Tool management-oriented agent task arrangement system and decision-making method

Abstract

The invention discloses a tool management-oriented agent task arrangement system and decision-making method, and relates to the technical field of tool management, the system comprises a terminal layer and a business logic layer, the terminal layer comprises a client, a server, a database and a communication module; the business logic layer comprises an identity verification module, a tool receiving module, a tool returning module, an operation recording and inquiring module and an early warning and reminding module; according to the invention, the efficiency and accuracy of tool management are improved through highly integrated multi-modal interaction and intelligent recognition technologies, and the system realizes rapid verification of user identity and accurate use and return of tools through fusion of face recognition, voice recognition and visual image recognition technologies, so that the system has high practicability. The user can complete the borrowing and returning process of the tool only through a simple voice instruction or visual operation, the tedious steps of traditional manual registration are simplified, and the working efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tool management, in particular to an agent task scheduling system and decision method for tool management. BACKGROUND

[0002] With the rapid development of intelligent manufacturing, tool management as a key link in the production process directly affects the overall production efficiency, and the efficiency and accuracy of the traditional tool management method rely on manual registration, paper account or simple RFID technology, which has the problems of complicated operation, easy error and difficult to track the tool state in real time, especially in the multi-tool and high flow operation scene, manual management is easy to cause tool loss, wrong taking or misuse due to negligence, and further cause production delay, safety hazard and resource waste, at the same time, the traditional technology lacks dynamic monitoring and intelligent analysis ability to tool use process, and it is difficult to meet the demand of modern production for efficient, transparent and traceable management, under this background, it is an urgent need of the industry to develop a system that can automatically and intelligently manage tools.

[0003] The traditional tool management technology mainly has the following defects: first, manual registration and paper account are easy to cause tool flow to be unknown due to human negligence or recording error, especially in the scene of multiple types of tools and frequent taking, the management efficiency is low and disputes are easy to be caused; second, although single RFID technology can realize tool identification, it cannot monitor the tool use state in real time, and has poor environmental adaptability; third, the traditional system lacks multi-modal interaction ability, users need to operate through fixed terminal or manual input, which is complicated and easy to make mistakes; finally, the traditional early warning mechanism is single, and cannot adjust the reminding strategy combined with the importance of tools and the dynamic factors of user historical behavior, resulting in insufficient timeliness and accuracy of early warning, these deficiencies make it difficult for the traditional technology to meet the requirements of modern production for efficient, safe and intelligent tool management. SUMMARY

[0004] The purpose of the present application is to make up for the deficiencies of the prior art, and to provide an agent task scheduling system and decision method for tool management, which realizes the full-process automation and intelligentization of tool taking, returning, monitoring and early warning by integrating multi-modal interaction and intelligent decision technology, and the system has dynamic permission fusion and feature adaptation ability, which can adapt to complex environment and reduce human errors; at the same time, combined with the importance of tools and the grading early warning mechanism of user behavior factors, the tool loss and misuse are effectively prevented.

[0005] To solve the above technical problems, the present application provides the following technical solutions: on the one hand, an agent task scheduling system for tool management, the system comprising a terminal layer and a business logic layer:

[0006] The terminal layer comprises a client, a server, a database and a communication module.

[0007] The business logic layer comprises an identity verification module, a tool taking module, a tool returning module, an operation record and query module, and an early warning and reminding module.

[0008] Further, in the terminal layer, the specific content of the client, the server, the database, and the communication module is:

[0009] The client: a touch screen terminal located locally on the tool cart, responsible for front-end face recognition collection, voice wake-up and command recognition, and visual image collection;

[0010] The server: deployed on a cloud server or an edge server, carrying a multi-modal large model for face / tool recognition and command analysis, and managing a database;

[0011] The database: used to store user identity information, tool and drawer mapping information, taking / returning records, early warning rules, and logs;

[0012] The communication module: the client and the server realize data interaction and model inference requests through RESTful API.

[0013] Further, in the terminal layer, the multi-modal large model in the server is constructed, and its calculation formula is: wherein, is the identity matching output, with a value range of: 0 represents complete mismatch, 1 represents complete match, when identity verification is passed, re-collection of the face or contact with the administrator is prompted, is the feature matching degree of the visual model to the input face image , is the light adaptation factor, is the historical face consistency coefficient, is the identity recognition weight, is the return tool recognition output, with a value range of: when tool type recognition is passed, is the tool name transcription matching degree of the voice model to the input voice , is the noise adaptation factor, is the historical taking correlation coefficient, is the voice recognition weight, is the return tool recognition output, with a value range of: when return tool type recognition is passed, ​​is a type matching degree of the visual model to the input tool image, is a placement angle adaptation factor, is a historical return correlation coefficient, is a visual recognition weight.

[0014] Further, in the business logic layer, the specific content of the identity verification module, the tool taking module, the tool returning module, the operation record and query module, and the early warning and reminding module is:

[0015] The identity verification module: the client collects a face image through a built-in camera and sends it to the server. The server compares feature vectors based on a multi-modal large model to verify identity and authority. After verification, an authorization token is generated and returned to the client, and the client enters an available state.

[0016] The tool taking module: the user speaks the tool name through voice, the client calls a dynamic context adaptation voice transcription model for preliminary transcription, and judges whether to upload to the server according to the confidence of the transcription result. A multi-modal large model is used for recognition. After recognizing the tool, the corresponding drawer number is obtained and unlocked. The drawer weight sensor detects the weight reduction of the tool, and then records the taking information and synchronizes it to the server database.

[0017] The tool returning module: the user returns the tool through vision, the client collects the tool placement area image, calls a dynamic feature adaptation visual detection model to detect the tool type, and judges whether to upload to the server according to the confidence of the detection result. A multi-modal large model is used for recognition. After recognizing the tool, the corresponding drawer number is obtained and unlocked. The user is prompted to put in the tool, and the drawer weight sensor detects the weight increase of the tool, and then records the returning information and synchronizes it to the server database.

[0018] The operation record and query module: the server maintains the tool state and operation log in real time, and the client provides a query interface. After the user inputs the query condition, the server is requested and the query result is displayed.

[0019] The early warning and reminding module: the client starts a timer after the drawer is closed. If the user does not leave within a set time, a door closing timeout reminder is given. The server periodically scans the database and gives a timeout unreturned tool reminder.

[0020] Further, in the business logic layer, the dynamic permission fusion face verification model in the tool taking module is constructed, and the calculation formula is: wherein is the final transcription confidence of the dynamic context adaptation voice transcription model, with a value of 0~1, is the voice signal intelligibility score, is the matching degree of the transcription text and the tool vocabulary, is the consistency with the user's historical voice transcription, is the context correlation factor, , , is the dynamic weight, , represents that the model verification is passed, and there is no need to upload to the server end; , the cloud multi-modal large model secondary recognition is triggered.

[0021] Further, in the business logic layer, the construction of the dynamic feature adaptive visual detection model in the tool return module has a calculation formula: , wherein is the final detection confidence of the dynamic feature adaptive visual detection model, is the matching degree of the tool visual feature and the template library, is the completeness score of the tool in the image, is the deviation degree of the tool placement position and the recognition area center, is the light compensation factor, is the feature consistency with the historical return image of the tool, , , is the dynamic weight, and , represents that the local model verification is passed, and there is no need to upload to the cloud end; if , the cloud large model secondary recognition is triggered.

[0022] Further, in the early warning and reminding module, the closed door timeout reminder: the client starts a timer after the drawer is closed, combines the tool importance, the user's historical operation record, and the timeout time to determine the priority of the reminder, and if the user does not leave after the set time is exceeded, the sound and light reminders are triggered according to the priority.

[0023] Further, in the early warning and reminding module, the timeout unreturned reminder: the server scans the database regularly every day, combines the tool importance, the user's performance, and the timeout time to determine the priority of the reminder, sends a short message reminder to the user according to the priority for the tool that exceeds the preset borrowing period, and pops up a warning information on the client, and the early warning record is stored synchronously to the server log.

[0024] On the other hand, the agent task arrangement decision method for tool management has the following specific steps:

[0025] S100, user identity verification: the user performs identity verification through the client, the client collects a face image and sends it to the server end, and the server end calculates using a multi-modal large model, if If yes, the identity verification is passed, the client enters an available state, otherwise, the user is prompted to re-collect the face or contact the administrator;

[0026] S200, tool taking operation: after the identity verification is passed, if the user needs to take a tool, the user speaks the tool name through voice, the client calls the dynamic context adaptive speech transcription model to calculate , if , the local recognition is passed, the corresponding drawer number is obtained and is unlocked, the weight sensor in the drawer detects that the weight of the tool is reduced, the taking information is recorded and is synchronized to the server database; if , the server is uploaded, the server uses the multi-modal large model to calculate , if , the recognition is passed, the subsequent taking operation is executed, otherwise, the user is prompted to re-input;

[0027] S300, tool returning operation: if the user needs to return the tool, the tool is placed in the specified area, the client collects the tool placement area image, calls the dynamic feature adaptive visual detection model to calculate , if , the local recognition is passed, the corresponding drawer number is obtained and is unlocked, the user is prompted to put the tool in, the weight sensor in the drawer detects that the weight of the tool is increased, the returning information is recorded and is synchronized to the server database; if , the server is uploaded, the server uses the multi-modal large model to calculate , if , the recognition is passed, the subsequent returning operation is executed, otherwise, the user is prompted to re-place;

[0028] S400, record query management: the operation record and query module maintains the tool state and operation log in real time, the user can query the related information through the client;

[0029] S500, early warning and reminder triggering: the early warning and reminder module starts a timer after the drawer is closed, if the user does not leave within the set time, the sound and light reminders are triggered according to the priority; the server scans the database regularly every day, sends a short message to the user according to the priority for the tool that is not returned within the time, pops up warning information on the client, and the early warning record is stored in the server log synchronously.

[0030] Compared with the prior art, the intelligent agent task scheduling system and decision method for tool management have the following beneficial effects:

[0031] I. The present application improves the efficiency and accuracy of tool management through highly integrated multi-modal interaction and intelligent recognition technology. The system realizes rapid verification of user identity and accurate borrowing and returning of tools by fusing face recognition, voice recognition and visual image recognition technology. Users can complete the tool borrowing and returning process through simple voice commands or visual operations, simplifying the cumbersome steps of traditional manual registration and improving work efficiency. Meanwhile, the system's built-in dynamic context adaptation and feature adaptation model can automatically adjust the recognition strategy according to the voice clarity and lighting conditions in different scenarios, ensuring high recognition rate in various complex environments and enhancing the robustness and adaptability of the system. This intelligent management method not only reduces human errors, but also improves the overall safety and reliability of tool management.

[0032] II. The present application effectively prevents tool loss and misuse through real-time monitoring and data analysis. The system automatically starts a timer after the drawer is closed, intelligently determines the priority of the reminder based on the importance level of the tool and the user's historical operation records, and timely triggers an audible and visual reminder when the user does not leave after a timeout, avoiding the tool being accidentally taken out of the work area. Meanwhile, the server regularly scans the database to accurately identify tools that have not been returned on time, and sends reminder information to the user through SMS and client pop-up windows based on the importance of the tool and the user's compliance factors, ensuring that the tool is returned on time and reducing resource waste and safety hazards. This comprehensive early warning and reminder system not only improves the refinement level of tool management, but also saves a large amount of additional costs for enterprises due to tool loss or misuse.

[0033] Other advantages, objects, and features of the present application will be apparent to those skilled in the art from the following specification, and it is intended to cover any alternatives, modifications, or equivalents included within the scope of the following claims. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creating any creative labor.

[0035] Figure 1 Terminal layer structure diagram of the intelligent agent task scheduling system for tool management;

[0036] Figure 2 Business logic layer module diagram of the intelligent agent task scheduling system for tool management;

[0037] Figure 3 Flowchart of an agent task orchestration decision-making method for tool-oriented management;

[0038] Figure 4 A schematic diagram illustrating the application logic of multimodal large models in tool management;

[0039] Figure 5 A diagram illustrating the user authentication process;

[0040] Figure 6 This is a schematic diagram illustrating the workflow for tool issuance and return. Detailed Implementation

[0041] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0042] Example 1:

[0043] System application scenarios for factory tool carts

[0044] To improve the efficiency of tool management in the workshop, an automobile manufacturing plant deployed an intelligent agent task orchestration system for tool management. The system's terminal layer and business logic layer work collaboratively in actual operation, with the specific process as follows:

[0045] Terminal layer operation: The tool cart is equipped with a local touch screen client with a built-in camera, microphone and weight sensor; the server is deployed on the factory edge server to carry multimodal large models; the database stores the facial information of workshop employees, the correspondence between wrenches, screwdrivers and drawers, and historical use / return records; the client and server transmit data in real time through RESTful API.

[0046] Business logic layer operation:

[0047] Identity verification: Employees scan their faces in front of the tool cart client. The client's camera captures the facial image and sends it to the server. The server uses a multimodal big data model to compare the facial information. The calculation formula for the multimodal big data model is as follows: ; ; ,in, This is the identity matching output, with a value range of: 0 indicates no match at all, 1 indicates a perfect match, when At that time, the identity verification is deemed successful. The system will then prompt you to re-collect your face or contact the administrator. It is a visual model of the input face image ( Feature matching degree It is the light adaptation factor. It is the historical facial consistency coefficient. It is the identity recognition weight. This is the output of the return tool, with a value range of: ,when At that time, the tool type identification passed. It is a speech model for input speech Tool name transcription matching degree It is a noise adaptation factor. It is the historical usage correlation coefficient. It is speech recognition weight. This is the output of the return tool, with a value range of: ,when At that time, the identification of the return tool type was successful. It is the type matching degree of the visual model to the input tool image. Placement angle adaptation factor It is the historical return correlation coefficient. It uses visual recognition weights to compare employee faces with information stored in the database. Once the verification is successful, an authorization token is generated, and the client unlocks and enters a usable state.

[0048] Tool Requisition: An employee says "I need a size 3 wrench" via voice. The client calls a dynamic context-adaptive speech-to-text model to initially recognize the speech content. The calculation formula for dynamic permission fusion with facial verification model is as follows: ,in This is the final transcription confidence score of the dynamic context-adaptive speech-to-text model, with a value ranging from 0 to 1. It is a speech signal clarity score. It is the degree of matching between the transcribed text and the tool's vocabulary. It is consistent with the user's historical speech transcription. It is a contextual factor. , , It is a dynamic weight. , This means the model has passed validation and does not need to be uploaded to the server. If the recognition fails, the cloud-based multimodal large model will trigger secondary recognition. If the recognition confidence level meets the local verification standard, the drawer number corresponding to wrench No. 3 will be directly obtained and the drawer will be unlocked. The weight sensor inside the drawer will detect that the wrench has been taken out (weight reduction), automatically record the time of receipt and employee information, and synchronize them to the server database. If the local recognition fails, the voice information will be uploaded to the server. The server will conduct secondary recognition through the receipt recognition model in the multimodal large model. After confirming that there are no errors, the unlocking and recording operations will be performed.

[0049] Tool return: the employee places the used No. 3 wrench in the designated area of the tool cart, the client camera captures the tool image, and the dynamic feature adaptation visual detection model is called for recognition. The calculation formula of the dynamic feature adaptation visual detection model is: wherein is the final detection confidence of the dynamic feature adaptation visual detection model, is the matching degree of the tool visual feature and the template library, is the completeness score of the tool in the image, is the deviation degree of the tool placement position from the center of the recognition area, is the light compensation factor, is the feature consistency with the historical return image of the tool, , , is the dynamic weight, and , represents that the local model verification is passed, and there is no need to upload to the cloud; if , the cloud large model secondary recognition is triggered, and if the recognition confidence meets the local verification standard, the corresponding drawer is automatically unlocked and the tool is prompted to be placed; the weight sensor in the drawer detects the weight increase, records the return time and synchronizes it to the database, if the local recognition fails, the image is uploaded to the server, the server performs secondary recognition through the return recognition model in the multi-modal large model, and the return process is completed after confirmation.

[0050] Operation record and query: the server updates the tool state (such as “No. 3 wrench has been taken” and “No. 5 screwdriver is in stock”) and operation log in real time, and the employee can query the current position of a tool and the information of the historical taking record through the client.

[0051] Warning and reminder: the client timer starts after the employee returns the tool without closing the drawer in time, and sound and light reminders are triggered after the set time is exceeded; if a special wrench taken by an employee is not returned for more than 7 days, the server scans the database every day, determines the priority according to the importance of the tool, sends a short message to the employee, and pops up a warning information on the tool cart client, which is recorded and synchronized to the server log.

[0052] In summary, the above system embodiment realizes the functions of identity verification, visitor management and abnormality processing of the intelligent access control through the cooperation of the terminal and the server, relying on the identity matching model and the temporary permission generation model core models. Each model plays a role in confirming identity, standardizing permission and ensuring system stability in the process, which not only improves the efficiency of community access, but also strengthens security management through the record tracing mechanism, and overall forms an intelligent access control management system that takes into account convenience and security.

[0053] Embodiment two:

[0054] Method application scenario for maintenance workshop tool management.

[0055] An automotive maintenance workshop uses an agent task scheduling decision method for tool management to manage maintenance tools (such as wrenches, torque wrenches, and fault detectors). The specific steps are as follows:

[0056] S100: User authentication: the maintenance personnel perform identity verification on the client of the workshop tool cabinet, the client camera captures their face image and sends it to the server, and the server calls the multi-modal large model to compare the face information. The calculation formula of the multi-modal large model is: ; ; If the verification is passed, the client is unlocked to enter the available state, and the maintenance personnel can perform subsequent operations; if it is not passed, the client prompts to re-collect the face or contact the workshop administrator. This step can strictly control the use of tool permissions and ensure that only qualified maintenance personnel can take tools to ensure the professionalism of maintenance work.

[0057] S200: Tool taking operation: after identity verification, if the maintenance personnel need to take tools, they speak the tool name to the client through voice, such as "I want to take a 15 wrench". The client calls the dynamic context adaptive speech recognition model to calculate the recognition confidence. The calculation formula of the dynamic permission fusion face verification model is: If it meets the local verification standard, the corresponding drawer number of the tool is directly obtained and the drawer is unlocked. The weight sensor in the drawer detects that the tool is taken out (the weight is reduced), records the taking information (such as taking time, maintenance personnel name, tool model) and synchronizes it to the server database. If the local confidence does not meet the standard, the client uploads the voice information to the server, and the server calls the taking recognition model in the multi-modal large model to calculate the recognition degree. If it meets the standard, the subsequent taking operation is performed, otherwise the maintenance personnel are prompted to speak the tool name again. This operation allows maintenance personnel to quickly get the required tools through convenient voice interaction, improving the continuity of maintenance work.

[0058] S300: Tool return operation: after the maintenance personnel finish using the tool, they place it in the designated return area of the tool cabinet. The client camera captures the tool image and calls the dynamic feature adaptive visual detection model to calculate the detection confidence. The dynamic feature adaptive visual detection model is constructed, and its calculation formula is: If the local verification criteria are met, the corresponding drawer is unlocked and the maintenance personnel are prompted to place the tool. The weight sensor in the drawer detects the weight increase, records the return information and synchronizes it to the database. If the local confidence does not meet the standard, the client uploads the image to the server, the server calls the return recognition model in the multi-modal large model to calculate the recognition degree. If the standard is met, the return process is completed, otherwise the maintenance personnel are prompted to reposition the tool. This process ensures that the tool can be accurately returned to the corresponding position, making it easy for subsequent personnel to find and use.

[0059] S400: Record query management: The operation record and query module maintains the state of the tool (such as "torque wrench has been returned" "fault detector is in stock") and operation log in real time. Maintenance personnel or management personnel can query relevant information through the client, such as querying the historical borrowing record of a fault detector to understand its usage, facilitating the maintenance and maintenance of the tool.

[0060] S500: Early warning trigger: The maintenance personnel return the tool without closing the drawer. The client starts a timer after the drawer is closed. If the maintenance personnel do not leave within the set time, the importance of the tool (such as high-priced fault detector, high-precision requirement, higher importance than ordinary wrench), historical operation record of the maintenance personnel and overtime duration are combined to determine the priority of the reminder, triggering the sound and light reminder of the tool cabinet. The server scans the database daily. If it finds that a torque wrench borrowed by a maintenance personnel has not been returned for more than a predetermined period, the importance of the tool, the maintenance personnel's performance and the overtime duration are combined to determine the priority, send a short message to the maintenance personnel, and pop up a warning information on the tool cabinet client. The early warning record is synchronized to the server log. Through timely reminders, the risk of tool loss or damage due to long-term borrowing is reduced, ensuring the normal development of maintenance work.

[0061] In summary, the method is applied in the maintenance workshop, starting with identity verification, through dynamic context adaptation of speech transcription model and dynamic feature adaptation of visual detection model to realize precise management of tool borrowing and returning, record query to facilitate tool maintenance, early warning to reduce overdue and damage of tools. Each step relies on the corresponding model, which not only ensures the professionalism and safety of tool use, but also improves the continuity of maintenance work through efficient process, providing a reliable solution for workshop tool management.

[0062] The above merely describes the preferred embodiments of the present application, and is not intended to limit the present application in any form. Although the present application has been disclosed with the preferred embodiments as above, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content to obtain equivalent embodiments with equivalent changes, as long as the changes or modifications do not deviate from the technical solution of the present application. Any modification, change, equivalent change and modification of the above embodiments made according to the technical essence of the present application still belong to the scope of the technical solution of the present application.

Claims

1. An agent-based task orchestration system for tool management, characterized in that, The system consists of a terminal layer and a business logic layer: The terminal layer includes: a client, a server, a database, and a communication module; The business logic layer includes: an identity verification module, a tool requisition module, a tool return module, an operation record and query module, and an early warning and reminder module.

2. The intelligent agent task orchestration system for tool management according to claim 1, characterized in that, The specific contents of the client, server, database, and communication module in the terminal layer are as follows: The client is a touchscreen terminal located on the tool vehicle, responsible for facial recognition front-end acquisition, voice wake-up and command recognition, and visual image acquisition. The server-side component is deployed on a cloud server or edge server, carrying out multimodal large-scale model recognition for face / tool ​​recognition and command parsing, and managing the database. The database is used to store user identity information, tool and drawer mapping information, usage / return records, warning rules, and logs. The communication module: The client and server communicate via RESTful API to exchange data and make model inference requests.

3. The intelligent agent task orchestration system for tool management according to claim 2, characterized in that, In the terminal layer, the calculation formula for constructing the multimodal large model in the server is as follows: ; ; ,in, This is the identity matching output, with a value range of: 0 indicates no match at all, 1 indicates a perfect match, when At that time, the identity verification is deemed successful. The system will then prompt you to re-collect your face or contact the administrator. It is a visual model of the input face image ( Feature matching degree It is the light adaptation factor. It is the historical facial consistency coefficient. It is the identity recognition weight. This is the output of the return tool, with a value range of: ,when At that time, the tool type identification passed. It is a speech model for input speech Tool name transcription matching degree It is a noise adaptation factor. It is the historical usage correlation coefficient. It is speech recognition weight. This is the output of the return tool, with a value range of: ,when At that time, the identification of the return tool type was successful. It is the type matching degree of the visual model to the input tool image. Placement angle adaptation factor It is the historical return correlation coefficient. It is the visual recognition weight.

4. The intelligent agent task orchestration system for tool management according to claim 1, characterized in that, The specific contents of the identity verification module, tool requisition module, tool return module, operation record and query module, and early warning and reminder module in the business logic layer are as follows: The identity verification module works as follows: the client captures facial images through the built-in camera and sends them to the server. The server compares feature vectors based on a multimodal large model to verify identity and permissions. After successful verification, an authorization token is generated and returned to the client, and the client enters the available state. The tool requisition module works as follows: the user speaks the name of the tool, the client calls the dynamic context-adaptive speech-to-text model to perform preliminary transcription, and determines whether to upload it to the server based on the confidence level of the transcription result. The multimodal large model is used for recognition. After the tool is recognized, the corresponding drawer number is obtained and the drawer is unlocked. If the weight sensor in the drawer detects that the tool is lighter, the requisition information is recorded and synchronized to the server database. The tool return module works as follows: the user returns the tool visually, the client collects an image of the tool placement area, calls a dynamic feature-adaptive visual detection model to detect the tool type, determines whether to upload to the server based on the confidence level of the detection result, uses a multimodal large model for identification, after the tool is identified, obtains the corresponding drawer number and unlocks it, prompts the user to put the tool back in, the weight sensor inside the drawer detects an increase in the weight of the tool, records the return information and synchronizes it to the server database; The operation record and query module: The server maintains the tool status and operation log in real time, and the client provides a query interface. After the user enters the query conditions, the server requests and displays the query results. The warning and reminder module: The client starts a timer after the drawer is closed. If the user is not detected leaving after the set time, a door-closing timeout reminder is issued. The server periodically scans the database and issues timeout reminders for tools that have not been returned.

5. The intelligent agent task orchestration system for tool management according to claim 4, characterized in that, In the business logic layer, the dynamic permission fusion face verification model constructed in the tool requisition module is calculated using the following formula: ,in This is the final transcription confidence score of the dynamic context-adaptive speech-to-text model, with a value ranging from 0 to 1. It is a speech signal clarity score. It is the degree of matching between the transcribed text and the tool's vocabulary. It is consistent with the user's historical speech transcription. It is a contextual factor. , , It is a dynamic weight. , This means the model has passed validation and does not need to be uploaded to the server. This triggers secondary recognition of the multimodal large model in the cloud.

6. The intelligent agent task orchestration system for tool management according to claim 4, characterized in that, In the business logic layer, the dynamic feature adaptation visual detection model constructed in the tool return module is calculated using the following formula: ,in It is the final detection confidence score of the dynamic feature-adaptive visual detection model. It refers to the degree of matching between the visual features of the tool and the template library. It is a tool for scoring the completeness of an image. It is the deviation between the tool placement position and the center of the recognition area. It is a light compensation factor. It is consistent with the characteristics of the images historically recovered by the tool. , , It is a dynamic weight, and , This indicates that the local model has passed validation and does not need to be uploaded to the cloud; if This triggers secondary recognition by the large cloud-based model.

7. The intelligent agent task orchestration system for tool management according to claim 4, characterized in that, The warning and reminder module includes a door closing timeout reminder: the client starts a timer after the drawer is closed, and determines the reminder priority based on the importance of the tool, the user's historical operation records, and the timeout duration. If the user is detected not to have left after the set time has elapsed, an audio-visual reminder is triggered according to the priority.

8. The intelligent agent task orchestration system for tool management according to claim 4, characterized in that, The warning and reminder module includes an overdue return reminder: The server scans the database daily and determines the reminder priority based on the importance of the tool, the user's fulfillment of the agreement, and the overdue duration. For tools that have exceeded the preset borrowing period, SMS reminders are sent to the user according to priority, and a warning message pops up on the client. The warning record is also stored in the server log.

9. A tool-management-oriented agent task orchestration decision-making method, wherein the method is used in the tool-management-oriented agent task orchestration system according to any one of claims 1-8, characterized in that, The specific steps of this decision-making method are as follows: S100, User Authentication: Users authenticate themselves through the client. The client captures facial images and sends them to the server. The server uses a multimodal large model to perform calculations. ,like If the identity verification is successful, the client will be available; otherwise, the client will be prompted to re-collect the face or contact the administrator. S200, Tool Requisition Operation: After identity verification, if a user needs to requisition a tool, they can speak the tool's name, and the client will invoke a dynamic context-adaptive speech-to-text model to calculate the tool's name. ,like If local identification is successful, the corresponding drawer number is obtained and the drawer is unlocked. The weight sensor inside the drawer detects a reduction in the weight of the tools, records the usage information, and synchronizes it to the server database. Then it is uploaded to the server, and the server uses the multimodal large model to calculate... ,like If the identification is successful, proceed with the subsequent requisition process; otherwise, prompt for re-entry. S300, Tool Return Operation: If the user needs to return a tool, they place it in a designated area. The client captures an image of the area where the tool was placed and calls the dynamic feature adaptation visual detection model to calculate... ,like If local identification is successful, the corresponding drawer number is obtained and the drawer is unlocked. The user is prompted to put the tool in. The weight sensor inside the drawer detects an increase in the weight of the tool, records the return information, and synchronizes it to the server database. Then it is uploaded to the server, and the server uses the multimodal large model to calculate... ,like If the identification is successful, the subsequent return operation will be performed; otherwise, the user will be prompted to place the item again. S400, Record Query Management: The operation record and query module maintains the tool status and operation log in real time, and users can query relevant information through the client; S500, Warning and Reminder Trigger: The warning and reminder module starts a timer after the drawer is closed. If the user is not detected to have left after the set time, an audio-visual reminder is triggered according to priority. The server scans the database daily and sends SMS reminders to users for tools that have not been returned within the time limit according to priority. Warning information also pops up on the client and the warning records are synchronously stored in the server log.

Images