Intelligent decision-making method for machine manufacturing production management based on AI Agent

Abstract

The invention discloses an intelligent decision-making method for machine manufacturing production management based on an AI Agent, and aims to realize efficient natural language interaction and intelligent data analysis by replacing complex operation of traditional industrial software with an intelligent agent. According to the method, a natural language query request of a user is analyzed through an intention recognition module, a structured database query statement is automatically generated, and a background data system is called to obtain key business data. The system automatically completes visual display of data, index attribution analysis and abnormal early warning, and generates auxiliary operation decision suggestions in combination with a rule engine. Multiple rounds of dialogues and context management are supported, the use threshold of an enterprise management layer is remarkably reduced, and the decision-making efficiency and the information response speed are improved. The method is suitable for a plurality of industries such as the manufacturing industry and has wide popularization and application values.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of industrial software and artificial intelligence, in particular to an intelligent decision-making method and business operation method for mechanical manufacturing production management based on an AI Agent, which specifically constructs an intelligent assistant Agent for enterprise executives (such as bosses), replaces the operation process of traditional industrial software, realizes functions such as report acquisition, data analysis, index monitoring and operation decision support, and improves the efficiency and intelligent level of industrial enterprises in production management and strategic decision-making. BACKGROUND

[0002] With the development of intelligent manufacturing, enterprises have increasingly high requirements for production efficiency, management decision efficiency and data visualization analysis capability. Although traditional industrial software (such as ERP, MES, WMS, SCM, etc.) has been widely used in production and manufacturing enterprises, these systems generally have problems such as complex operation, high learning cost, low cross-system linkage efficiency, and dependence on professional personnel for data analysis. When enterprise managers acquire reports, issue orders, perform data analysis or manage procurement and sales processes, they often need to rely on multiple systems and multiple roles to complete the operation, which seriously hinders the response speed and management efficiency.

[0003] In recent years, with the development of natural language processing (NLP), semantic understanding, NL2SQL (natural language to structured query language) and multi-modal chart generation technology, it is possible to reconstruct the human-computer interaction mode of industrial software. AI Agent (Intelligent Agent) is a task-oriented autonomous interaction system that can understand natural language, call background data, perform analysis tasks and provide feedback in a natural way. However, there is currently a lack of a complete and generalizable technical solution that can systematically cover and replace the key operation processes of various traditional industrial software in enterprises and provide dialog-based, intelligent and visual decision support capabilities.

[0004] Therefore, a "boss assistant Agent" for enterprise senior managers is proposed, which acquires user intent through natural language dialogue and automatically performs tasks such as data query, index analysis, attribution diagnosis and chart generation, and realizes the reconstruction of the operation mode of traditional industrial software. The Agent not only can complete data query and display of traditional BI systems, but also has the ability to track changes in key indicators, provide early warning information and assist in operation decision-making, greatly improving the decision-making efficiency of senior management personnel and the intelligent level of enterprise operation. SUMMARY

[0005] The application provides an AI Agent-based intelligent decision-making method and business operation method for mechanical manufacturing production management, aiming to replace the complex operation process of traditional industrial software with an intelligent agent, and improve the efficiency and intelligent level of enterprise management in information acquisition, data analysis and decision-making. The method constructs a "boss assistant agent" for enterprise executives, which understands user needs through natural language dialogue, and automatically performs data query, chart generation, attribution analysis and decision-making suggestions, etc., to break through the whole link of data acquisition and decision support.

[0006] In the application, users can directly ask the boss assistant agent about business-related questions in daily language, such as "help me check the recent profit changes" or "what is the reason for the decline in sales". The agent first performs semantic analysis on the user input through a natural language processing model, identifies the core intent and related parameters, and automatically generates a corresponding data query statement in combination with enterprise knowledge graph or database structure information. This process does not require users to understand complex data table structures or master professional BI system operation logic, thereby significantly reducing the threshold for data acquisition.

[0007] The query statement generated by the intelligent agent will be used to extract structured data from the enterprise backend system. After standardization processing, the extracted data will automatically trigger the generation process of visual charts. The system will intelligently select appropriate chart types such as trend charts, comparison charts, and proportion charts based on the query content, allowing users to grasp key indicator changes at a glance.

[0008] The application also introduces an attribution analysis mechanism based on business logic and historical data. When the user focuses on indicators that fluctuate, the intelligent agent can further analyze the main causes of the changes, such as the rise in the price of a raw material, the shutdown of a key production line, or the reduction of regional orders, and present the impact factors to the user in the form of explanatory text or highlighted charts. Based on this, the system can configure warning rules to detect abnormalities in key indicators and provide immediate reminders, helping enterprises to identify potential problems in advance.

[0009] In addition, the boss assistant agent in the application has certain strategic judgment ability, and can provide targeted business recommendations based on data analysis results and established enterprise rule models. For example, when detecting slow inventory turnover, the system can suggest adjusting the procurement rhythm or launching promotional activities; when the production yield decreases, potential human or equipment problems can be inferred, and checking the corresponding production line is suggested.

[0010] The entire interaction process supports context preservation and multi-round dialogue capability, users can add conditions, change time ranges or delve into the details of a certain indicator in continuous queries, thereby realizing a human-like dialogue-style data exploration experience.

[0011] Through the above scheme, the application effectively breaks through the limitations of traditional industrial software in high use threshold, low response efficiency, and difficult data interpretation. The intelligent agent not only significantly improves the information acquisition and decision-making ability of the enterprise top management, but also provides a new method path for the intelligent upgrading of industrial management. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 Workflow diagram of the boss assistant agent. DETAILED DESCRIPTION

[0013] To make the technical solutions of the application clearer and more explicit, the "boss assistant agent" proposed by the application is described in detail below in conjunction with specific embodiments. This embodiment is based on the integrated application of current natural language understanding, natural language generation, NL2SQL (large language model) technology, data chart visualization module, index attribution analysis algorithm model, and enterprise knowledge graph. Through multi-module collaboration, a human-computer interaction intelligent system is realized to replace the operation of traditional industrial software.

[0014] In this embodiment, the boss assistant agent system generally includes five core modules: intent recognition module, semantic analysis and NL2SQL module, data interface access module, chart generation and visualization module, and attribution analysis and decision recommendation module. In addition, the system also includes a conversation context management module and a permission control subsystem to enhance user experience and ensure data security.

[0015] In terms of system architecture, this embodiment adopts a microservice architecture for deployment. Each core module runs as an independent service, and communication between modules is achieved through an asynchronous message queue to ensure the scalability and high availability of the system. The system is deployed in a cloud computing environment that supports GPU inference. The front end uses a Web interface and a mobile app interface, and users can interact through natural language text or voice.

[0016] The user first inputs a natural language request through the interface. The input content is processed by the intent recognition module. This module uses a BERT or RoBERTa pre-trained language model as the underlying semantic encoder, combined with fine-tuned multi-classifiers for intent classification, and identifies key slot parameters such as query targets, time ranges, and data dimensions. For example, for the request "I want to see the sales trend in the last three months", the system can identify "view trend" as the intent, "sales" as the query object, and "the last three months" as the time parameter, and output them in a structured format.

[0017] The structured semantic information is transmitted into the NL2SQL module. Based on the metadata structure of the enterprise database, the module constructs a semantic-table structure mapping atlas, and uses a self-developed large language model to generate SQL statements. The input of the model is the parsed semantic structure and the database metadata vector, and the output is an executable standard SQL query statement.

[0018] The query statement is sent to the data interface access module, which provides unified abstract access capability to enterprise databases or data platforms. To improve real-time response speed, the embodiment introduces a caching strategy and an asynchronous preloading mechanism at the query layer, and constructs a multi-level index for frequently used data indicators.

[0019] When the data is returned, the system transmits it into the chart generation and visualization module, which automatically generates charts based on the open-source ECharts graphics engine or commercial BI graphics rendering components. In terms of chart type selection, the system uses a Heuristic rule weighted chart recommendation engine to automatically match appropriate charts such as line charts, bar charts, or stacked charts, based on indicator types (such as time series, classification comparison, numerical distribution) and user preferences. Charts support user interaction, such as zooming, drilling, and conditional filtering.

[0020] Based on data display, the system enters the attribution analysis and decision-making suggestion module. This module uses various attribution analysis methods, including difference decomposition algorithms (such as Shapley value decomposition) for KPI fluctuations, influence factor regression analysis models, and causal graph reasoning logic, to explain abnormal changes in data. For example, when sales decline, the system can identify that it is caused by a sharp drop in orders in the East China region, and mark the contribution of this factor to the overall fluctuation. If the system detects that the abnormal fluctuation exceeds the threshold, it will automatically trigger an early warning logic and notify the user through message pushing.

[0021] Further, combined with the enterprise's preset rule engine and strategy knowledge base, the system can generate auxiliary decision-making suggestions. For example, if it detects that the production capacity of a certain production line has decreased for three consecutive days and the yield fluctuation exceeds 20%, the system will refer to historical cases to suggest personnel inspection or equipment inspection, and attach supporting data.

[0022] During the entire interaction process, the conversation context management module maintains the user's historical intent, queried conditions, and presented chart state, supporting the user to ask follow-up questions in a continuation manner, such as "What about the same period last year?" and "Help me only look at the South China region." The system associates historical semantic information based on the Session Token, and automatically completes the omitted content based on the context, ensuring a continuous dialogue experience.

[0023] In order to guarantee enterprise data security and system stability, the embodiment introduces a permission control subsystem and an operation audit mechanism. All SQL execution operations are subjected to semantic whitelist verification and parameter injection protection to avoid unauthorized queries and injection risks. User access control is based on the RBAC permission model, which assigns data access ranges according to roles to ensure that management personnel at different levels can only access authorized content.

[0024] Through the above multi-module collaborative implementation scheme, the boss assistant Agent of the application realizes complete link reconstruction from natural language to data insight, not only reduces the operation threshold of traditional industrial software, but also greatly improves the decision-making efficiency and information response ability of enterprise top management. The method can be widely used in manufacturing and other industrial scenarios that rely on big data to drive decision-making, and has strong application value.

Claims

1. An intelligent decision making method for AI Agent based mechanical manufacturing production management characterized in that, Comprising the following steps: Receiving user query requests input through natural language and performing semantic analysis on user input through an intent recognition module to identify query intent and key parameters; Based on the analysis results, using a natural language to structured query language (NL2SQL) model to automatically generate corresponding database query statements; Sending query statements to enterprise backend databases through a data interface access module to obtain structured business data; Automatically processing the obtained data and generating visual charts to display key business indicators; Attribution analysis of indicator fluctuations to identify influencing factors and trigger abnormal pre-warning according to set rules; Generating targeted auxiliary business decision-making suggestions in combination with enterprise rule base and historical data; Supporting multi-round dialogue context management to realize continuous query and condition addition of natural language interaction.

2. The method of claim 1, wherein the intent recognition module employs a deep learning network based on a pre-trained language model for intent classification and slot extraction.

3. The method of claim 1, wherein the NL2SQL model is based on a large language model, combined with enterprise database metadata for semantic mapping and SQL statement generation.

4. The method of claim 1, wherein the data interface access module supports multiple enterprise backend systems, with cache and index optimization mechanisms to improve query performance.

5. The method of claim 1, wherein the attribution analysis includes difference decomposition, regression analysis, and causal reasoning to identify the main influencing factors of key indicator changes.

6. The method of claim 1, wherein the auxiliary business decision-making suggestions are generated based on a rule engine and a historical data model, and improvement suggestions are proposed for key business links such as production capacity, sales, and inventory.

7. The method of claim 1, wherein the multi-round dialogue is supported through a conversation context management module, maintaining user interaction history and supporting continuous condition addition and semantic completion.

8. The method of claim 1, further comprising a permission control subsystem, based on role-based access control policies to limit user data access permissions and perform operation auditing.

Images