Manufacturing industry ERP system and method based on dynamic module loading and anti-noise voice control

Abstract

The invention relates to an industrial intelligent ERP (Enterprise Resource Planning) production management system and a key technology, belongs to the technical field of industrial informatization software, artificial intelligence voice interaction and equipment predictive maintenance, and aims at solving the problems that a traditional industrial ERP system is insufficient in flexibility, workshop voice interaction is difficult, and equipment maintenance is passive and low in efficiency. The core innovation of the invention lies in that: a module hot plug architecture is adopted, so that a user can switch a full-function ERP system / a lightweight chemical single system / a purchase-sale-stock combined system within 5 seconds; according to an anti-noise voice engine, the instruction recognition rate gt in the 90dB workshop environment is determined; 92%, response delay lt; 200 ms; a closed loop is predictively maintained, an equipment health model is called in real time in work order execution, and shutdown loss is reduced by 40%. The problems that a traditional ERP system cannot be flexibly split, the workshop voice interaction rate is low, and equipment maintenance is passively responded are solved.

Description

Technical Field

[0001] This invention relates to the fields of industrial software technology, intelligent interaction technology, and equipment maintenance technology, specifically to a hot-swappable ERP software module system for manufacturing, a method for implementing a workshop noise-resistant voice engine based on this system, and a comprehensive technical solution integrating predictive maintenance functions for equipment. This solution, through dynamic module management, intelligent voice interaction, and data analysis models, enables flexible configuration, efficient interaction, and equipment fault prediction in industrial software systems, and is applicable to information management and intelligent production scenarios in the manufacturing industry. Background Technology

[0002] In traditional industrial ERP systems, the deployment and updates of software modules typically rely on manual configuration or system restarts, resulting in low module loading efficiency, poor configuration flexibility, and frequent user operation interruptions. For example, when production process adjustments require temporarily enabling or disabling modules such as sales and inventory, current technologies often necessitate manual modification of configuration files or a system restart, leading to business continuity disruptions. Furthermore, the complex acoustic environment of a workshop, including equipment operating noise and electromagnetic interference, significantly impacts traditional voice interaction systems, resulting in low command recognition accuracy and failing to meet the real-time interaction needs of industrial settings. In addition, equipment failure prediction generally relies on manual experience or offline data analysis, leading to delayed warnings, high maintenance costs, and an inability to achieve real-time fault prediction and proactive handling.

[0003] While some existing technologies employ dynamic module loading, they lack a mechanism for linked updates to access control and interface routing. Voice interaction systems are often designed for general scenarios and lack noise-resistant algorithms optimized for industrial noise environments. Predictive maintenance solutions generally suffer from insufficient model generalization capabilities and low coupling with business systems. Therefore, there is an urgent need for a comprehensive solution integrating hot-swappable modules, noise-resistant voice interaction, and intelligent predictive maintenance to improve the flexibility, usability, and intelligence of industrial ERP systems. Summary of the Invention

[0004] To address the problems mentioned in the background, this application provides a manufacturing ERP system and method based on dynamic module loading and noise-resistant voice control.

[0005] To achieve the aforementioned objectives, this invention proposes a technical solution for manufacturing ERP based on dynamic module loading and noise-resistant voice control, comprising: I. Hot-swappable Industrial ERP Software Module System This system achieves flexible module configuration and real-time activation through client interaction, microservice calls, and dynamic loading technology, specifically including the following steps: Command reception and parsing: The computer client, web browser and mobile client listen to user input in real time, support voice commands (such as "activate inventory module") or interface triggered actions (such as clicking module switch), and generate module configuration commands after parsing; Microservice module invocation: Based on the instruction content, retrieve the corresponding business module (such as sales management, production scheduling, inventory control, etc.) from the preset microservice repository, and obtain the module code package and configuration parameters through RESTful API or RPC interface; Dynamic loading and system adaptation: The target module is injected into the local runtime environment using a software loader, and the built-in permission tree (defining the user's operation permissions for each module) and API gateway routing table (configuring the data interaction interface between modules) are updated synchronously to ensure that the module functions are seamlessly integrated with the system architecture; Real-time interface refresh: After loading is complete, the system interface dynamically renders the function menu of the activated module and hides the entry of the inactive module. Users can use the new configuration function immediately without restarting the system.

[0006] II. Implementation Method of Workshop Noise-Reducing Voice Engine This engine is integrated into the system of claim 1, and solves the challenges of voice interaction in industrial scenarios through hardware optimization, algorithm processing, and cross-module communication, specifically including: Noise reduction and signal processing: Utilizes single / dual microphone arrays for directional sound pickup, combined with Fourier transform (FFT) and filtering algorithms (such as Wiener filtering) to separate environmental noise (such as mechanical vibration and equipment alarm sounds) from effective speech signals, suppress impulse noise (such as instantaneous impact during motor start-up and shutdown), and improve the signal-to-noise ratio (SNR) to over 20dB. Voice command recognition mechanism: A predefined industrial-specific vocabulary (containing core command words such as "skilled worker", "check inventory", and "report work") is used to build a keyword wake-up engine and feature extraction training based on deep learning, so as to achieve high-precision recognition of short command sentences (accuracy ≥ 95%). Cross-system command transmission: After the voice command is parsed, the structured data is encapsulated through the Model Context Protocol (MCP) and transmitted to the AI ​​big model for semantic understanding and intent matching. Finally, it is parsed into API commands that the system can call (such as "query the October inventory report"), realizing a closed-loop interaction of "user voice → system function".

[0007] III. Predictive Maintenance Integration Approach This method combines equipment status monitoring with intelligent algorithms to achieve early warning of faults and proactive maintenance. The specific process is as follows: Real-time data acquisition: The "Production Work Order" module subscribes to device IoT sensor data (vibration acceleration, bearing temperature, motor current, etc.) via the MQTT protocol, with a sampling frequency ≥1kHz, and supports multi-protocol (Modbus, OPC UA) conversion and data cleaning (noise reduction, outlier filtering). Fault probability modeling: The anomaly detection model based on LSTM (Long Short-Term Memory Network) is called. This model is trained with 1 million historical operating data of injection molding machines (including normal operation and failure samples). The input layer includes time series features (30-minute sliding window) and equipment operating parameters (such as load rate and running time). The output layer predicts the probability of equipment failure (0-1 interval). Intelligent decision-making and execution: When the real-time calculated failure probability exceeds the threshold (85%), the system automatically freezes the associated work orders (prohibiting new operations) and pushes maintenance instructions (including equipment number, failure type prediction, and suggested downtime) to the workshop dashboard via the WebSocket protocol. At the same time, the work order system is triggered to generate a maintenance dispatch order, realizing an automated closed loop of "prediction-early warning-handling".

[0008] Compared with existing traditional ERP systems, the manufacturing ERP system based on dynamic module loading and noise-resistant voice control combines hot-swappable technology to improve system configuration efficiency (module loading time ≤ 2 seconds), a noise-resistant voice engine to reduce command misrecognition rate (≤ 5% in noisy environments), and a predictive maintenance interface method to shorten equipment downtime by more than 40%. It realizes functions such as modular and detachable ERP, full voice control, and fault early warning, significantly enhancing the intelligence level and production collaboration capabilities of industrial ERP systems. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the manufacturing ERP system based on dynamic module loading and noise-resistant voice control, as described in this application.

[0010] Figure 2 This is a schematic diagram illustrating a hot-swappable module process in this application.

[0011] Figure 3 This is a schematic diagram of the noise-resistant voice engine workflow in the factory workshop of this application.

[0012] Figure 4 This is a schematic diagram of the predictive maintenance interface model structure of this application.

[0013] Figure 5 This is a schematic diagram comparing the effects of a full-featured ERP system with a production work order-only ERP system based on the hot-swappable module combination of this application. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and examples, focusing on its logical architecture and applications. It should be understood that the examples described herein are merely illustrative and not intended to limit the scope of the invention.

[0015] like Figure 1 As shown, the workflow and method of a manufacturing ERP system based on dynamic module loading and noise-resistant voice control include: S1: Construct a business coupling module based on a hot-swappable microservice architecture The system's top-level design adopts a modular, hot-swappable microservice architecture, forming highly decoupled business modules. This architecture uses sales, procurement, production, inventory, finance, and reporting modules as core components. Each module achieves independent deployment and dynamic collaboration through standardized interfaces, supporting flexible expansion and reconfiguration of business functions. Inter-module interaction relies on an AI big data model and the MCP protocol for intelligent collaboration. The AI ​​big data model provides data analysis and decision support capabilities, while the MCP protocol ensures the standardization and security of data transmission, thereby constructing a digital management framework covering core business scenarios of the enterprise.

[0016] S2: Deploy a noise reduction engine for voice communication in factory workshops. Between the business coupling module and the underlying execution layer, a factory workshop production voice noise reduction engine is integrated, specifically optimized for voice communication needs in high-noise industrial environments. This engine uses acoustic noise reduction algorithms, speech recognition enhancement technology, and a dynamic environmental noise compensation mechanism to filter out interference signals such as workshop equipment noise and environmental background noise in real time. This ensures clear acquisition and accurate transmission of production instructions and equipment interaction voice, providing reliable voice interaction support for the underlying execution system and improving human-machine collaboration efficiency and the accuracy of production instruction execution.

[0017] S3: Build an industrial ERP production management system and achieve end-to-end integration The final product is an industrial ERP production management system based on hot-swappable modules and noise-resistant voice control. Through a layered architecture, it achieves deep integration of technical capabilities and business functions: the upper-layer microservice architecture ensures flexible expansion and efficient collaboration of business modules; the middle-layer voice noise-resistant engine addresses interaction pain points in industrial environments; and the lower-layer system integrates end-to-end production data, covering core aspects such as sales orders, procurement supply chain, manufacturing, inventory management, financial settlement, and report analysis. The system, through standardized interfaces and intelligent technologies, constructs an integrated solution of "modular business configuration – voice interaction noise-resistant processing – full production process control," helping industrial enterprises achieve digital, intelligent, and flexible upgrades in production management.

[0018] Technical effects: This invention improves system configuration efficiency through hot-swappable module technology (module loading time ≤ 2 seconds), reduces command misrecognition rate through noise-resistant voice engine (≤ 5% in noisy environments), and reduces equipment downtime by more than 40% through predictive maintenance scheme, significantly enhancing the intelligence level and production collaboration capabilities of industrial ERP systems.

[0019] Figure 2 The diagram shown is an example of a hot-swappable module process. The specific logic of the layered architecture design for the business module is described below: In the factory-level business module initialization phase, the business modules for factories A, B, and C are deployed independently: each factory (A, B, C) deploys its own dedicated modules based on its business needs. Factory A requires a "sales module" and a "production module," factory B requires a "procurement module" and a "production module," and factory C requires a "production module" and a "reporting module." This phase achieves initial modularization of business operations, supporting independent operation of factory-level businesses. However, it also presents issues such as redundant construction of similar modules across factories (e.g., the "production module") and data interoperability barriers.

[0020] The interface loader function enables standardized integration of interfaces, providing unified access across factory module interfaces: The interface loader standardizes heterogeneous business modules across factories, establishing unified interface protocols and data interaction specifications. First, it extracts common functional interfaces from each factory module (such as data query and command issuance), shielding underlying implementation differences. Then, it establishes communication bridges between modules, resolving interoperability issues between factory A / B / C modules (e.g., synchronizing supply and demand data between factory B's "procurement module" and factory C's "production module"). Finally, it outputs a standardized interface service catalog, providing a unified access point for upper-layer containerized deployments.

[0021] Containerized deployment and resource scheduling layer construction (supported by Docker technology platform), infrastructure layer virtualization and resource management: The underlying runtime environment is built on Docker container technology, realizing two core functions: "containers" and "resource scheduling". Containerization encapsulation encapsulates standardized business modules integrated by the interface loader into independent container units. Each container contains the complete environment (code, dependencies, configuration) required for the module to run, ensuring consistent deployment across environments. Resource scheduling optimization dynamically allocates computing, storage, and network resources through the resource scheduling module. Based on load balancing algorithms (such as CPU / memory utilization and request concurrency), it enables elastic scaling of container instances, improving the utilization of infrastructure resources.

[0022] In the business restructuring and aggregation phase of microservice architecture, domain-driven microservice decomposition and integration are implemented: On a containerized platform, standardized business modules are restructured into microservices to form specialized service clusters. Module deduplication and decoupling are performed, integrating duplicate modules from various factories (e.g., merging multiple factories' "production modules" into a unified microservice), and splitting them into independent microservices according to business domains (sales, procurement, production, inventory, finance, reporting). Each service focuses on a single function and supports independent deployment and iteration. Communication between microservices is then achieved through an API gateway (implicitly extended by the interface loader function), using event-driven or RESTful interfaces to complete data flow (e.g., an order placed in the "sales module" triggers inventory deduction in the "inventory module") as a service collaboration mechanism. Finally, based on the characteristics of Docker containers, on-demand horizontal scaling of microservices is supported (e.g., dynamic scaling of the "sales module" container instance during promotional periods) to meet the needs of fluctuating business traffic.

[0023] The value of this evolving technical architecture lies in: modular reuse, eliminating redundancy across factory modules through interface loaders, and enabling domain-level functional reuse through microservice architecture; unified technology stack, with Docker containerization shielding differences in the underlying environment and supporting smooth migration of heterogeneous systems; elastic deployment, combining resource scheduling with container technology to build a highly available and easily scalable business support platform; and business agility, with the independent deployment characteristics of microservices shortening iteration cycles and supporting rapid response to changes in market demands. This architecture, through a logical chain of "layered abstraction – interface integration – container empowerment – ​​microservice implementation," achieves the evolution from factory-level business modules to an enterprise-level elastic platform, balancing business standardization with technical flexibility.

[0024] Figure 3 The diagram shows the workflow of a workshop noise-reducing voice engine. The specific architecture logic is described as follows: The application scenario layer focuses on two core application areas: the management scenario addresses the management needs of internal organizational processes, decision support, and resource scheduling, improving operational efficiency and management accuracy through intelligent tools; the production scenario targets production links such as industrial manufacturing and process automation, realizing an intelligent closed loop of equipment interaction, data collection, and production optimization to support the implementation of intelligent manufacturing.

[0025] The terminal layer serves as the human-computer interaction entry point, constructing a multi-terminal adaptation system: terminal forms cover various types of devices such as PCs, mobile devices, tablets, and large screens, meeting the operating habits and interaction needs in different scenarios (e.g., PCs are suitable for complex data management, mobile devices are adapted for mobile inspection, and large screens support visual monitoring); terminal wake-up achieves intelligent activation across terminal devices through a unified protocol, supporting multi-modal wake-up methods such as voice commands, QR code scanning, and physical triggers, ensuring the convenience and consistency of the interaction entry point.

[0026] The noise reduction layer implements multi-level signal optimization processing to address the characteristics of voice data transmission: Fourier transform noise reduction converts the time-domain voice signal into frequency-domain data through Fourier transform, and separates effective voice from background noise based on voiceprint feature modeling to achieve noise suppression at the spectral level; noise suppression is combined with a real-time environmental noise detection algorithm to dynamically adjust noise reduction parameters, minimizing the impact of environmental noise (such as mechanical noise and electromagnetic interference) on subsequent processing while preserving the semantic integrity of the voice, thereby improving signal quality.

[0027] The data processing layer relies on AI cloud servers to build the core computing power hub, realizing intelligent processing of the entire data chain: AI cloud servers provide distributed computing resources and elastic expansion capabilities to support parallel processing of large-scale data and model training, ensuring high availability and low latency response of the system. The NLP module performs basic data preprocessing on the cleaned speech-text data, including word segmentation, part-of-speech tagging, and syntactic analysis, to construct a standardized data format and provide structured input for subsequent parsing. It then integrates multi-domain and multilingual corpora and expands the sample dimension set of the text dataset through data augmentation techniques (such as synonym replacement and scene annotation) to support improved model generalization capabilities. Finally, it combines dialect parsing and deploys dialect feature recognition algorithms to address regional language differences, achieving the mapping and conversion from dialect speech to standard text and overcoming cross-language and cross-accent interaction barriers.

[0028] The text parsing layer standardizes semantic understanding and data interaction through standardized protocols: The MCP large model protocol is based on the self-developed large model communication protocol, which defines structured interaction rules for text data, supports cross-module semantic alignment, context association and multi-turn dialogue logic, and ensures the naturalness of human-computer interaction and the consistency of business logic.

[0029] The database layer constructs an industrial-grade data storage and management foundation: The industrial model database adopts a distributed storage architecture to classify and store core industrial data such as industry knowledge graphs, equipment operating parameters, and business rule models. It supports high-concurrency access, version control, and data security auditing, providing reliable data support and knowledge empowerment for upper-layer applications.

[0030] The value of the architecture design: Through the closed-loop logic of "scenario definition - terminal interaction - signal optimization - intelligent processing - semantic parsing - data accumulation", each layer realizes the seamless connection from user input to business implementation. This not only ensures the flexibility of multimodal interaction, but also enhances the stability and professionalism of industrial-grade scenarios, providing a scalable and highly compatible technical support system for intelligent applications.

[0031] Figure 4 The image shows a predictive maintenance interface model, and its system logic flow is described as follows: Equipment data acquisition utilizes the equipment's sensor modules as the system's front-end data entry point, collecting three core operating parameters of the industrial equipment in real time: vibration data, temperature data, and current data. This data is preprocessed through the sensors' built-in signal conditioning and analog-to-digital conversion modules to form standardized time-series data, providing the raw signal source for subsequent analysis.

[0032] The data transmission layer reliably transmits the collected multi-dimensional device data via the MQTT protocol (Lightweight IoT Communication Protocol). This protocol, based on a publish-subscribe mechanism, features low power consumption and high reliability, ensuring stable data transmission in a network environment. Raw data, after being encapsulated by the protocol, is transmitted to the backend server in the form of a message queue, completing the pre-processing for data access and storage.

[0033] LSTM Anomaly Detection Model Construction: An anomaly detection model based on an LSTM neural network is deployed on the server side. This model is trained using historical equipment operation data to form a dataset, which is used to learn the characteristic patterns under normal equipment operation. The model's input layer receives multi-dimensional time-series data such as vibration, temperature, and current. Through temporal feature extraction and pattern matching in the hidden layers, the model outputs an assessment result of the probability of anomalies in equipment operation.

[0034] Real-time inference and fault probability calculation: Real-time access to device data is preprocessed and then input into a trained LSTM model. Based on the difference analysis between the current state and historical normal patterns, the model calculates and outputs a device fault probability value. This probability value serves as the core basis for system decision-making, reflecting the confidence level that the device's current operating state deviates from the normal range.

[0035] The decision control layer, based on the fault probability calculation results, executes branch logic judgments: when the fault probability is ≤85%, it is determined that the equipment operating status is within the normal fluctuation range or the initial stage of potential risk, and the system maintains the current monitoring strategy, continuously collects data and updates the model input; when the fault probability is >85%, a high-risk warning mechanism is triggered, and the system automatically executes the freeze work order operation, suspends the production task scheduling of related equipment, and avoids the risk from expanding.

[0036] The decision results of the visualization interaction layer are presented visually through the push system dashboard: For devices that trigger high-risk warnings, the dashboard displays the faulty device number, fault probability, abnormal parameter trends, and frozen work order status in real time, providing maintenance personnel with an intuitive risk management interface; the dashboard supports data capture and historical tracing, and can display the entire life cycle operation data of the device to assist the maintenance team in formulating maintenance strategies.

[0037] System logic closed loop description: The entire process forms a complete closed loop of "data acquisition - transmission - analysis - decision-making - feedback". The equipment sensors perceive the operating status in real time, and the data is efficiently transferred through the MQTT protocol. The LSTM model completes the intelligent identification of abnormal states. The decision-making mechanism based on risk threshold drives work order management. Finally, the human-machine collaborative operation and maintenance management is realized through the visualization dashboard, providing technical support for predictive maintenance of industrial equipment.

[0038] Figure 5 The image shows a comparison between a full-featured ERP system with hot-swappable modules and a production order-only ERP system. The description is as follows: Menu structure module logic description: The full-featured ERP model constructs an integrated management framework covering the core business chain of an enterprise. Its menu structure is business process-oriented and integrates six core modules: Sales Management: Enables functions such as customer resource management, order processing, and sales data analysis to support front-end business expansion; Production management: encompasses the entire process of production planning, work order scheduling, and process management, ensuring efficient collaboration in the manufacturing process; Procurement Management: Based on inventory data and production needs, it enables supplier management, purchase order execution, and cost optimization; Inventory management: Real-time monitoring of material inbound and outbound operations, inventory alerts, and supply chain collaboration to ensure maximum inventory turnover; Financial Management: Integrating accounting, cost accounting, fund management, and financial statement generation to build an integrated business and finance system; Report Management: Through multi-dimensional data modeling, generate visual reports such as business analysis, production progress, and financial performance to support decision-making.

[0039] The production work order mode focuses on production execution, streamlining the menu structure to a single "Production Work Order" module, and concentrating on three real-time control scenarios: Status management: Real-time synchronization of work order execution progress, process completion status, and anomaly warnings; Quantity management: Dynamically track material consumption, work-in-process inventory, and finished product delivery data; Equipment monitoring: Access device IoT data to enable operational status monitoring, fault alarms, and OEE (Overall Equipment Effectiveness) analysis.

[0040] This model ensures that on-site personnel can quickly respond to work order execution needs by focusing on key functions, thereby reducing unnecessary operational interference.

[0041] Main workspace module logic description: Full-featured ERP integrated management hub The main workspace of a full-featured ERP system is positioned as a "digital dashboard" for enterprise operations, constructing a multi-level management view through three main components: Comprehensive dashboard: Integrates key performance indicators (KPIs) across the entire business chain, presenting core data such as sales trends, production progress, inventory levels, and financial health in a graphical format, and supports cross-module data correlation analysis; KPI cards: Focus on core business objectives and display key information such as target achievement rate and anomaly warnings in real time in the form of visual cards, so that management can quickly grasp the business situation; To-do items: Integrate task nodes in business processes (such as approval, work order processing, data entry, etc.) and achieve closed-loop management of tasks through an intelligent reminder mechanism to improve collaboration efficiency.

[0042] Real-time operation platform in production work order mode only The main work area in the production work order-only mode focuses on efficient execution on the production floor, building a closed-loop system of "data collection - real-time feedback - rapid response": Work order list: Displays currently active work orders sorted by timeline or priority, supports quick search, status filtering, and work order detail traceability, ensuring visualized management of production tasks; Real-time work reporting panel: Provides a lightweight data entry interface, allowing workers to provide real-time feedback on process completion status through scanning codes, clicking buttons, etc., synchronously updating production progress and triggering subsequent process linkages, achieving a "what you see is what you get" on-site operation experience.

[0043] This design simplifies the interaction hierarchy, shortens the operation path, and adapts to high-frequency, standardized production reporting scenarios.

[0044] AI Functional Module Logical Description Full-featured ERP model intelligent decision-making ecosystem The AI ​​functionality of the full-featured ERP model aims for "end-to-end intelligence" and constructs a multi-scenario voice interaction system: Full voice control: Integrating natural language processing (NLP) technology, it supports users to retrieve data from any module via voice commands (such as "query this month's sales report" and "generate production work order schedule"), execute functional operations (such as approval processes and data filtering), and trigger business processes (such as starting inventory counts and creating purchase requests), achieving "seamless" intelligent office work, especially suitable for high-frequency data query and management decision-making scenarios.

[0045] Lightweight intelligent interaction in production work order mode only AI functionality in the production work order mode focuses on improving efficiency on the production floor, providing a single-scenario voice solution: Voice-only reporting: By using a pre-set standardized instruction set (such as "Work order X started" or "Work order Y completed"), workers can quickly report their work via voice. The system automatically verifies the legality of the instructions and updates the work order status synchronously, reducing manual input errors and optimizing the "paperless" operation experience on the production site. It is suitable for production line scenarios with high requirements for ease of operation and relatively fixed business processes.

[0046] Summary: Differences in Models and Applicable Scenarios The two models differentiate themselves through feature tailoring and scenario focus: Full-featured ERP mode: With "full business coverage and full management integration" as its core, it is suitable for most enterprises to build an end-to-end digital management system and support cross-departmental collaboration and strategic decision-making; Production work order only mode: With the goal of "lightweight production execution and maximum operational efficiency", it is only suitable for small manufacturing enterprises or independent deployment in production workshops, and can quickly realize digital control of the entire work order process and improve on-site efficiency.

[0047] Both retain expansion interfaces through modular design, allowing for flexible upgrades or modular combinations based on the stage of enterprise digital transformation.

[0048] It should be understood that the above-described examples of module combinations are for succinct description and to aid in the explanation of the present invention, and are not intended to limit the invention. The combinations of the above-described technical features can be freely combined as long as they do not contradict each other. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. Hot-swappable ERP software modules for manufacturing • Computer client software, web browser software, and mobile client software receive module configuration commands triggered by user voice or interface; • Call the corresponding business module (i.e., the sales module / production module / inventory module / procurement module / finance module / reporting module in the industrial ERP system) from the microservice repository according to the instructions; • Dynamically inject into the local runtime environment of the software system through the interface loader, and synchronously update the software's built-in permission tree and API gateway routing table; • The software system interface refreshes in real time, displaying only the function menus of the activated modules.

2. A workshop noise-reducing voice engine, as described in claim 1, characterized in that... The voice control module specifically includes: • Single / dual microphone directional sound pickup, combined with Fourier filtering, suppresses environmental and equipment impulse noise in the audio signal acquired by the computer, thereby improving the accuracy of human-computer voice interaction; • Define a software dictionary; voice commands are triggered by pre-trained keywords that wake up the engine (dictionary: {"small worker", "check inventory", "report work"}); • The instructions are delivered through the workshop noise-resistant speech engine → Model Context Protocol (MCP) → AI large model → software system user, providing full-process assistance for software users to communicate with the software system.

3. A predictive maintenance integration method, the computer software system as described in claim 1, characterized in that: The "Production Work Order" module in the software receives real-time IoT (Internet of Things) data from the equipment (vibration / temperature / current). • Utilize an LSTM (Recurrent Temporal Neural Network) anomaly detection model (training data: 1 million historical injection molding machine operation records); • Calculate the probability of equipment failure using a pre-trained dataset; • When the predicted failure probability is >85%, the work order will be automatically frozen and a maintenance instruction will be pushed to the dashboard.

Images