition unit 13, a question distribution agent 14, an additional information acquisition agent 15, and an action processing unit 18 other than AI (Artificial Intelligence). The additional information acquisition agent 15 has agent α 16 and agent β 17.

[0022] The maintenance support automated response system 1 is a general-purpose computer and its hardware includes a central control unit, input devices such as a keyboard, output devices such as a display, main memory, auxiliary storage, and communication devices (see Figure 10 below for the hardware configuration). The auxiliary storage stores each DB shown in Figure 1.

[0023] The chatbot application 11 in Figure 1, as well as each part and agent contained therein, are programs as software. In the following explanation, when a subject is indicated as "part ○" or "agent ○", it means that the central control unit reads each program from the auxiliary storage device into the main memory and executes the processing described in advance in that program.

[0024] In Figure 1, the automated maintenance support response system 1 is represented as a single enclosure. However, the automated maintenance support response system 1 may be configured as a distributed system (cloud) across multiple enclosures. In that case, each component of the automated maintenance support response system 1 as shown in Figure 1 will be contained within any enclosure on the cloud. For example, each database may be stored in an enclosure independent of the automated maintenance support response system 1. Note that Figure 1 is also common to embodiments 2 to 5 described later.

[0025] Figure 2 is a diagram illustrating the functions of the main components of the maintenance support automated response system of Embodiment 1 in a stepwise manner. In Figure 2, one step (indicated by symbols such as "104") may correspond to multiple configurations (rectangular and cylindrical). The same applies to other embodiments.

[0026] <Stage 101> The customer device 2 is a means by which the customer accesses the Web server 4 via a browser or dedicated application and sends questions to the maintenance support automated response system 1. Specifically, it is a personal computer, tablet, smartphone, etc. When the customer enters a maintenance-related question into the customer device 2, the customer device 2 sends the question to the Web server 4. Among the customer devices 2, those used exclusively by a particular customer are indicated as "Customer A's device 2a," and so on.

[0027] <Stage 102> The Web server 4 receives an HTTP request from the customer, analyzes the customer's question, and forwards (routes) it to the application server 3. This allows appropriate processing according to the content of the question to be carried out in subsequent stages.

[0028] <Stage 103> The application server 3 functions as the central hub of the entire maintenance support automated response system 1 and manages the data and information necessary to process customer inquiries. First, the application server 3 accesses the customer information DB 21 and obtains basic information about the customer (customer ID (Identifier), customer name, owned machine name, contract details, etc.). Note that within the customer information DB 21, the section that stores information for a specific customer is indicated as "Customer A Information DB 21a," for example. This prepares the basic information necessary to provide individualized answers for each customer. Next, the application server 3 transfers the customer inquiry to the chatbot application 11 via the customer inquiry acquisition unit 13.

[0029] <Stage 104> The chatbot application 11 receives questions from customers via the customer question acquisition unit 13 and analyzes their content. During the analysis process, the question distribution agent 14 analyzes the question content in detail in order to determine which database information should be retrieved from.

[0030] In this process, the chatbot application 11 uses a question distribution agent 14 to retrieve appropriate information from multiple databases. The question distribution agent 14 obtains the necessary information from one or more of the general-purpose knowledge database 22, the maintenance domain knowledge database 23, and the customer machine-specific knowledge database 24. By collecting knowledge from multiple sources rather than relying on a single database in this way, it becomes possible to provide broader and more accurate answers.

[0031] The General Knowledge DB 22 is a database that stores knowledge not belonging to a specific domain, and it stores general knowledge regarding technical information and basic operating principles. The General Knowledge DB 22 can also handle questions not directly related to machine maintenance. The Maintenance Domain Knowledge DB 23 stores knowledge related to machine maintenance. For example, if the machine is a centrifugal compressor, the Maintenance Domain Knowledge DB 23 stores maintenance information and technical knowledge for centrifugal compressors, as well as failure modes and countermeasures. The Maintenance Domain Knowledge DB 23 provides appropriate answers to technical questions about the machine. The Customer Machine-Specific Knowledge DB 24 stores specific data about the customer's machine (maintenance history, operation history, data sheets, instruction manuals, troubleshooting information). The Customer Machine-Specific Knowledge DB 24 may store such specific data for each customer. In this case, an area of ​​the Customer Machine-Specific Knowledge DB 24 that is independent for a specific customer will be indicated as "Customer A Machine-Specific Knowledge DB 24a". The Customer Machine-Specific Knowledge DB 24 generates customized answers for each customer.

[0032] <Stage 105> The information acquired by the question sorting agent 14 is passed to the answer generation unit 12. The answer generation unit 12 integrates the information from multiple databases received from the question sorting agent 14 and uses natural language processing to generate an answer in a format that is easy for the customer to understand. This allows the customer to quickly obtain specialized information regarding the maintenance of their machine.

[0033] For example, if a customer asks, "When should the next maintenance check be performed?", the answer generation unit 12 takes into account the maintenance history, operating history, and manufacturer-recommended maintenance intervals obtained from the customer machine-specific knowledge DB 24 to propose the optimal inspection timing. Furthermore, the answer generation unit 12 also takes into account general maintenance cycle information obtained from the maintenance domain knowledge DB 23 to create the final answer.

[0034] As shown above, Figure 2 visually illustrates the overall configuration of the automated maintenance support response system 1, clearly showing how each component works in cooperation with others. The automated maintenance support response system 1 enables customers to quickly receive expert and personalized support regarding maintenance. The entire chatbot application 11, or a part of its configuration, can be implemented as artificial intelligence (AI) that takes questions as input and outputs answers. Various databases may be used to deep learn this AI. Components shown in Figure 1 that are not explained in Figure 2 will be described later.

[0035] Figure 3 shows the processing flow of Embodiment 1. This processing flow visually explains the series of processes from when a customer inputs a question into the maintenance support automated response system 1 until the final answer is provided. Due to space limitations, in many cases, the entity that performs the processing is indicated in the rectangles showing each step in Figure 3 instead of the processing content.

[0036] In step S201, the customer device 2 receives a question from the customer regarding maintenance via a browser or dedicated application. The customer device 2 then sends this question to the web server 4. In step S202, the web server 4 forwards the results of parsing the HTTP request and the question to the application server 3.

[0037] In step S203, the application server 3 accesses the customer information DB 21 and obtains necessary information about the customer (customer ID, customer name, owned machine name, contract details, etc.). This information is essential for providing appropriate answers to questions and serves as the basis for generating personalized answers in subsequent processing. Then, based on the acquired customer information, the application server 3 passes the question to the chatbot application 11.

[0038] In step S204, the chatbot application 11 receives a question from the customer via the customer question acquisition unit 13 and processes its content. Specifically, the chatbot application 11 sends the content of the question to the question distribution agent 14.

[0039] In step S205, the question distribution agent 14 determines which database to retrieve information from based on the content of the question. For example, if the information related to the question is spread across multiple databases, such as the general-purpose knowledge database 22, the maintenance domain knowledge database 23, and the customer machine-specific knowledge database 24, the question distribution agent 14 accesses each database simultaneously to collect the most appropriate information. This allows the maintenance support automated response system 1 to provide more accurate and comprehensive answers without relying on a single database.

[0040] In step S206, the question distribution agent 14 starts a database search and retrieves information from the selected database. This search process is performed to quickly compare accumulated knowledge and customer machine-specific information to derive the most appropriate answer. In step S207, the answer generation unit 12 receives the search results from the question distribution agent 14. Based on the retrieved information, the answer generation unit 12 uses natural language processing to generate an answer in a format that is easy for the customer to understand.

[0041] In step S208, the answer generation unit 12 uses LLM to complement the answer generation process. That is, the answer generation unit 12 uses LLM to further refine the search results and provide contextually relevant, natural, and accurate answers. This means that highly accurate answers to customer questions are possible. In step S209, the answer generation unit 12 sends the generated answer back to the customer device 2. Here, the answer generation unit 12, acting as a chatbot, provides answers in a conversational format and sends maintenance-related information to the customer device 2 in real time. This allows the customer to quickly obtain specific guidance to proceed with maintenance work.

[0042] Figure 3 shows such a processing flow, clearly depicting the procedure for generating appropriate answers to customer questions while each step cooperates. A process is described in which each component functions in a coordinated manner and operates efficiently and accurately as a whole.

[0043] <Summary of Embodiment 1> As is clear from the above, the maintenance support automatic answering system 1 of Embodiment 1 includes a maintenance domain knowledge DB 23 that stores knowledge, technical information, and troubleshooting information related to general maintenance of mechanical equipment such as centrifugal compressors. The maintenance domain knowledge DB 23 includes component replacement cycles, maintenance procedures, general failure modes, and countermeasures in machines such as centrifugal compressors, and functions as a basis for providing general maintenance information that does not depend on specific customer machines. Thereby, it can respond to questions regarding general mechanical maintenance.

[0044] Next, the maintenance support automatic answering system {1} of Embodiment 1 includes a general knowledge DB 22 that stores general knowledge not belonging to a specific domain, enabling the system to respond to a wide range of questions other than mechanical maintenance. The general knowledge DB 22 stores technical information, operating principles, and knowledge regarding general business procedures that are not directly related to mechanical maintenance, enabling flexible responses to various questions.

[0045] Further, the maintenance support automatic answering system 1 of Embodiment 1 includes a customer machine-specific knowledge DB 24 that stores the operation history, maintenance history, data sheets indicating the specifications of the machine, operation manuals, and troubleshooting information specific to the customer machine. The customer machine-specific knowledge DB 24 stores detailed operation and maintenance histories related to a specific machine owned by the customer and functions as an information source for generating customized answers specialized for the customer machine. Thereby, detailed responses regarding individual machines become possible.

[0046] As is clear from the above, the maintenance support automatic answering system 1 of Embodiment 1 includes a chatbot application 11 that contains a natural language processing module (not shown), analyzes questions from customers, and determines which DB should be referred to based on the content of the questions. This natural language processing module forms a basis for extracting relevant information from an appropriate DB and providing an answer to the customer's question.

[0047] <Embodiment 2> Figure 4 is a diagram for explaining step by step the functions of the main configuration of the maintenance support automatic answering system of Embodiment 2. The maintenance support automatic answering system 1 of Embodiment 2 extracts information not only from natural language forms but also from documents such as table forms and graph forms based on questions from customers and provides an optimal answer.

[0048] <Step 301> The device 2a of customer A transmits a question to the maintenance support automatic answering system 1 via the Web server 4. The Web server 4 receives an HTTP request and routes it to the application server 3. <Step 302> The application server 3 receives the question from customer A and manages which DB and agent should be used to generate an answer. The customer information DB 21 stores information such as the ID for each customer, customer name, name of the owned machine, contract content, etc., and at this time, provides information necessary for identifying the customer and processing the content of the question.

[0049] <Step 303> The chatbot application 11 acquires the question from the customer and transfers it to the question distribution agent 14. The question distribution agent 14 analyzes the category of the question from the customer and determines which agent should be used to refer to which DB. Depending on the content of the question, information from multiple DBs may be integrated.

[0050] The chatbot application 11 has an additional information acquisition agent 15 in addition to the question distribution agent 14. The additional information acquisition agent 15 plays an additional role in acquiring specific information. Agent α 16 of the additional information acquisition agent 15 acquires troubleshooting information from the asset knowledge DB 25. The asset knowledge DB 25 stores technical information regarding countermeasures and repair procedures for specific defects in tabular or graphical format, and provides accurate answers to customer questions.

[0051] Furthermore, agent β17, one of the additional information acquisition agents 15, acquires detailed machine specification information from the machine specification DB 26. The machine specification DB 26 stores information on machine design specifications, machine parameters, parts compatibility, and parts performance in tabular or graphical format. In this way, the agents work in conjunction with each DB to extract accurate information in response to specific questions.

[0052] <Stage 304> The response generation unit 12 generates a response in natural language format using LLM based on the information collected from the agent and DB. This response is not merely a presentation of data, but provides comprehensive information that addresses the customer's question. <Stage 305> The customer machine-specific knowledge DB 24 stores maintenance history, operating data, and detailed specifications for each customer's machine, and includes customer A machine-specific knowledge DB 24a. This makes it possible to provide personalized information about a specific customer's machine.

[0053] <Stage 306> The maintenance support automated response system 1 is equipped with a maintenance domain knowledge DB 23 that stores general knowledge about maintenance. By integrating this general knowledge with machine-specific information, more sophisticated answers are generated. <Stage 307> The non-AI action processing unit 18 performs actions other than AI answers (for example, providing contact information) based on the question category for specific questions, thereby achieving a more flexible response. With the above configuration, the maintenance support automated response system 1 in Figure 4 uses multiple DBs and agents to provide the optimal answer to customer questions.

[0054] Figure 5 shows the processing flow of Embodiment 2. In step S401, customer A's device 2a receives an HTTP request from customer A regarding a question and maintenance. In step S402, the web server 4 receives the HTTP request and routes it to the application server 3.

[0055] In step S403, the application server 3 processes the question content, accesses the customer information DB 21, and refers to customer A's ID, contract details, and information on owned machinery. In step S404, the chatbot application 11 analyzes the content of the question. Based on customer A's question, the chatbot application 11 sends the analysis results to the question distribution agent 14.

[0056] In step S405, the question sorting agent 14 determines the category of the question and decides which database to access or which other agent to use. The question sorting agent 14 accesses the customer machine-specific knowledge database 24 to obtain historical data about customer A's machine and, if necessary, activates the additional information acquisition agent 15 to obtain additional information.

[0057] In step S406, the additional information acquisition agent 15 decides to acquire additional data using other agents. In step S407, agent α16 acquires troubleshooting information from the asset knowledge DB 25. The asset knowledge DB 25 includes past maintenance and repair history and is used as reference information to identify the cause and countermeasures for the problem.

[0058] In step S408, agent β17 collects detailed technical data of the machine from the machine specification DB26. This data is used to provide specification information necessary for maintenance work and repairs. In step S409, the response generation unit 12 collects this information and generates an appropriate response.

[0059] In step S410, the response generation unit 12 uses LLM to create a response in natural language. In this process, the response generation unit 12 integrates specialized knowledge and technical information related to maintenance obtained by referring to the maintenance domain knowledge DB 23, as well as general knowledge obtained by referring to the general knowledge DB 22, into the response generated in step S409.

[0060] If an action other than an AI response is required, in step S411, the non-AI action processing unit 18 performs action processing based on the question category. At this time, the response generation unit 12 sends the final response after the action processing to the customer device 2 via the chatbot application 11. The customer receives the response in real time. As is clear from the above, the maintenance support automated response system 1 as a whole seamlessly links a series of processes such as question analysis, information acquisition, and response generation to accurately and quickly answer the customer's maintenance-related questions.

[0061] <Summary of Embodiment 2> As is clear from the above, the maintenance support automatic response system 1 of Embodiment 2 includes an asset knowledge DB 25 that stores technical information regarding countermeasures and repair procedures for specific malfunctions, and a machine specification DB 26 that stores information regarding machine design specifications, machine parameters, parts compatibility, and parts performance. The response generation unit 12 generates a response by integrating information obtained from at least one of the asset knowledge DB 25 and machine specification DB 26 according to the content of the question with information obtained from the maintenance domain knowledge DB 23, general-purpose knowledge DB 22, and customer machine-specific knowledge DB 24 according to the content of the question.

[0062] Furthermore, the asset knowledge DB 25 and the machine specification DB 26 each store information in tabular and graphical formats, and the response generation unit 12 generates responses in natural language format as well as in at least one of the tabular and graphical formats.

[0063] Furthermore, the maintenance support automatic response system 1 of Embodiment 2 realizes the maintenance domain knowledge DB 23, general-purpose knowledge DB 22, customer machine-specific knowledge DB 24, and response generation unit 12 through a combination of a large-scale language model and search extension generation.

[0064] <Embodiment 3> Figure 6 is a diagram illustrating the functions of the main components of the automated maintenance support response system of Embodiment 3 in a stepwise manner. The automated maintenance support response system 1 of Embodiment 3 physically separates and manages the data of customer A and customer B, preventing the accidental disclosure of one customer's information to another customer. In particular, in the generation AI-based data processing, data leakage can be prevented by clearly switching the information source for each customer in order to minimize the risk of unintended data being accessed.

[0065] <Stage 501> First, Customer A and Customer B access the maintenance support automated response system 1 using Customer A's device 2a and Customer B's device 2b, respectively. Customer A's device 2a and Customer B's device 2b connect to the Web server 4 and send HTTP requests. The Web server 4 receives these HTTP requests and routes them to the application server 3.

[0066] <Stage 502> The application server 3 retrieves information about customer A from the customer A information DB 21a and performs appropriate data processing based on the customer's questions (the same applies to customer B). Each customer's maintenance history and contract details are managed independently and are not mixed with data from other customers. Furthermore, a generative AI such as LLM is used for this data processing, which can generate appropriate answers in natural language to customer questions.

[0067] <Stage 503> The chatbot application 11 receives and analyzes the customer's question. <Stage 504> The question distribution agent 14 analyzes the content of the customer's question and, based on that, accesses the appropriate database to generate an answer. For questions from customer A, the question distribution agent 14 accesses customer A machine-specific knowledge DB 24a, and for questions from customer B, it accesses customer B machine-specific knowledge DB 24b. This access method ensures that each customer's data is completely isolated and prevents it from being mistakenly provided to other customers.

[0068] <Stage 505> The additional information acquisition agent 15 is called by the question distribution agent 14 and collects the necessary information from each DB. In this process, agent α 16 acquires troubleshooting information from the asset knowledge DB 25, and agent β 17 acquires detailed machine specification information from the machine specification DB 26.

[0069] <Stage 506> The answer generation unit 12 integrates this information to generate the optimal answer. During the answer generation process, the answer generation unit 12 also acquires information from the maintenance domain knowledge DB 23 as needed. This ensures that the customer is provided with expert and accurate answers regarding machine maintenance.

[0070] <Stage 507> If a customer's question cannot be resolved by AI-generated answers alone, the answer generation unit 12 performs processing based on the question category and takes appropriate actions such as providing contact information. In this way, a system is realized that provides efficient and secure maintenance support while ensuring security and privacy, with customer information kept separate.

[0071] The automated maintenance support response system 1 manages data individually for each customer, preventing the mixing of information from different customers and minimizing the risk of data leakage during maintenance operations. Furthermore, the automated maintenance support response system 1 provides each customer with prompt and accurate maintenance information under appropriate data management.

[0072] <Summary of Embodiment 3> As is clear from the above, in the maintenance support automatic response system 1 of Embodiment 3, the customer machine-specific knowledge DB 24 is independent for each customer. The response generation unit 12 uses the information stored in the customer X machine-specific knowledge DB 24x corresponding to a specific customer X to generate a response for that specific customer. This provides a response based on data specific to the customer machine and prevents confusion with information from other customers. This mechanism enables accurate and safe individualized support for each customer and can increase customer satisfaction.

[0073] <Embodiment 4> Figure 7 is a diagram showing the processing flow of Embodiment 4. The maintenance support automated response system 1 of Embodiment 4 updates the operation history, maintenance history, and failure history from the customer and generates a more accurate response based on them.

[0074] In step S601, the customer device 2 receives the customer's access to the maintenance support automated response system 1 and inputs the latest operating history, maintenance history, and failure history information for its machine. In step S602, the customer device 2 stores this data in the customer machine-specific knowledge DB 24. The customer machine-specific knowledge DB 24 accumulates the operating history, maintenance history, failure history, etc., of the customer's machine.

[0075] In step S603, the customer machine-specific knowledge DB 24 updates its stored operational history, maintenance history, and failure history. This update may be performed sequentially in response to the occurrence of predetermined events, or it may be performed periodically. Specifically, the customer machine-specific knowledge DB 24 may be updated whenever a malfunction occurs in the customer machine, or it may be updated in nighttime batch processing, end-of-month batch processing, etc. As a result, the customer machine-specific knowledge DB 24 is always kept up-to-date. In step S604, the answer generation unit 12 generates an answer to the customer's question based on the updated data. Through this process, the maintenance support automatic answer system 1 accurately reflects the latest operational status and maintenance history of the customer machine and provides the customer with an appropriate answer. Since the updated data reflects the latest data from the customer, the answer also reflects the latest data from the customer, and its accuracy is high.

[0076] In step S605, the customer device 2 ultimately outputs a highly accurate response. The customer can receive the optimal maintenance method and appropriate countermeasures for their machine in real time, based on the latest data. Since the features of Embodiment 4 are concise, the "Summary of Embodiment 4" is omitted.

[0077] <Embodiment 5> Figure 8 is a diagram illustrating the functions of the main components of the automated maintenance support response system of Embodiment 5 in a stepwise manner. The automated maintenance support response system 1 of Embodiment 5 combines internal information from the customer machine-specific knowledge DB 24 and external information from the external portal site 5 to generate comprehensive and expert answers.

[0078] <Stage 701> The customer device 2 accepts the customer's access to the customer machine-specific knowledge DB 24 and references internal information. <Stage 702> The customer machine-specific knowledge DB 24 stores information specific to the machine owned by the customer (operation history, maintenance history, failure history, etc.) and plays an important role in generating responses by the maintenance support automated response system 1.

[0079] <Stage 703> Furthermore, the maintenance support automated response system 1 also references external information from an external portal site 5. The external portal site 5 includes publicly available maintenance information, technical documents, industry standard guidelines, and other publicly available data related to machine maintenance. By using the external portal site 5, the maintenance support automated response system 1 can generate responses based on a wider range of information, rather than relying solely on internal information.

[0080] <Stage 704> The answer generation unit 12 integrates internal and external information to generate the optimal answer to the customer's question. The answer generation unit 12 refers not only to customer machine-specific data but also to external information, providing a comprehensive answer based on the latest industry information and standards, rather than simply an answer based on past data.

[0081] <Stage 705> Finally, the maintenance support automated response system 1 provides the customer with a comprehensive and expert answer. This answer is the result of utilizing both internal and external information, and can provide the information necessary for the customer's maintenance work with greater accuracy and expertise. In this way, it is possible to increase the efficiency of the customer's maintenance work and improve the speed of problem solving. With this configuration, the maintenance support automated response system 1 provides more valuable answers to the customer and improves reliability in machine maintenance by flexibly utilizing external information without relying solely on internal information. The features of Embodiment 5 are concise, so the "Summary of Embodiment 5" is omitted.

[0082] (Chat Screen) Figure 9 is a diagram illustrating the chat screen. In the left column of the chat screen 31 displayed by the customer device 2, the answer generation unit 12 displays questions 1, 2, and 3 (reference numerals 41a, 41b, and 41c) posed by the customer in chronological order. In the right column of the chat screen 31, the answer generation unit 12 displays answers 1, 2, and 3 (reference numerals 42a, 42b, and 42c) posed by itself in chronological order. Answer 1 corresponds to question 1, answer 2 corresponds to question 2, and answer 3 corresponds to question 3. Note that questions do not necessarily have the form of interrogative sentences. Here, in a broad sense, the words posed by the customer are called "questions." Similarly, answers do not necessarily have the form of answers to questions and may include follow-up questions to the customer. Here, in a broad sense, the words posed by the answer generation unit 12 are called "answers."

[0083] The general-purpose knowledge DB 22 stores knowledge 32, "Identify the area the customer is focusing on." Knowledge 32 is not knowledge belonging to a specific domain, but rather very general knowledge for eliciting customer requests. The maintenance domain knowledge DB 23 stores knowledge 33, "If an abnormal noise occurs, check its frequency." Knowledge 33 is knowledge belonging to the maintenance domain and is used to narrow down the necessary maintenance. The customer machine-specific knowledge DB 24 stores knowledge 34, "In one instance, an object was placed around customer A's outdoor unit, resulting in poor ventilation and causing the compressor to intermittently emit abnormal noises." Knowledge 34 is knowledge belonging to the specific machine of that customer and is used to specifically resolve customer-specific problems.

[0084] The answer generation unit 12 generates answer 1 to question 1 by referring to knowledge 32. The answer generation unit 12 generates answer 2 to question 2 by referring to knowledge 33. The answer generation unit 12 generates answer 3 to question 3 by referring to knowledge 34.

[0085] (Knowledge sharing with other customers) As is clear from Figure 9, whether a customer can obtain truly accurate answers depends on the quality of the knowledge 34 stored in the customer machine-specific knowledge DB 24. A specific customer A would be satisfied if they could obtain answers that reflect the knowledge 34 about their own machine (operation history, maintenance history, failure history, etc.), but it would be even more convenient if the answers also reflected the knowledge 34 about similar machines used by other customers.

[0086] Therefore, the customer question acquisition unit 13 may obtain prior consent from the customer to "share knowledge about similar machines among customers." Such customers are called "knowledge-opening customers." Now, suppose customer A changes from a regular customer to a knowledge-opening customer. Then, the question distribution agent 14 modifies the knowledge 34 as follows.

[0087] (Knowledge 34 before correction) Objects were placed around customer A's outdoor unit, resulting in poor ventilation and causing the compressor to make intermittent noises. (Knowledge 34b after correction) Objects were placed around the outdoor unit [ID=○○], resulting in poor ventilation and causing the compressor to make intermittent noises.

[0088] In the revised knowledge 34b, "Customer A" is removed. Furthermore, an identifier that uniquely identifies the model (not the individual unit) of the "outdoor unit" is assigned. Then, the answer generation unit 12 provides answers that reflect the revised knowledge 34b to questions from customers other than Customer A. If other customers are using that model, the answer generation unit 12 can provide more accurate answers to those customers. In this way, knowledge that appears to be unique to a particular customer can be anonymized and then reflected in answers to other customers.

[0089] (Application to centrifugal compressors) The following is an example of the operation of the maintenance support automatic response system 1 applied to centrifugal compressors. Unlike mass-produced products, centrifugal compressors are individually designed and manufactured and operated in customer plants, etc., for a long period of time. Centrifugal compressors are equipped with various sensors. These sensors collect operational data including inlet pressure, outlet pressure, operating temperature, cooling characteristics, rotational speed, and vibration tendencies of the rotor or bearing parts in relation to load conditions during operation.

[0090] This operational data is periodically stored in the Customer Machine-Specific Knowledge Database 24. The Customer Machine-Specific Knowledge Database 24 stores information regarding the aging characteristics and operational status of centrifugal compressors, including design change history, modification history, and the supply status of replacement and repair parts. This information is regularly updated by the system administrator, ensuring that the latest operational data is reflected, thereby maintaining the overall accuracy and effectiveness of the system.

[0091] When a customer inputs a maintenance-related question into the automated maintenance support answering system 1, the system analyzes the content of the question using natural language processing technology. Then, the automated maintenance support answering system 1 searches for relevant information from the maintenance domain knowledge DB 23 and the customer machine-specific knowledge DB 24 to generate an answer to the analyzed question. In this process, the automated maintenance support answering system 1 makes recommendations for inspections and parts replacements based on pressure and temperature fluctuation data, proposes preventive maintenance based on vibration data, provides operating guides according to rotational conditions, and presents methods for preventing surging. For example, if the outlet pressure of a centrifugal compressor is fluctuating significantly over time, the automated maintenance support answering system 1 estimates that the cause is either a leak in the seals inside the centrifugal compressor or deterioration of the piping, and recommends specific inspection points and necessary parts replacements.

[0092] Furthermore, if the operating temperature trend data suggests a decrease in the efficiency of the cooling system, the maintenance support automated response system 1 will propose an inspection of the cooling system. In addition, if the vibration trend of the rotor or bearing part changes, the maintenance support automated response system 1 will prompt early repair or parts replacement as preventive maintenance based on operational data such as the machine's operating hours. If a history of design changes or modifications is accumulated, the maintenance support automated response system 1 will make maintenance proposals that correspond to the latest information based on these changes. In this way, the maintenance support automated response system 1 improves the efficiency of customer maintenance work and the reliability of the compressor by providing the time-series information necessary to support the long-term operation of the centrifugal compressor.

[0093] Figure 10 is a hardware configuration diagram of a computer applied to an automated maintenance support response system. The application server 11 in Figure 1 is composed of one or more computers as shown in Figure 10. The computer 900 shown in Figure 10 is configured as a computer 900 having a CPU 901, RAM 902, ROM 903, HDD 904, communication I / F 905, input / output I / F 906, and media I / F 907. The communication I / F 905 is connected to an external communication device 915. The input / output I / F 906 is connected to an input / output device 916. The media I / F 907 reads and writes data to the recording medium 917. Furthermore, the CPU 901 executes the program (application) loaded into the RAM 902 to realize the various functions shown in Figure 1, such as the response generation unit 12, the customer question acquisition unit 13, and the question distribution agent 14. This program can also be distributed via communication lines or by recording it on a recording medium 917 such as a CD-ROM.

[0094] It should be noted that the present invention is not limited to the embodiments described above, and various modifications are included. For example, the embodiments described above are described in detail for the purpose of clearly illustrating the present invention, and the present invention is not necessarily limited to having all the configurations described. Furthermore, it is possible to add, delete, or replace some of the configurations of the embodiments described above with other configurations.

[0095] 1. Maintenance support automated response system 2. Customer device 3. Application server 4. Web server 5. External portal site 11. Chatbot application 12. Response generation unit 13. Customer question acquisition unit 14. Question distribution agent 15. Additional information acquisition agent 16. Agent α 17. Agent β 18. Action processing unit other than AI 21. Customer information database (DB) 22. General-purpose knowledge database (DB) 23. Maintenance domain knowledge database (DB) 24. Customer machine-specific knowledge database (DB) 25. Asset knowledge database (DB) 26. Machine specification database (DB)

Claims

1. A maintenance support automated response system comprising: a maintenance domain knowledge database that stores knowledge related to machine maintenance; a general-purpose knowledge database that stores knowledge not belonging to a specific domain; a customer machine-specific knowledge database that stores the operating history of the customer's machine; and an answer generation unit that generates answers in natural language to customer questions regarding machine maintenance based on the information stored in the maintenance domain knowledge database, the general-purpose knowledge database, and the customer machine-specific knowledge database.

2. The maintenance support automatic answering system according to claim 1, comprising: an asset knowledge database storing technical information relating to countermeasures and repair procedures for specific defects; and a machine specification database storing information relating to machine design specifications, machine parameters, parts compatibility, and parts performance, wherein the answer generation unit generates the answer by integrating information obtained from at least one of the asset knowledge database and the machine specification database according to the content of the question with information obtained from the maintenance domain knowledge database, the general-purpose knowledge database and the customer machine-specific knowledge database according to the content of the question.

3. The maintenance support automatic response system according to claim 2, characterized in that the asset knowledge database and the machine specification database store information in tabular and graph format, and the response generation unit generates the response in natural language format as well as in at least one of tabular and graph format.

4. The maintenance support automatic response system according to claim 1, characterized in that the maintenance domain knowledge database, the general-purpose knowledge database, the customer machine-specific knowledge database, and the response generation unit are realized by a combination of a large-scale language model and search extension generation.

5. The maintenance support automated response system according to claim 1, characterized in that the customer machine-specific knowledge database is independent for each customer, and the response generation unit generates a response for a specific customer using information stored in the customer machine-specific knowledge database corresponding to that specific customer.

6. The maintenance support automatic response system according to claim 1, characterized in that the customer machine-specific knowledge database periodically updates the operating history, maintenance history, and failure history of the machines used by the customer, which are stored therein, either periodically or when a predetermined event occurs.

7. The automatic maintenance support response system according to claim 1, characterized in that the response generation unit generates a response by referring to information provided on an external portal site related to machine maintenance.

8. The maintenance support automatic response system according to claim 1, characterized in that the machine is a centrifugal compressor, and the customer machine-specific knowledge database stores the operating history of the centrifugal compressor regarding pressure, temperature, or vibration.

9. An automated maintenance support response method characterized by an automated maintenance support response system that generates answers in natural language to customer questions regarding machine maintenance based on information stored in a maintenance domain knowledge database that stores knowledge related to machine maintenance, a general-purpose knowledge database that stores knowledge not belonging to a specific domain, and a customer machine-specific knowledge database that stores the operating history of the customer's machine.

Images