Building quality information acquisition method, robot and computer program product

Abstract

The invention relates to a building quality information acquisition method, a robot and a computer program product, belongs to the field of building quality detection, and aims to solve the problem of building component quality information acquisition. The method comprises the steps that a building quality information acquisition robot matches input content with detection item characteristic values in a detection item list, and the input content comprises measurement data of a detection tool; if the matched detection item characteristic value exists, corresponding detection items are filled according to the input content, if the matched detection item characteristic value does not exist, a corresponding method list is determined according to the currently executed detection task, whether a method in the method list is triggered or not is judged according to the input content, and if the method is not triggered, the corresponding detection items are filled according to the input content. And if yes, selecting information related to the input content from the knowledge base and outputting the information. According to the invention, intelligent identification and automatic filling of detection data can be realized, the efficiency and accuracy of detection data acquisition are improved, and related auxiliary reference information can be provided for detection personnel.

Description

Technical Field

[0001] The present invention relates to the technical field of building quality detection, and in particular to a building quality information collection method, a robot and a computer program product. Background Art

[0002] In the quality inspection process of the construction industry, the collection of inspection data is a crucial step. However, many inspection data collection scenarios currently still require manual operation, and are manually filled in and submitted in the system after measurement. This traditional method of collecting building quality information has many shortcomings, as shown in the following two points: First, the on-site environment of construction sites and factories is often complex and changeable, which poses a great challenge to the operation of manual reporting systems. Operators need to measure on-site, but due to the instability of the on-site environment and the operating requirements of the reporting system, operators usually need to return to a relatively stable environment such as the office to fill in the data. This model not only increases the workload of operators, but may also affect the timeliness and accuracy of data collection.

[0003] Secondly, in the process of manually filling in data, human factors such as negligence, fatigue or misunderstanding may lead to data errors or omissions, which not only affects the quality and integrity of the data, but may also mislead subsequent quality analysis and decision-making.

[0004] Therefore, the current method of collecting building quality information needs to be improved. Summary of the invention

[0005] In order to solve the above problems, the inventors have made the present invention, which provides a method for collecting building quality information, a robot and a computer program product, aiming to realize intelligent recognition and automatic reporting of test data by combining the detection tool with a natural language processing model, a database, etc., and improve the efficiency and accuracy of test data collection. In addition, the present invention can also provide relevant specifications, processes and parameter consultations during the detection process, and provide auxiliary reference information for staff who collect building quality information.

[0006] In a first aspect, the present invention provides a method for collecting building quality information, comprising the following steps: Determine the detection task currently being performed; The building quality information collection robot determines the corresponding inspection item list according to the inspection task currently being performed; The building quality information collection robot includes an AI module that performs natural language processing on input content; the building quality information collection robot is connected to a building quality information database; the database stores characteristic values ​​of detection task types, characteristic values ​​of detection items, a list of detection items and a list of methods corresponding to the detection tasks, and a knowledge base; The building quality information collection robot matches input content with characteristic values ​​of detection items in the detection item list, and the input content includes measurement data of the detection tool; If there is a matching detection item feature value, fill in the corresponding detection item according to the input content. If there is no matching detection item feature value, the corresponding method list is determined based on the currently executed detection task, and whether the method in the method list is triggered is determined based on the input content. If the method is not triggered, information related to the input content is selected from the knowledge base for output.

[0007] Specifically, determining the detection task currently being executed includes the following steps: Obtain the information of the component to be detected, query the task library according to the information of the component to be detected, and determine the detection task currently being executed; The task library stores the detection task information and the association information between the detection tasks and components.

[0008] Specifically, the building quality information collection method also includes the following steps: The building quality information collection robot interacts with users in real time through instant messaging tools; the AI ​​module includes the LLM model, text vector library and knowledge graph. The building quality information collection robot performs semantic analysis on the input content through the AI ​​module, extracts the reported content, or triggers methods, or feeds back information related to the input content.

[0009] Specifically, the building quality information collection method also includes the following steps: The detection tool sends the measurement data to the edge device of the Internet of Things, and the edge device pre-processes the measurement data to obtain the measurement data in a preset format; Send the measurement data in a preset format to the building quality information collection robot.

[0010] Specifically, fill in the corresponding detection items according to the input content, including the following steps: Determine the data reporting method of the corresponding detection item, extract the reporting content from the input content, report the corresponding detection item in the preset data reporting method, and update the detection item list; Output the next detection item name based on the detection item list.

[0011] Specifically, the building quality information collection method also includes the following steps: A template library is established in the database, prompt word templates for different scenarios are established in the template library, and corresponding roles, profiles, skills and rules are configured for each prompt word template; The template library stores the association information between the prompt word templates and the components.

[0012] Specifically, the building quality information collection method also includes constructing a knowledge base, and constructing the knowledge base specifically includes: Collect text content of professional books, papers, specifications, standards, cases and experiences in the field of building inspection; Convert text content into multiple text vectors; According to data features of the multiple text vectors, the multiple text vectors are classified into different scenarios, and each text vector corresponds to at least one scenario.

[0013] Specifically, selecting information related to the input content from the knowledge base for output includes the following steps: According to the similarity between the input content and the text vector, select the text vector and scene related to the input content; According to the selected text vector and scene, the corresponding text content is output.

[0014] In a second aspect, the present invention provides a building quality information collection robot, comprising an interaction layer, a data layer and a collection layer; The interactive layer is used for input and output of interactive content; The data layer includes a building quality information database, which stores characteristic values ​​of the detection task type, characteristic values ​​of the detection items, a list of detection items and a list of methods corresponding to the detection task, and a knowledge base; The acquisition layer is used to determine the corresponding detection item list according to the currently executed detection task; match the input content with the detection item feature value in the detection item list. The input content includes the measurement data of the detection tool. If there is a matching detection item feature value, the corresponding detection item is filled in according to the input content. If there is no matching detection item feature value, the corresponding method list is determined according to the currently executed detection task. According to the input content, it is determined whether to trigger the method in the method list. If the method is not triggered, the information related to the input content is selected from the knowledge base and sent to the interaction layer for output.

[0015] Based on the same inventive concept, the present invention also provides a computer program product, including a computer program / instruction, which implements the aforementioned building quality information collection method when executed by a processor.

[0016] The beneficial effects of the above technical solution provided by the present invention include at least: By combining the detection tool with a natural language processing model, a database, etc., the intelligent recognition and automatic reporting of the detection data are realized, and the efficiency and accuracy of the detection data collection are improved. In addition, the present invention can also provide relevant knowledge information during the detection process, providing auxiliary reference for the staff who collect the building quality information.

[0017] Other features and advantages of the present invention will be set forth in the description which follows.

[0018] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 A schematic diagram of a solution target in an embodiment of the present invention; Figure 2 This is a flow chart of a method for collecting building quality information in an embodiment of the present invention; Figure 3 A schematic diagram of a robot system framework in an embodiment of the present invention; Figure 4 This is a schematic diagram of binding a detection tool and filling in data in an embodiment of the present invention; Figure 5 This is a schematic diagram of a detection task startup method in an embodiment of the present invention; Figure 6 A schematic diagram of detection items and intelligent detection work in an embodiment of the present invention; Figure 7 It is a schematic diagram of intelligent assistance for detection items in an embodiment of the present invention.

[0020] Figure 4 and Figure 5 The two-dimensional code pattern in the figure is a legend and is not recognizable. When scanning the two-dimensional code pattern in the figure, no content can be scanned. The implementation of the present invention does not rely on Figure 4 , Figure 5 The QR code pattern in . DETAILED DESCRIPTION

[0021] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0022] In order to solve the problems existing in the prior art, the embodiments of the present invention provide a method for collecting building quality information, a robot and a computer program product. Taking the field of prefabricated concrete building quality inspection as an example, on the one hand, in order to solve the current situation of separation of measurement operation and data reporting operation in the quality inspection link of prefabricated concrete components, the measurement results are collected by human-robot interaction, avoiding the operation of manually logging into the business system to report data, and improving the efficiency and accuracy of information collection; on the other hand, through the application of artificial intelligence knowledge base and large language model, it can provide instant question-answering services for building quality inspectors for related issues in the inspection process. Figure 1 The robot avatar in the figure represents the robot implemented in this solution.

[0023] The embodiment of the present invention provides a method for collecting building quality information, the process of which is as follows: Figure 2 As shown, the following steps are included: Step S1: Determine the inspection task currently being performed; the building quality information collection robot determines the corresponding inspection item list based on the inspection task currently being performed; the building quality information collection robot includes an AI module that performs natural language processing on the input content; the building quality information collection robot is connected to a building quality information database; the database stores the inspection task type characteristic value, the inspection item characteristic value, the inspection item list and method list corresponding to the inspection task, and the knowledge base.

[0024] The system framework of the building quality information collection robot is as follows Figure 3 As shown in the figure, it includes the interaction layer, collection layer, database layer and application system. Among them, the application system includes the original business management system of the enterprise, such as the factory system, task scheduling management, etc. At the same time, in order to receive the data results from the interactive confirmation and the measurement results directly sent by the detection tool, the application system also develops a corresponding data collection service based on the asset management shell (AAS), which can automatically distinguish and process various types of collection results. The data layer collects various business data from the application system and knowledge data related to intelligent services. Among them, the data of the task library comes from the task scheduling management function of the application system, which can obtain the relevant task information required for quality inspection. The template library is a customized configuration of different detection content templates combined with different detection scenarios. It is a data model formed after the detection scenario is concretized. Through the template library, the robot obtains the data required for the current task and then carries out the corresponding collection work. The knowledge base contains various structured and unstructured knowledge information. This part of the content complements the text vector library and knowledge graph in the AI ​​module. In specific scenarios, it can provide more complete and professional information support for artificial intelligence. The method library contains relevant information about various methods that need to be called when the application module obtains event instructions, which is used to support the robot to automatically perform related operations.

[0025] At the collection layer, such as Figure 3 As shown in the figure, the building quality information collection function is mainly divided into three parts: AI module, application module and detection tool. The AI ​​module includes the application of AI agent based on prompt words, and the knowledge service of the agent related to the reporting process, which mainly completes the answers to relevant questions, the call of execution standards and the rationality check of the content when filling in the information. The application module is responsible for the collection and submission of the user's reporting information in an interactive mode, and is responsible for the call of related functions completed with the AI ​​module, and the background management of related objects. At the same time, the application module also supports receiving the result information sent from the networkable detection tool, and combines the information with the user tasks bound to the tool to realize the automatic reporting of the detection results.

[0026] Testing tools refer to various indicator measurement tools used in the process of collecting quality information of prefabricated concrete components, such as common laser rangefinders, electronic thermometers, rebound testers, etc. If these tools have networking capabilities, the collection results can be sent directly to the robot through edge devices for reporting. If they cannot be connected to the Internet, the content can also be reported through human-computer interaction. Figure 4 As shown, if the detection tool can be connected to the Internet, a QR code with the unique identification information of the device can be generated in advance. After the construction quality inspector passes the QR code to the robot, the corresponding relationship between the tool and the person can be established. After that, the measurement results submitted by the tool will be passed to the person's current quality inspection task.

[0027] The interaction layer is mainly responsible for real-time interaction with users through instant messaging tools (IM). This layer is the user-side embodiment of the robot. The tool can use existing IM tools, such as WeChat for Enterprise, DingTalk, Feishu, etc., or it can customize the company's own IM tools as needed. The implementation of the present invention does not depend on specific instant messaging tools or specific large language models. This solution takes DingTalk as an example of a research and verification tool. Based on the chatbot API interface of the DingTalk developer platform, develop a chatbot that interacts with people. This robot can receive user input and make corresponding responses. Figure 3 In the illustrated embodiment, the interaction layer performs user identity authentication, sending and receiving real-time messages, language conversion, authority management, sending and receiving files, image recognition, and other functions that can be integrated into instant messaging tools.

[0028] The interaction process also includes: Authentication and authorization of API interfaces to ensure that only legitimate users can access the interfaces. Use authentication protocols such as OAuth2.0 and combine them with DingTalk's user system for authentication. Encrypt key data for transmission to ensure data confidentiality and integrity. Use secure protocols such as HTTPS for data transmission.

[0029] Specifically, register a developer account on the DingTalk developer platform, create an application, and obtain the relevant API key and token to provide the necessary authentication information for subsequent development. Use the robot API interface provided by DingTalk to receive messages sent by users through HTTP requests. Parse the received HTTP request to extract key information such as message content, sender information, and message type. Parse the extracted message content according to DingTalk's message format specification. The message content may include multiple formats such as text, pictures, voice, and video, which need to be processed separately. For text messages, the AI ​​module will be called to obtain relevant responses. After processing according to the AI ​​module and subsequent functions, the system will send the corresponding response content to the user in a specified manner through the DingTalk robot API interface. Such as text, cards, tables, pictures, etc. Connect the developed chatbot backend service with the DingTalk robot API interface. Configure the robot's Webhook address and related security settings on the DingTalk developer platform. Perform multiple compatibility tests on the robot and the DingTalk platform to ensure that messages can be received and sent normally.

[0030] In some specific embodiments, determining the currently executed detection task includes the following steps: obtaining information of the component to be detected, querying a task library based on the information of the component to be detected, and determining the currently executed detection task; the task library stores the detection task information and the association information between the detection task and the component.

[0031] The detection task is the starting point for the detection work. Personnel can start the detection work in two ways, such as Figure 5 As shown in the figure, one is to view the task list and select the work to be started based on the interactive card of the instant messaging tool, and the other is to scan or take a photo of the QR code of the component to be detected. The application module can automatically query the task to be detected of the current component based on the component information of the QR code and feedback to the user. Both methods are based on the task data obtained from the task library. After the detection work is completed, the robot will call the method in the method library to complete the task status update.

[0032] In some optional embodiments, querying for detecting a person or a building component comprises the following steps: Establish a database containing all inspection tasks and component information. Generate a unique QR code for each component and record the corresponding inspection task information in the factory system. Users scan the QR code through DingTalk to trigger the query operation. Develop an interface that connects to the factory system to query the inspection task information. When the corresponding inspection task is found, notify the user through the instant messaging tool and start the inspection work.

[0033] Specifically, a MySQL relational database is selected to store inspection tasks and component information. Design the tasks table to store basic information of inspection tasks, such as task ID, task name, component ID, inspection requirements, creation time, etc. Design the components table to store basic information of components, such as component ID, component name, specification model, manufacturer, etc. Design other related tables according to business needs, and establish associations between tables. After the database is established, import initial data, including existing inspection tasks and component information. Use the ETL (Extract Transform Load) tool to regularly synchronize information such as organizations, personnel, and component types related to inspection. Use Python's qrcode library as a QR code generation tool to generate a unique QR code for each component. The information encoded in the QR code includes the component ID and other unique identifiers. Print or paste the generated QR code on the component, and initialize the relevant inspection task information in the factory system. The user scans or takes a photo of the QR code on the component in DingTalk, triggering the interaction layer to recognize the QR code information. Obtain the work task information corresponding to the scan result (i.e. the encoded information in the QR code) through the API interface provided by the factory system. Specifically, the API interface is a RESTful API interface, which is used to receive query requests sent by DingTalk and return the corresponding detection task information. The input parameters of the interface include unique identifiers such as component ID. The output parameters of the interface include basic information of the detection task, detection status, current detection results, etc. The obtained results are saved in the local database as the basic data for the detection work. If a corresponding detection task is found, the user is notified through the interaction layer (such as DingTalk's chatbot). If no results are found, the corresponding information will be fed back to the user. While notifying the user, the startup process of the detection work can be triggered to start data collection.

[0034] In some specific embodiments, the building quality information collection method also includes constructing a building quality information collection robot, which interacts with users in real time through instant messaging tools; the AI ​​module includes an LLM model, a text vector library and a knowledge graph, and the building quality information collection robot performs semantic analysis on the input content through the AI ​​module, extracts the reported content, or triggers methods, or feeds back information related to the input content.

[0035] Specifically, a clear feature value is defined for each task category to be detected to distinguish different detection tasks. The task type to be detected is clarified, and the set of items to be detected is obtained through the task type. The method information that may be used is preset for each task type to form a method list. A unique feature value is set for each detection item, including information such as the detection result type, unit, and conversion method. The feature value of the detection task type and the feature value of the detection item are stored in the database to facilitate system reading and comparison. The user input information received by the user end is sent to the server end, and the server end is deployed with an AI module. On the server end, the detection task exclusive mode is set for the current dialog user. The correspondence between the user and the detection task is stored in the Redis database (Remote Dictionary Server, Redis is a high-performance key-value database) to clarify whether the quality inspection work is currently being carried out. After receiving the information input by the user, first check whether the user is currently performing the detection task and whether the input information meets the feature value of the item to be detected (such as a number plus a unit). If the input information meets the feature value of the detection item, it is compared with the preset input unit and other specifications to determine whether the input information meets the detection requirements. If the detection item needs to be converted, the conversion method is called to convert the input information into a standard format. If the input information does not meet the detection item feature value requirements, the background knowledge of the current detection task, the possible method information, and the context information are sent to the AI ​​module. The AI ​​module organizes the received information and calls the GPT model to analyze it. Taking the "Zhi Pu Qing Yan" large language model as an example, the background knowledge corresponds to the system role in the context, and the method information used corresponds to the function type input. Check whether the analysis result of the AI ​​model triggers a method preset in a certain scenario, such as a method for updating the detection result. If the analysis result triggers a method, the corresponding method is executed through the method library. If the analysis result of the AI ​​model triggers the update detection result method, the detection result of the specific detection item is updated according to the analysis result. And automatically trigger the next link of the item to be detected. In this way, interactive detection result submission without tool input is realized. If the analysis result does not trigger the method in the method list, the information related to the input content is selected from the knowledge base for output according to the input information.

[0036] In some specific embodiments, the method for collecting building quality information also includes the following steps: the detection tool sends the measurement data to an edge device of the Internet of Things (IoT), the edge device preprocesses the measurement data to obtain measurement data in a preset format; and the measurement data in the preset format is sent to the building quality information collection robot.

[0037] In some optional embodiments, the process of collecting building quality information includes: The user starts the inspection by taking a photo of the QR code or by asking the robot to do the task list. Establish a task library, template library and method library to store and manage the information of inspection tasks, templates and methods. Through the application interface, integrate the semantic analysis results of the AI ​​module, receive the inspection data and results, or execute the instructions issued by the AI ​​module or the interaction layer, such as submitting the inspection results. By developing the application interface corresponding to the edge device, integrate the edge device data and receive the data results submitted by the edge device. Through standard protocols and methods, pair the edge device with the IoT inspection tool and exchange data. Furthermore, through the API interface, support integration and docking with other tools, such as sensors, measuring instruments, etc. Develop edge device software, support Bluetooth and other protocols, integrate with the inspection tool, pre-process the received data information and organize it into data results, and submit the results to the collection layer.

[0038] In some specific implementations, the process of collecting building quality information includes: A relational database (such as MySQL) is used to store information about detection tasks, task templates, and methods. According to the task type, the database is queried to obtain the list of items to be detected for the task to be detected. If the current component needs to be tested before delivery, the query result will return all the detection item information for the pre-delivery test. The list of items to be detected is stored in the Redis database to facilitate adjustment of the detection results. The data layer uses RESTful API to communicate with the AI ​​module and edge devices. When the user uses the detection tool for detection, the device QR code is first sent to the server through instant messaging tools such as DingTalk. After parsing, the server saves the corresponding relationship between the device and the user in the Redis database. The subsequent data sent by the device will be used as the detection result of the detection task currently executed by the user. The result data of the detection tool is first sent to the edge device through standard protocols (such as MQTT, CoAP) and methods. The edge device includes dedicated software to support integration with other tools (such as sensors and measuring instruments), such as supporting protocols such as Bluetooth, and integrating with the detection tool. The edge device pre-processes the received data information, organizes it into data results in a specified format, and then encrypts the data results using a specified method (such as AES, Advanced Encryption Standard). The data reporting dedicated API is called through https to send the encrypted data to the server, which receives and decrypts the data to obtain the data detection results. When all the detection results of the current detection item list are completed, or a submission instruction confirmed by the user is received, the AI ​​module issues a submission instruction. When the AI ​​module sends a submission instruction, all the detection results of the current detection item list are retrieved from the Redis database, organized into a standard format for data reporting, and sent to the data reporting dedicated API. After receiving the data results, the data reporting dedicated API will execute the data processing program, update the data results and status, and synchronize the relevant information to the factory system through the factory system API.

[0039] In some optional implementations, when the server-side application receives data reported from different methods, it first retrieves the attributes of the item to be detected that the user is currently working on from the Redis database to determine whether the current item to be detected allows data to be reported in this way. If the data reporting method is not allowed by the item to be detected, feedback is given through a text message. If the data reporting method is allowed by the item to be detected, the detection result table in the Redis database is updated according to the instruction content. According to the list of detection items, the next item to be detected is automatically queried, and feedback is given to the user through a text message, instructing the user to carry out the next detection work.

[0040] Step S2: The building quality information collection robot matches the input content with the detection item feature values ​​in the detection item list, and the input content includes the measurement data of the detection tool; if there is a matching detection item feature value, the corresponding detection item is filled in according to the input content; if there is no matching detection item feature value, the corresponding method list is determined according to the currently executed detection task, and whether the method in the method list is triggered according to the input content; if the method is not triggered, information related to the input content is selected from the knowledge base for output.

[0041] In some specific embodiments, filling in corresponding detection items according to input content includes the following steps: determining a data reporting method for the corresponding detection items, extracting reporting content from the input content, filling in corresponding detection items in a preset data reporting method, and updating a detection item list; outputting the next detection item name according to the detection item list.

[0042] Inspection items refer to specific items that need to be reported numerically when carrying out inspection work according to different components and corresponding inspection requirements. The information that needs to be inspected for each component can be obtained by associating the information in the task library and the template library. The inspection item list is ultimately reflected in a structured data format, such as forming a string in JSON format as a list of inspection items. After that, the robot will automatically report the received measurement results to the value of the corresponding inspection item to complete the reporting process.

[0043] Intelligent detection process, such as Figure 6 As shown in the figure, it is a process of combining the user's natural language with the structured data (such as JSON) formed by the extraction and combination of detection items to obtain the final result through intelligent analysis. In this process, in order to understand the user's intention as accurately as possible, it is necessary to combine the large language model (LLM) to process natural language and structured data. In this process, the use of prompt words, thinking chains and other means has a great impact on improving the reasoning accuracy of the large language model. Therefore, for different components, prompt word templates related to component features can be set. When the robot calls the LLM tool for reasoning analysis, the specific features of the component will help the robot obtain more accurate results through prompt word templates, result self-verification, related AI algorithms, etc.

[0044] In some specific embodiments, the method for collecting building quality information also includes the following steps: establishing a template library in a database, establishing prompt word templates for different scenarios in the template library, and configuring corresponding roles, introductions, skills and rules for each prompt word template; the template library stores the association information between the prompt word templates and the components.

[0045] For different scenarios, establishing scenario-related prompt word templates can effectively improve the accuracy of the robot's answering questions in different scenarios. According to the template requirements, it is necessary to define the role, introduction, skills, rules, etc. of each intelligent agent, and generate prompt words through template library information matching. Taking the scenario of filling in structured (JSON) data as an example, the template design example is as follows: In some specific embodiments, the method for collecting building quality information also includes constructing a knowledge base, and constructing the knowledge base specifically includes: collecting text content of professional books, papers, specifications, standards, cases and experiences in the field of building inspection; converting the text content into multiple text vectors; and classifying the multiple text vectors into different scenarios based on data features of the multiple text vectors, each text vector corresponding to at least one scenario.

[0046] During the detection process, the robot also needs to answer the user's questions related to the scene during operation. Therefore, through the retrieval enhanced document question and answer function, this kind of service can be provided to the user. The process is as follows: Figure 7 As shown. The user input content is encoded to obtain the question vector encoding, which is input into the text vector database for retrieval, and the relevant text fragments are found to obtain a text fragment set, and the text fragment set is encoded to obtain a new text vector encoding, and the new text vector encoding is added to the text vector database. The large language model of the AI ​​module is trained and learned based on the updated text vector database, and context enhancement processing is performed to obtain a more accurate answer text, which is output to the user through the robot's interactive layer.

[0047] In some specific embodiments, information related to the input content is selected from a knowledge base for output, including the following steps: based on the similarity between the input content and the text vector, a text vector and a scene related to the input content are selected; based on the selected text vector and scene, the corresponding text content is output.

[0048] In some optional implementations, the construction quality information collection method includes the following steps to implement the accompanying inspection question and answer process: Establish a knowledge base of detection data vectors, including all knowledge and information related to detection. Use RAG (Retrieval-Augmented Generation) to answer questions, that is, retrieve relevant information from the knowledge base based on the questions entered by the user, and use a large language model to generate answers. Update the knowledge base regularly to ensure that it contains the latest detection knowledge and information. At the same time, continuously optimize the quality of RAG's answers based on user feedback and AI analysis results.

[0049] In some optional implementations, the companion question and answer implementation method is: Collect text content in the field of building inspection. The text content includes professional books, papers, specifications, standards, cases, etc., as well as experience sharing by industry experts and inspectors. Use text recognition and text deduplication methods to remove irrelevant information, such as advertisements and redundant content, from the text content to ensure data quality. Use GPT's pre-trained language model (such as ChatGLM2) to convert text data into vector representation for subsequent retrieval and generation. For quality inspection scenarios, extract content related to the scenario, such as precautions and inspection points for the inspection scenario, to form the inspection scenario background information. Use the GPT large language model to classify the background information of each scenario into items to form a background knowledge item set for the scenario. Establish a mapping relationship between the background knowledge item set related to the task type. Use natural language processing technology to perform word segmentation, part-of-speech tagging, syntactic analysis, etc. on the questions entered by the user to understand their semantics. According to the questions entered by the user, extract the background knowledge item set, and retrieve the text vector related to it in the knowledge base. By calculating the similarity, select the most matching records as candidate answers. The GPT generative language model is used to generate the final answer based on the entry set and the retrieved candidate answers, combined with the context of the user's question. The generated answer is passed to the interaction layer and sent to the user as a response.

[0050] In the above method of this embodiment, by combining big model technology and interactive tools, the quality inspection process is made intelligent and efficient, which not only significantly improves the convenience of the inspection work, but also greatly shortens the time of data circulation, avoiding the cumbersome process in the traditional way that the inspectors need to complete the quality inspection work on site and then return to the office to log in to the system to fill in the result data.

[0051] Specifically, the robot uses the instant messaging function to enable inspectors to submit result data directly and interactively on site. In addition, it also supports the binding of measurement tools to the robot, thereby realizing the automatic reporting of measurement results, further improving the accuracy and efficiency of data entry. It not only optimizes the inspection process, but also reduces human errors, providing strong technical support for the quality management of prefabricated concrete buildings.

[0052] With the continuous development of artificial intelligence and Internet of Things technology, building quality information collection robots have broad application prospects. The robot's interactive interface and user experience can be further optimized to make it more in line with the actual needs of inspectors. On the other hand, more advanced sensors and data analysis technologies can be integrated into the system to achieve more accurate and comprehensive quality inspections. In addition, it is also possible to consider integrating the system with other types of information management systems to form a robot-assisted system with a wider range of applications.

[0053] Those skilled in the art can change the above sequence without departing from the protection scope of the present invention.

[0054] Another embodiment of the present invention provides a building quality information collection robot, comprising an interaction layer, a data layer and a collection layer; The interactive layer is used for input and output of interactive content; The data layer includes a building quality information database, which stores characteristic values ​​of the detection task type, characteristic values ​​of the detection items, a list of detection items and a list of methods corresponding to the detection task, and a knowledge base; The acquisition layer is used to determine the corresponding detection item list according to the currently executed detection task; match the input content with the detection item feature value in the detection item list. The input content includes the measurement data of the detection tool. If there is a matching detection item feature value, the corresponding detection item is filled in according to the input content. If there is no matching detection item feature value, the corresponding method list is determined according to the currently executed detection task. According to the input content, it is determined whether to trigger the method in the method list. If the method is not triggered, the information related to the input content is selected from the knowledge base and sent to the interaction layer for output.

[0056] Regarding the building quality information collection robot in the above embodiment, its specific implementation method has been described in detail in the embodiment of the method, and will not be elaborated here.

[0057] Based on the same inventive concept, the present invention also provides a computer program product, including a computer program / instruction, which implements the aforementioned building quality information collection method when executed by a processor.

[0058] In this embodiment, by combining the detection tool with a natural language processing model, a database, etc., intelligent recognition and automatic reporting of detection data are achieved, thereby improving the efficiency and accuracy of detection data collection. In addition, the present invention can also provide relevant knowledge information during the detection process, providing auxiliary reference for staff who collect building quality information.

[0059] Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention shall still fall within the patent scope of the present invention.

Claims

1. A method for collecting building quality information, characterized in that: The following steps are involved: Determine the detection task currently being performed; The building quality information collection robot determines the corresponding inspection item list according to the inspection task currently being performed; The robot includes an AI module that performs natural language processing on input content; the robot is connected to a building quality information database; the database stores characteristic values ​​of detection task types, characteristic values ​​of detection items, a list of detection items and a list of methods corresponding to the detection task, and a knowledge base; The robot matches input content with detection item feature values ​​in the detection item list, wherein the input content includes measurement data of a detection tool; If there is a matching detection item feature value, fill in the corresponding detection item according to the input content, If there is no matching detection item feature value, the corresponding method list is determined based on the currently executed detection task, and whether to trigger the method in the method list is determined based on the input content. If the method is not triggered, information related to the input content is selected from the knowledge base for output.

2. The method according to claim 1, characterized in that Determine the detection task currently being performed, including the following steps: Obtain the information of the component to be detected, query the task library according to the information of the component to be detected, and determine the detection task currently being executed; The task library stores detection task information and association information between detection tasks and components.

3. The method according to claim 1, characterized in that The building quality information collection method further comprises the following steps: The robot interacts with the user in real time through an instant messaging tool; the AI ​​module includes an LLM model, a text vector library and a knowledge graph, and the robot performs semantic analysis on the input content through the AI ​​module, extracts the reported content, or triggers the method, or feeds back information related to the input content.

4. The method according to claim 1, characterized in that The building quality information collection method further comprises the following steps: The detection tool sends the measurement data to an edge device of the Internet of Things, and the edge device preprocesses the measurement data to obtain the measurement data in a preset format; Send the measurement data in a preset format to the building quality information collection robot.

5. The method according to claim 1, characterized in that Filling in the corresponding detection items according to the input content includes the following steps: Determine the data reporting method of the corresponding detection item, extract the reporting content from the input content, report the corresponding detection item in a preset data reporting method, and update the detection item list; Output the next detection item name based on the detection item list.

6. The method according to claim 1, characterized in that The building quality information collection method further comprises the following steps: Establishing a template library in the database, establishing prompt word templates for different scenarios in the template library, and configuring corresponding roles, introductions, skills and rules for each prompt word template; The template library stores association information between prompt word templates and components.

7. The method according to claim 1, characterized in that The building quality information collection method further includes constructing the knowledge base, and constructing the knowledge base specifically includes: Collect text content of professional books, papers, specifications, standards, cases and experiences in the field of building inspection; Convert the text content into multiple text vectors; According to data features of the multiple text vectors, the multiple text vectors are classified into different scenarios, and each text vector corresponds to at least one scenario.

8. The method according to claim 7, characterized in that Selecting information related to the input content from the knowledge base for output includes the following steps: Selecting a text vector and a scene related to the input content according to the similarity between the input content and the text vector; According to the selected text vector and scene, the corresponding text content is output.

9. A building quality information collection robot, characterized in that: Includes interaction layer, data layer and collection layer; The interaction layer is used for inputting and outputting interactive content; The data layer includes a building quality information database, which stores characteristic values ​​of detection task types, characteristic values ​​of detection items, a list of detection items and a list of methods corresponding to the detection tasks, and a knowledge base; The acquisition layer is used to determine the corresponding detection item list according to the currently executed detection task; match the input content with the detection item feature value in the detection item list, the input content includes the measurement data of the detection tool, if there is a matching detection item feature value, fill in the corresponding detection item according to the input content, if there is no matching detection item feature value, then determine the corresponding method list according to the currently executed detection task, and judge whether to trigger the method in the method list according to the input content; if the method is not triggered, select information related to the input content from the knowledge base and send it to the interaction layer for output.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the building quality information collection method described in any one of claims 1 to 8 is implemented.

Images