An industrial software-oriented cloud intelligent assisted learning system and method

Abstract

The application discloses an industrial software-oriented cloud intelligent auxiliary learning system and method, and belongs to the technical field of industrial software auxiliary. The system comprises the following modules: individualized configuration, self-adaptive picture acquisition and detection, local offline character recognition, remote cloud intelligent analysis and content generation, multi-modal synchronous display, safe operation and data reservation, etc. The method is as follows: individualized configuration is executed; industrial software windows are silently scanned and locked, and software pictures are acquired and change detection is performed; local offline OCR recognition is triggered when changes occur, English scene text is extracted and structured; the encrypted transmission is transmitted to a remote server, a vertical large model professional knowledge base is called for analysis, scene adaptive recognition and RAG intelligent question and answer double branch line logic are adopted to generate auxiliary content and multi-modal display. Through the local light operation and remote heavy operation architecture, the user level adaptation, the intranet offline deployment and the abnormal fault tolerance mechanism are matched, and the intelligent auxiliary learning of the English industrial software is realized.

Description

Technical Field

[0001] This invention relates to the field of industrial software assistance technology, and in particular to a cloud-based intelligent assisted learning system and method for industrial software. Background Technology

[0002] Against the backdrop of industrial digital transformation, foreign industrial software, with its mature technology system and complete functional modules, has long been widely used in core industrial scenarios such as industrial design, simulation analysis, and CNC machining. It is a core tool for production and R&D in enterprises in high-end manufacturing, aerospace, and mechanical engineering. These industrial software programs all use an all-English interface and system, and are characterized by complex functional hierarchies, obscure professional English terminology, rigorous parameter configuration logic, and highly specialized error messages. This makes them extremely difficult to learn for university students, beginners, and front-line engineering technicians.

[0003] Currently, the industry pain points for learning general industrial software mainly focus on three aspects: First, the interface cognition barrier, where users struggle to understand the English semantics of menus, toolbars, and function buttons, making it impossible to complete basic operations; second, the parameter cognition barrier, where process parameters, simulation parameters, etc., involve complex technical terms, and industrial software often involves multiple disciplines, while the English interface, error messages, and labels lack easy-to-understand explanations, easily leading to parameter configuration errors; and third, the troubleshooting barrier, where industrial software errors are related to analysis type, industry, load conditions, boundary conditions, etc., and are not only expressed in technical English, but also often involve difficult basic theories and mechanics, making it difficult for users to quickly locate the cause of the fault and formulate a solution.

[0004] Currently, the mainstream technical solutions in the field of industrial software-assisted learning can be mainly divided into three categories: (1) Static document and fixed tutorial solutions. The content of this type of solution is completely fixed and can only provide general basic operation explanations. It cannot dynamically match the guidance content according to the user's real-time operation scenarios, interface switching, pop-up error and parameter changes. The Q&A teaching is seriously lagging behind and cannot adapt to the complex learning scenarios of multiple categories and versions of English industrial software.

[0005] (2) General OCR translation assistance solutions. For example, Gong Yaling invented an artificial intelligence-based English intelligent learning assistance system (patent number CN202510926756.7), which can only realize the literal translation of English text. It lacks the professional interpretation logic specific to industrial software and cannot interpret the meaning of professional parameters, process principles, and the root causes and solutions of error faults. At the same time, it cannot adapt to the real-time changing scenarios of the software interface. Moreover, it relies on high computing power throughout the process, has no lightweight design, has high local resource consumption, and is prone to interfering with the normal operation of industrial software.

[0006] (3) Cloud-based Q&A solution for general large models. This type of solution requires users to actively copy or screenshot questions to the general large model and answer them based on a general knowledge base. However, it is not very specialized in the industrial field and the output content is not very practical. It relies on high-frequency cloud interaction, resulting in high response latency and high computing power consumption. It also requires continuous uploading of user operation screens and learning data, which poses a risk of leakage of industrial privacy data.

[0007] In summary, existing technical solutions not only fail to achieve intelligentization, but also generally suffer from deficiencies such as insufficient scenario adaptability, low accuracy of professional interpretation, high equipment operating load, lack of personalized adaptation, and insufficient data security and stability. They cannot meet the intelligent assisted learning needs of multi-category English industrial software for high precision, real-time operation, lightweight design, and security. Summary of the Invention

[0008] The purpose of this invention is to provide a cloud-based intelligent assisted learning system and method for industrial software. Relying on lightweight distributed computing, scene adaptive recognition, industry knowledge base matching, dynamic user level adaptation, multimodal intelligent display and security management technologies, it enables real-time, accurate, lightweight and secure learning and operation assistance for multiple categories of English industrial software.

[0009] To achieve the above objectives, the present invention provides a cloud-based intelligent assisted learning system for industrial software, comprising: The personalized configuration module is used to complete the system initialization configuration, generate or read the personalized configuration file stored locally by the user; and during the long-term operation phase, it uses a quantitative algorithm to collect four types of core data in real time: the difficulty of the user's operation scenario, the breadth of software function usage, the learning frequency, and the depth of Q&A, to dynamically determine the user's learning level and complete the level promotion and configuration parameter iterative update. An adaptive image acquisition and detection module is used to carry a software feature recognition engine. The engine has a built-in feature library of English industrial software. It can silently scan and identify and lock the core operation areas of different English industrial software in the background. It continuously acquires the locked software screen and performs differential comparison on the continuously acquired screen frames through the built-in screen change detection and comparison algorithm. It only triggers subsequent recognition calculations when substantial screen changes are identified, and automatically skips the calculations when the screen is still. The local offline text recognition module deploys a lightweight offline OCR recognition engine adapted to the industrial software interface. After receiving a signal triggered by a change in the screen, it performs full-domain offline extraction of English text information in the software and completes text deduplication, classification and normalization, and invalid character removal through post-processing logic, outputting structured English scene text data. The remote cloud-based intelligent parsing and content generation module receives structured English scene text data that has been encrypted and desensitized, and completes scene semantic parsing on the remote server; it calls upon the industrial software professional knowledge base and the semantic inference capabilities of the large model to perform intelligent inference based on the current scene and generate personalized guidance adapted to the real-time work scenario. The multimodal synchronous display module is used to synchronously display the generated auxiliary content through custom floating subtitles, real-time speech synthesis and broadcasting, and dynamic virtual avatar linkage explanation; The secure operation and data retention module adopts a distributed architecture of local light computing and remote heavy computing. The local terminal is only responsible for low computing power operations, while high computing power operations are completed by the remote server. Only anonymized text data is transmitted in encrypted form, user privacy data is only retained locally, and temporary data is destroyed immediately after use.

[0010] Preferably, in the adaptive image acquisition and detection module, the English industrial software-specific feature library includes process features, window title features, and interface layout features; the core operation area includes a main operation window, a parameter setting pop-up window, a fault reporting window, a function menu bar, and a toolbar; the acquisition frequency can be customized and adjusted, and the acquisition process supports dynamic adaptation such as window scaling, movement, and interface switching.

[0011] Preferably, the remote cloud-based intelligent parsing and content generation module has a dual-branch content generation logic: The scene adaptive recognition branch will judge the user's current operation behavior based on real-time scene data. The background completes scene status analysis and scene data pre-caching. It runs silently throughout the process without interfering with the user's normal settings. After the user triggers the display command, the corresponding scene data is retrieved to generate scene-based guidance content. The RAG intelligent question-answering branch, when users manually input learning questions or actively trigger question-answering queries in the visual operation interface, will combine real-time scenario recognition with a layered question-answering mechanism: First, it will retrieve relevant industrial software expert experience content; if no matching expert experience is found, it will supplement the content by retrieving the corresponding English industrial software professional knowledge base through RAG semantic retrieval, recalling relevant knowledge fragments. The system will then feed the user's question, the identified scenario, expert experience materials, and knowledge base retrieved fragments into the LLM large model for inference, generating targeted question-answering content including fault cause analysis, parameter configuration suggestions, and operation method instructions. When neither expert experience nor the knowledge base provides a valid match and cannot meet the question-answering needs, the LLM large model will be activated as a fallback to ensure full coverage of question-answering requirements.

[0012] Preferably, the personalized configuration module has four standardized learning configuration templates built-in: novice, intermediate, proficient, and VIP, and is pre-linked to an English industrial software scenario library, professional interpretation rules, Q&A style, voice parameters, and interface display style; the quantitative algorithm realizes dynamic level promotion and iterative update of configuration parameters by continuously collecting user setting data.

[0013] Preferably, in the secure operation and data retention module, low-computing-power operations include image capture, offline recognition, and content display; high-computing-power operations include semantic parsing, knowledge base matching, content generation, and speech synthesis; locally retained data includes user-personalized configurations, learning records, operation data, and core privacy data; and data that is destroyed immediately includes temporary Q&A data and voice cache data.

[0014] Preferably, it also includes a visual and convenient management and fault tolerance module. The visual and convenient management and fault tolerance module is configured with a visual operation interface, integrating one-click start / stop recognition, voice switch, Q&A query, parameter configuration function entry, and built-in fault tolerance mechanism; it can identify network anomalies, text recognition anomalies, and knowledge base content matching failures, and automatically execute the retry function. If the retry fails, it records the anomaly log, stops the current task, and then issues a warning message to the user, while ensuring the continuous operation of other auxiliary processes.

[0015] This invention also provides a cloud-based intelligent assisted learning method for industrial software, applied to the aforementioned system, comprising the following steps: S1. Perform personalized initialization configuration: Upon first launch, the program generates a personalized configuration file for the user; subsequent launches read the local historical configuration file. The program also initializes and loads the interface, virtual avatar, and subtitle parameters. S2, Adaptive Image Acquisition and Change Detection: The background automatically scans the device's running program and, through a software feature recognition engine equipped with a dedicated feature library for English industrial software, automatically identifies and locks the active English industrial software window, enabling silent real-time image acquisition and image change detection; subsequent recognition calculations are triggered only when a substantial image change is detected, and calculations are automatically skipped when the image is still. S3, Local Offline Text Recognition and Text Shaping: When the screen undergoes substantial changes, the local offline lightweight OCR recognition engine is triggered to extract English text information from the software screen across the entire domain. The post-processing logic is then used to complete text deduplication, classification and shaping, and invalid character removal, outputting structured English scene text data. S4. Data transmission and auxiliary content generation through dual-branch logic: After the structured English scenario text data is encrypted and desensitized, it is transmitted to the remote server. Relying on the English industrial software professional knowledge base and the semantic inference capability of the large model, the dual-branch content generation logic is executed to generate learning guidance or Q&A content adapted to the real-time operation scenario. S5, Multimodal Display Output: The generated auxiliary content is displayed synchronously in a multimodal manner through custom floating subtitles, real-time voice synthesis and broadcasting, and dynamic virtual avatar linkage explanation; S6. Distributed Computing Power Scheduling and Data Security Management: Adopting a local-remote distributed computing power allocation mode, low-computing power operations such as image capture, offline recognition, and content display are performed locally; high-computing power tasks, including semantic parsing, knowledge base matching, content generation, and speech synthesis, are handled by a remote server. The remote server is only responsible for receiving non-privacy intermediate data and returning task calculation results; original captured images and user privacy data are not uploaded to the cloud. Highly confidential privacy data, including user personalized configurations, learning records, operation data, and core privacy data, are stored locally only, and temporary computation data is destroyed immediately after use. Simultaneously, the terminal hardware configuration is assessed; low-spec terminals use an ultra-lightweight mode to reduce image capture frequency and simplify display effects, while high-spec terminals use a full-function mode. An intranet offline execution mode is supported, deploying the industrial software professional knowledge base, lightweight OCR recognition model, and lightweight semantic parsing model on the local terminal. All recognition and semantic parsing operations are completed locally without uploading data to a remote server. User editing commands are received, and knowledge base entries are added, modified, and retained locally. S7. Fault Tolerance: Real-time monitoring of operational anomalies. When network anomalies, text recognition failures, or knowledge base matching failures are detected, automatic retry is performed. If the retry fails, an exception log is recorded, the current task is terminated, and a warning message is sent to the user, while maintaining the continued operation of other business processes.

[0016] Therefore, the cloud-based intelligent assisted learning system and method for industrial software described above have the following beneficial effects: 1) Multi-category adaptation and scenario-based real-time assistance: Through silent background scanning and industrial software feature library recognition, it can be adapted to various English industrial software; combined with screen change detection and dual-branch content generation logic, it can realize operation guidance and Q&A based on the current operation scenario, eliminating the need for users to frequently switch between tutorials and software, significantly improving learning efficiency.

[0017] 2) Distributed architecture balances efficiency and security: It adopts a local light computing-remote heavy computing architecture, which completes image acquisition and offline recognition locally, and performs semantic parsing and content generation remotely. This ensures recognition accuracy and response speed, and effectively prevents the leakage of core enterprise design data through de-identified transmission, local retention of private data, and immediate destruction of temporary data.

[0018] 3) Personalized adaptive learning: By collecting user operation data, the learning level is dynamically determined and the corresponding configuration parameters are automatically matched to achieve differentiated assistance from novice to expert to VIP user, avoiding the assistance content being too simple or too complicated.

[0019] 4) Highly reliable operation and multi-scenario deployment: It has an anomaly tolerance mechanism, automatically retrying when there is network anomaly or identification failure, recording retry failure and alarming to ensure the continuous operation of auxiliary processes; at the same time, it supports the intranet offline deployment mode, which can meet the fully localized operation requirements in high-security scenarios.

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0021] Figure 1 This is the overall flowchart of the cloud-based intelligent assisted learning method for industrial software according to the present invention; Figure 2 This is a flowchart of the personalized initialization and dynamic configuration of learning levels of this invention; Figure 3 This is a flowchart of the adaptive silent acquisition and change detection process for industrial software screens in this invention. Figure 4 This is a flowchart of the local offline text recognition and text normalization process of this invention; Figure 5 This is a flowchart of the data transmission and auxiliary content generation process of the present invention via dual-branch logic; Figure 6 This is a flowchart of the multimodal display output of the present invention; Figure 7 This is a flowchart of the distributed computing power scheduling and data security management process of the present invention. Detailed Implementation

[0022] The technical solutions of the present invention will be further described below with reference to the accompanying drawings and embodiments. Unless otherwise defined, the technical or scientific terms used in this invention should be understood in their ordinary sense by those skilled in the art. The described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0023] Example 1: A Complete Lightweight Distributed Cloud-Based Intelligent Assisted Learning System This embodiment provides an auxiliary learning system for industrial software. Deployed on a user terminal, the system includes a personalized configuration module, an adaptive image acquisition and detection module, a local offline text recognition module, a remote cloud-based intelligent parsing and content generation module, a multimodal synchronous display module, a secure operation and data retention module, and a visual, convenient management and fault tolerance module. The system adopts a distributed architecture of local lightweight computing and remote heavy computing. The local terminal handles low-computing operations such as image acquisition, offline recognition, and content display, while high-computing operations such as semantic parsing, knowledge base matching, content generation, and speech synthesis are performed by a remote server.

[0024] The following is combined Figures 1 to 7 This embodiment provides a detailed description of the system and its complete workflow. Figure 1 This is the overall flowchart.

[0025] S1. Perform personalized initialization configuration.

[0026] like Figure 2 As shown, after the program starts, the personalized configuration module first determines whether it is the first launch. If it is the first launch, the default level template is loaded, which defaults to the standardized learning configuration template for beginners. This template is pre-associated with the English industrial software scenario library, professional interpretation rules, Q&A style, voice parameters, and interface display style to complete the full-dimensional initialization configuration and generate a local configuration file. If it is not the first launch, the local persistent configuration file is read to restore the user's historical configuration parameters, including interface style, virtual avatar parameters, subtitle parameters, learning level, and corresponding configuration.

[0027] After configuration, the main program will run. During long-term operation, the personalized configuration module collects four core data categories in real time using a quantitative algorithm: difficulty of the operation scenario, breadth of software function usage, learning frequency, and depth of Q&A. These four data categories are then weighted and quantitatively scored. When the score reaches a promotion threshold, the user's learning level is automatically upgraded, progressing from novice to intermediate, and from intermediate to proficient. Corresponding configuration parameters are iteratively updated to ensure that the depth, breadth, and presentation style of the supplementary content match the user's current learning level.

[0028] S2, Adaptive image acquisition and change detection.

[0029] like Figure 3 As shown, the adaptive image acquisition and detection module is equipped with a software feature recognition engine. This engine incorporates a feature library specifically for English industrial software, which includes process features, window title features, and interface layout features. The module silently scans all running programs on the device in the background, matching the scanned process names and window titles with the feature library for English industrial software.

[0030] When a match is successful and the target English industrial software window is identified, the core operation area of ​​the software is automatically identified and locked. The core operation area includes the main operation window, parameter setting pop-up window, fault reporting window, function menu bar, and toolbar, and the main window and each sub-area are marked. Then, the software screen is continuously captured at a set frequency. The acquisition frequency can be customized and adjusted by the user, and the acquisition process supports dynamic adaptation for window scaling, movement, and interface switching.

[0031] The system compares the differences between consecutively captured frames and calculates the difference value. When the difference value exceeds a preset threshold, it is determined that a substantial change in the image has been detected. Substantial changes include interface switching, pop-up windows, parameter changes, and error messages. The frame is then marked as a valid change frame and sent to the local offline text recognition module to trigger subsequent recognition operations. When the difference value does not exceed the threshold, the image is considered static, and subsequent recognition operations are automatically skipped, thereby reducing unnecessary computational power consumption.

[0032] S3, local offline text recognition and text straightening.

[0033] like Figure 4 As shown, the local offline text recognition module deploys a lightweight offline OCR recognition engine adapted to the industrial software interface. Upon receiving a screen change trigger signal and a valid change frame, the screen is first divided into multiple image blocks according to sub-regions. Then, the lightweight OCR recognition engine is called to perform full-domain offline extraction of English text information in each image block.

[0034] After extraction, the recognition results are processed through post-processing logic, including: removing invalid characters and redundant information, deduplicating duplicate text, classifying and organizing the text according to its functional area, and finally structuring the processed text into JSON format scene data, outputting standard English scene text, and waiting for encrypted transmission.

[0035] S4. Data is transmitted and auxiliary content is generated through dual-branch logic.

[0036] like Figure 5 As shown, the secure operation and data retention module encrypts and de-identifies the structured English scenario text data before transmitting it to the remote server through a secure channel. The remote cloud-based intelligent parsing and content generation module receives the encrypted and de-identified structured English scenario text data, invokes the English industrial software professional knowledge base and the semantic inference capabilities of the large model, and executes a dual-branch content generation logic.

[0037] Scene Adaptive Recognition Branch: Based on real-time scene data, it judges the user's current operation behavior. The background completes scene status analysis and scene data pre-caching, running silently throughout without interfering with the user's normal operation. After the user triggers the display command through the visual and convenient management and fault tolerance module, the corresponding scene data is retrieved, and scene-based guidance content is generated and output. The guidance content includes an explanation of the current interface functions, explanation of parameter meanings, and operation steps.

[0038] RAG Intelligent Question Answering Branch: When users manually input learning questions or actively trigger question-and-answer query commands on the visual operation interface, a layered question-and-answer mechanism is adopted based on real-time scenario recognition: First, relevant industrial software expert experience content is retrieved. If no matching expert experience is found, semantic retrieval of the corresponding English industrial software professional knowledge base is performed through RAG semantic retrieval to recall relevant knowledge fragments. The system then feeds the user's question, the recognized scenario, expert experience materials, and knowledge base retrieved fragments into the LLM large-scale model for inference, generating targeted question-and-answer content, including fault cause analysis, parameter configuration suggestions, and operation method instructions. When neither expert experience nor the knowledge base provides a valid match and cannot meet the question-and-answer requirements, the LLM large-scale model's independent question-and-answer mechanism is activated as a fallback to ensure full coverage of question-and-answer needs.

[0039] S5, multimodal display output.

[0040] like Figure 6 As shown, the multimodal synchronous display module receives auxiliary content output from the remote cloud intelligent parsing and content generation module and displays it synchronously in three ways: First, custom floating subtitles display text guidance content as a floating window above the software interface, and the position, font size, color, and transparency of the subtitles can be customized by the user; second, real-time speech synthesis broadcasting converts the text content into speech for real-time broadcasting, and the speech speed, timbre, and volume can be adjusted; third, dynamic virtual image linkage explanation, through the virtual image's lip movements, facial expressions, and actions synchronized with the speech content, enhancing the intuitiveness and immersion of the explanation.

[0041] S6, Distributed computing power scheduling and data security management.

[0042] like Figure 7 As shown, the secure operation and data retention module adopts a distributed architecture of local lightweight computing and remote heavy computing. The local terminal is only responsible for low-computing operations such as image capture, offline recognition, and content display; high-computing operations such as semantic parsing, knowledge base matching, content generation, and speech synthesis are handled by the remote server. Only anonymized text data is transmitted to the remote server in encrypted form; the original images are not uploaded. Privacy data such as user-personalized configurations, learning records, operation data, and core privacy data are stored locally only; temporary Q&A data and voice cache data are destroyed immediately after use.

[0043] Meanwhile, the system identifies the terminal hardware configuration and automatically adjusts the operating mode according to the configuration: low-configuration terminals enable the ultra-lightweight mode, reduce the screen capture frequency, turn off the virtual avatar, and simplify the display effect to reduce local resource consumption; high-configuration terminals enable the full-function mode and enable all auxiliary functions.

[0044] The system supports an intranet offline execution mode: In high-security scenarios, a streamlined industrial software professional knowledge base, a lightweight OCR recognition model, and a lightweight semantic parsing model are deployed on a local terminal. All recognition and semantic parsing operations are completed locally, without uploading any data to a remote server. Furthermore, the system receives user editing commands, allowing users to add, modify, and save knowledge base entries locally to adapt to the personalized needs of specific enterprises or software.

[0045] S7. Exception handling.

[0046] The visualized and convenient management and fault-tolerant module features a visual operation interface, integrating functions such as one-click start / stop recognition, voice on / off, Q&A query, and parameter configuration, and includes a built-in fault-tolerant mechanism. The module monitors the system's operating status in real time. When it detects network anomalies, text recognition errors, or knowledge base content matching failures, it automatically retryes. If a retry fails, it logs the error, suspends the current task, and issues a warning message to the user, ensuring the continuous operation of other auxiliary processes and preventing the entire auxiliary system from being interrupted due to a single point of failure.

[0047] Example 2: Real-time user assistance for large-scale general-purpose industrial simulation software For fields such as aerospace, civil engineering, mechanical engineering, and intelligent manufacturing, large-scale general-purpose industrial simulation software is essential for design and development. It uses virtual digital experiments to assess the strength, stiffness, and stability of structures under external loads. However, in learning scenarios involving modeling, simulation analysis, and load application to solve real-world engineering problems using general-purpose industrial simulation software, users often encounter the following problems: first, they struggle to quickly recognize the numerous technical terms in the English interface; second, the underlying technological principles and parameter definitions are difficult to understand; and third, they lack effective means to troubleshoot software errors independently. Meanwhile, traditional learning aids have several shortcomings: static text and image tutorials are outdated and cannot keep pace with the user's real-time work scenarios; general-purpose translation tools only provide literal translations, lacking the professional contextual understanding required for industrial simulation, which can lead to translation errors and affect subsequent setup and analysis.

[0048] To address the above issues, this embodiment is adapted to low-spec office computer hardware environments and targets fields such as aerospace, civil engineering, mechanical engineering, and intelligent manufacturing. It covers users from basic beginners to general users and advanced VIP users with industry knowledge, focusing on modeling, analysis, and setup scenarios for various general-purpose English-interface simulation analysis software (Abaqus, ANSYS, Fluent, Fe-safe, EDEM, Isight, etc.). It provides a low-load, offline-available, real-time-response, and professional process knowledge base-based end-to-end support system. The following uses Abaqus simulation software as a typical application case to illustrate the specific implementation process.

[0049] When a user launches the cloud-based intelligent assisted learning system for the first time, the personalized configuration module automatically matches the beginner-level configuration template, enabling subtitles, floating subtitles, and voice broadcast mode by default, while also activating the standby virtual avatar display function. The adaptive image acquisition and detection module automatically locks onto the simulation software's dedicated processes and window characteristics, completing the local persistent retention of all basic parameters without requiring manual configuration from the user, adapting to the user's introductory scenario with a single click.

[0050] Throughout the entire process of users completing analysis using industrial software, the cloud-based intelligent assisted learning system silently scans the equipment's operation process in the background, accurately identifying core industrial software operation interfaces such as the simulation software's main interface, parameter setting pop-ups, and mesh generation operation windows. It then activates a low-frequency adaptive screen acquisition mode to monitor various screen changes in real time, such as interface switching, parameter clicks, function triggers, and pop-up windows, capturing scene information corresponding to each step of the user's operation and settings.

[0051] Taking the Abaqus software meshing parameter setting interface as a typical application scenario, the adaptive image acquisition and detection module, upon detecting an update to the interface, immediately triggers the local offline text recognition module. This module offline recognizes core professional English parameter labels such as Element Size, Mesh Quality, and Smoothing, along with their accompanying annotations. Through post-processing logic, redundant information such as interface watermarks, blank characters, and invalid icons is automatically removed, transforming scattered text into standardized, structured parameter text data. This recognition logic possesses universal portability and can be quickly adapted to other mainstream simulation-based English industrial software.

[0052] The remote cloud-based intelligent analysis and content generation module receives structured recognition data and, relying on a dedicated knowledge base for simulation analysis enhanced by industry experts, analyzes user settings behavior through scene-adaptive recognition branches to determine the user's operational purpose in configuring grid parameters. It then automatically generates professional explanations of each parameter, configuration ranges of standard parameters, grid optimization solutions and steps, recommended operations, and operational suggestions and guidance to avoid pitfalls and solve practical problems.

[0053] When a user hovers an auxiliary icon or text area requiring assistance, the multimodal synchronous display module presents professional explanations for each parameter using floating Chinese subtitles. The subtitles can be dragged and dropped as needed without obscuring the core operation area. Simultaneously, voice broadcasts key configuration points, and the virtual avatar switches to corresponding explanation actions in sync. The entire process achieves an immersive intelligent learning experience with no interface obstruction, no software lag, and no operational interference.

[0054] After the single-step operation guide ends, the security operation and data retention module immediately destroys temporary recognition data and voice cache data to prevent data residue; user operation and setting process data and operation records are only stored on the local device to protect user privacy and security. At the same time, the personalized configuration module automatically records the difficulty of the user's operation and settings, dynamically scores the user's ability level, and provides data support for subsequent learning, adaptation and optimization.

[0055] Example 3: Cloud-based intelligent user learning assistance for English industrial software used in geometric modeling in the field of intelligent construction. In the field of intelligent construction, establishing geometric models is fundamental and crucial. For users ranging from beginners to advanced levels, the following problems are frequently encountered during 3D modeling operations and learning: firstly, insufficient ability to independently troubleshoot modeling software malfunctions; and secondly, a lack of thorough understanding of the underlying engineering and technological logic. Existing auxiliary learning methods struggle to meet users' advanced learning needs. Furthermore, the solutions provided by general-purpose large models are prone to providing "illusions," lacking depth and practical guidance from a professional perspective, and sometimes even failing to meet the specifications and standards of industrial 3D modeling. Traditional paper-based learning materials or electronic tutorials mostly only explain the main basic operating procedures, lacking scenario-based and personalized troubleshooting solutions, and failing to address core needs such as advanced and difficult problems and problem tracing.

[0056] To address the above issues, this embodiment is adapted to mid-range engineering office equipment and targets users in fields such as intelligent manufacturing. It focuses on modeling operations, fault diagnosis, and troubleshooting scenarios using various English-interface 3D modeling industrial software (UG, Pro / E, Solidworks, CATIA). SolidWorks is used as a typical application example of 3D modeling software, and the execution process is described in detail.

[0057] After the cloud-based intelligent assisted learning system is launched, the personalized configuration module automatically reads the user's historical configuration file stored locally, determines the current user's level through a quantitative algorithm, and then switches to an exclusive operating strategy: automatically loading floating subtitles, standard-speed voice broadcasting, and automatically adopting the corresponding explanation framework, focusing the core output on fault tracing and troubleshooting, recommendations for advanced modeling techniques, model optimization, and breakdown of modeling process logic, accurately aligning with the usage needs of users at each level.

[0058] When a user is building a model, the adaptive image acquisition and detection module will lock the process and window of the target 3D modeling software in the background, continuously monitoring the core operations of the entire process, such as sketching, feature extrusion, assembly constraints, and dimensioning. If the software displays an error message such as "Feature creation failed" or encounters a problem such as an abnormal solution calculation, the module can instantly capture the interface anomaly, automatically start image acquisition, and freeze the moment the fault is triggered and the state of the software interface.

[0059] Relying on the locally deployed offline text recognition module, the cloud-based intelligent assisted learning system will perform offline recognition of all information, including the text in the error pop-up window, the interface parameters of the modeling operation, and the records of the pre-fault operation. Through post-processing logic, it will complete the classification of error categories, the matching of fault association function modules, and filter out interference information such as watermarks, invalid icons, and blank characters, and output a standardized structured fault dataset, providing a reliable data foundation for assisted learning and explanation, source tracing analysis, etc.

[0060] When a user discovers a modeling anomaly, clicking the Q&A trigger button will invoke the auxiliary functions. The cloud-based intelligent analysis and content generation module will combine complete on-site scene information and, through the RAG intelligent Q&A branch, retrieve a dedicated knowledge base for 3D modeling fault diagnosis. It will intelligently pinpoint the root cause of feature creation failures, such as sketch constraint conflicts, dimensions exceeding defined ranges, mutual interference between model features, and parameter association matching errors. It will then output a step-by-step troubleshooting path, directly implementable repair solutions, and methods for preventing and avoiding similar problems. This fault analysis architecture possesses excellent cross-software compatibility and can be simultaneously adapted to internationally recognized English 3D modeling industrial software such as Creo and CATIA.

[0061] Considering the actual needs of users at all levels who pursue high modeling efficiency and fewer problems, the cloud-based content generation module will automatically remove redundant background content and highlight the fault handling operation steps and key parameter correction specifications; the multimodal synchronous display module will highlight the core operation instructions at the top of the screen, ultimately achieving a high-level assisted learning effect with real-time performance, lightweight design, and high accuracy.

[0062] Example 4: Lightweight Cloud-Based Intelligent User Learning Assistance for CNC Machining Software In routine industrial CNC programming production scenarios, when skilled technicians perform CAM CNC programming and toolpath debugging, traditional auxiliary tools generally incur significant computational overhead, easily consuming CNC software resources, interfering with normal programming workflows, and consequently affecting the accuracy of CNC programming and machining parameters. Furthermore, existing auxiliary learning solutions are mostly basic introductory operation guides with limited functionality, failing to meet the advanced production needs of technicians at different skill levels for process optimization, precision fine-tuning, and efficiency improvement. In addition, general-purpose auxiliary learning tools lack industrial-grade data security protection mechanisms, and core programming parameters, workpiece machining processes, and other production data are at risk of leakage, failing to meet the safety and compliance requirements of industrial production.

[0063] To address the above issues, this embodiment is adapted to industrial production equipment and targets technical personnel at all levels using CNC machining software. It focuses on English-language industrial CAM CNC machining software such as UG NX, Mastercam, and PowerMill, covering production scenarios including CNC programming, toolpath parameter configuration, and machining accuracy adjustment. This reduces the computational load on equipment and protects the security of core industrial production data. The following section uses UG NX software as an example to illustrate the specific implementation process.

[0064] After the cloud-based intelligent assisted learning system is launched, the personalized configuration module automatically identifies the user level, automatically completes the initial configuration of personalized parameters, and enables adaptation modes such as subtitle display and fast voice broadcast. At the same time, it is equipped with high-precision professional parameter interpretation logic, focusing on professional content such as process accuracy optimization, tool path optimization, precise debugging of machining parameters, production efficiency improvement, and machining loss control, matching the programming operation needs of industrial mass production scenarios.

[0065] Throughout the entire process of user CNC programming and settings, the adaptive image acquisition and detection module continuously monitors the CNC machining software's operating status in an ultra-low power mode. It abandons the high-load computing method of traditional full-screen image acquisition and adopts a scenario-based precise triggering mechanism. For dynamic changes in core operation interfaces such as toolpath parameter settings, machining accuracy adjustment, spindle speed configuration, and feed rate debugging, it initiates image acquisition and data processing. For static and inactive interfaces, it automatically skips redundant calculations. It achieves an extremely lightweight design from the bottom layer of operation, without occupying the effective computing resources of industrial production equipment or affecting the stable operation of the CNC software.

[0066] Leveraging an optimized local offline text recognition module, the system achieves targeted recognition of English parameters specific to the CNC machining industry. It accurately captures core machining parameter identifiers, default thresholds, and parameter constraints such as Spindle Speed, Feed Rate, and Cut Depth. Through post-processing logic, redundant information is automatically removed from the interface, transforming scattered interface data into standardized, structured, high-precision parameter data, providing accurate data support for subsequent intelligent process analysis. This recognition system is universal, quickly compatible with the parameter systems of multiple mainstream CNC programming software such as Mastercam and PowerMill, and adaptable to various industrial CNC programming scenarios.

[0067] During operation, users can click the Q&A trigger button or ask questions. The remote cloud-based intelligent analysis and content generation module retrieves a complete CNC machining process knowledge base. It combines the user's real-time programming scenario, workpiece machining characteristics, and equipment operating parameters to perform intelligent process analysis and generate advanced guidance content such as optimal parameter configuration ranges, adaptation parameter standards for workpieces of different materials, fine toolpath optimization techniques, machining accuracy fine-tuning schemes, and raw material loss control strategies. This content is tailored to the core operational needs of skilled technicians to optimize machining processes, adjust machining accuracy, and improve production efficiency.

[0068] At the interactive display level, the multimodal synchronous display module is fully adapted to high-intensity, long-term continuous industrial operation scenarios, completely eliminating redundant display effects and retaining only a minimalist display mode with bilingual (Chinese and English) subtitles for core parameters. The subtitle layout does not obscure the core operation area of ​​the software. Simultaneously, an on-demand triggered voice broadcast mechanism is adopted, broadcasting core process points and optimization suggestions only when the user completes key operations such as setting critical parameters and debugging core toolpaths. This eliminates unnecessary interactive interference, interface obstruction, and redundant computational consumption, ensuring that technicians can efficiently and continuously carry out CNC programming and production operations.

[0069] Example 5: Pure Local Deployment Assistance in Industrial Scenarios Without Network Intranet Factory-level industrial work scenarios are mostly based on closed intranet architectures, with terminal devices generally lacking external network support. Traditional industrial software assistance solutions heavily rely on cloud network transmission, failing to start or function offline, and are unsuitable for workshop operating conditions. Furthermore, general-purpose auxiliary tools have poor adaptability to industrial scenarios, lacking industry-specific knowledge bases and unable to accurately identify the professional parameters and process terminology of various industrial software. They also struggle to provide professional, scenario-based operating guidance for English industrial software, hindering the efficiency of core processes such as workshop drawing, workpiece processing, and equipment debugging.

[0070] This embodiment focuses on closed production environments without external network access in factories. It abandons the traditional cloud-dependent auxiliary mode and adopts a purely local lightweight deployment architecture to achieve fully offline, full-featured, and highly secure intelligent auxiliary services for various English industrial software, solving the tool compatibility problem in network-constrained scenarios. Using AutoCAD industrial drawing software as an example application, the offline deployment and implementation process is described in detail.

[0071] Before users begin operations, they can complete the localization of the system on industrial terminal equipment in advance. This involves deploying a dedicated process knowledge base highly adapted to industrial scenarios, an optimized lightweight OCR recognition model, and a local lightweight semantic parsing model all locally to the terminal equipment. All core algorithms and data resources are stored locally, laying a solid foundation for fully offline operation.

[0072] After the user starts the system program, the entire initialization process can be completed without an internet connection. The personalized configuration module directly reads the preset adaptation parameters locally, automatically matches the English interface features and window process parameters of the target CAD software, and autonomously completes a series of initialization operations such as offline scene recognition, intelligent content matching, and software parameter adaptation. The entire process is without cloud interaction and requires no network support, making it suitable for closed intranet operating environments.

[0073] Throughout the user's drawing process, the adaptive image acquisition and detection module silently monitors and captures dynamic changes in the core operation scenarios of the CAD software, including the drawing interface, layer settings interface, annotation parameter interface, and style debugging interface. The local offline text recognition module independently completes the entire process of accurate English text recognition, invalid information removal from the interface, and classification and organization of scattered text locally. Then, it uses a local lightweight semantic parsing model in conjunction with the built-in industrial knowledge base to complete intelligent matching. All image processing, text parsing, and semantic analysis are completed independently on the local terminal, without the need for cloud server data processing, completely eliminating network dependence.

[0074] The remote cloud-based intelligent analysis and content generation module is based on the locally built-in CAD process knowledge base and industrial drawing standard knowledge base. Combined with the user's real-time work scenario, it intelligently generates professional interpretations of layer parameters, industrial annotation standards, in-depth interpretations of drawing instructions, parameter debugging techniques, and guidance on avoiding high-frequency drawing mistakes through dual-branch logic. The entire process has no cloud data transmission, no network latency, and no functional limitations, and is suitable for the normalized operation scenario of the factory's closed intranet.

[0075] At the interactive output level, the multimodal synchronous display module independently completes multimodal auxiliary functions such as rendering Chinese and English parallel floating subtitles, real-time voice synthesis broadcasting, and visual operation guidance display on the local terminal. The display module and operation auxiliary functions support offline independent operation without functional loss or experience degradation, and are fully adapted to the closed production operation environment of the factory without external network.

[0076] Meanwhile, the visualized and convenient management and fault tolerance modules support the independent expansion and iteration of the local knowledge base. Users can independently add various CAD drawing parameters, customized process terms, special annotation specifications, scenario-based work skills, etc. through the visualized configuration operation interface. The system receives user editing instructions and completes the addition, modification and retention of knowledge base entries locally, continuously improving the knowledge base system and adapting to various personalized and customized industrial drawing needs of the factory.

[0077] Therefore, the present invention adopts the above-mentioned cloud-based intelligent assisted learning system and method for industrial software. Through adaptive image acquisition and change detection, local offline text recognition, remote content generation, multimodal synchronous display, and distributed security management, it realizes real-time, accurate, lightweight, and secure assisted learning for English industrial software. It effectively solves the problems of insufficient scene adaptation capability, low accuracy of professional interpretation, high equipment operating load, lack of personalized adaptation, and insufficient data security and stability in the existing technology.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A cloud-based intelligent assisted learning system for industrial software, characterized in that, include: The personalized configuration module is used to complete the system initialization configuration and generate or read personalized configuration files stored locally by the user. In addition, during the long-term operation phase, the system collects four core data categories in real time through quantitative algorithms: the difficulty of user operation scenarios, the breadth of software function usage, the frequency of learning, and the depth of Q&A. This dynamically determines the user's learning level and completes level promotion and configuration parameter iteration updates. An adaptive image acquisition and detection module is used to carry a software feature recognition engine. The engine has a built-in feature library of English industrial software. It can silently scan and identify and lock the core operation areas of different English industrial software in the background. It continuously acquires the locked software screen and performs differential comparison on the continuously acquired screen frames through the built-in screen change detection and comparison algorithm. It only triggers subsequent recognition calculations when substantial screen changes are identified, and automatically skips the calculations when the screen is still. The local offline text recognition module deploys a lightweight offline OCR recognition engine adapted to the industrial software interface. After receiving a signal triggered by a change in the screen, it performs full-domain offline extraction of English text information in the software and completes text deduplication, classification and normalization, and invalid character removal through post-processing logic, outputting structured English scene text data. The remote cloud-based intelligent parsing and content generation module receives structured English scene text data that has been encrypted and desensitized, and completes scene semantic parsing on the remote server; it calls upon the industrial software professional knowledge base and the semantic inference capabilities of the large model to perform intelligent inference based on the current scene and generate personalized guidance adapted to the real-time work scenario. The multimodal synchronous display module is used to synchronously display the generated auxiliary content through custom floating subtitles, real-time speech synthesis and broadcasting, and dynamic virtual avatar linkage explanation; The secure operation and data retention module adopts a distributed architecture of local light computing and remote heavy computing. The local terminal is only responsible for low computing power operations, while high computing power operations are completed by the remote server. Only anonymized text data is transmitted in encrypted form, user privacy data is only retained locally, and temporary data is destroyed immediately after use.

2. The cloud-based intelligent assisted learning system for industrial software according to claim 1, characterized in that, In the adaptive image acquisition and detection module, the English industrial software-specific feature library includes process features, window title features, and interface layout features; the core operation area includes the main operation window, parameter setting pop-up window, fault reporting window, function menu bar, and toolbar; the acquisition frequency can be customized and adjusted, and the acquisition process supports dynamic adaptation of window scaling, movement, and interface switching.

3. The cloud-based intelligent assisted learning system for industrial software according to claim 1, characterized in that, The remote cloud-based intelligent parsing and content generation module has a dual-branch content generation logic: The scene adaptive recognition branch will judge the user's current operation behavior based on real-time scene data. The background completes scene status analysis and scene data pre-caching. It runs silently throughout the process without interfering with the user's normal settings. After the user triggers the display command, the corresponding scene data is retrieved to generate scene-based guidance content. The RAG intelligent question-answering branch, when users manually input learning questions or actively trigger question-answering queries in the visual operation interface, will combine real-time scenario recognition with a layered question-answering mechanism: First, it will retrieve relevant industrial software expert experience content; if no matching expert experience is found, it will supplement the content by retrieving the corresponding English industrial software professional knowledge base through RAG semantic retrieval, recalling relevant knowledge fragments. The system will then feed the user's question, the identified scenario, expert experience materials, and knowledge base retrieved fragments into the LLM large model for inference, generating targeted question-answering content including fault cause analysis, parameter configuration suggestions, and operation method instructions. When neither expert experience nor the knowledge base provides a valid match and cannot meet the question-answering needs, the LLM large model will be activated as a fallback to ensure full coverage of question-answering requirements.

4. The cloud-based intelligent assisted learning system for industrial software according to claim 1, characterized in that, The personalized configuration module includes four standardized learning configuration templates for beginners, intermediate learners, proficient learners, and VIP learners. It is pre-linked with an English industrial software scenario library, professional interpretation rules, Q&A style, voice parameters, and interface display style. The quantitative algorithm continuously collects user setting data to achieve dynamic level advancement and iterative updates of configuration parameters.

5. The cloud-based intelligent assisted learning system for industrial software according to claim 1, characterized in that, In the secure operation and data retention module, low-computing-power operations include image capture, offline recognition, and content display; high-computing-power operations include semantic parsing, knowledge base matching, content generation, and speech synthesis; locally retained data includes user-personalized configurations, learning records, operation data, and core privacy data; and data that is destroyed immediately includes temporary Q&A data and voice cache data.

6. The cloud-based intelligent assisted learning system for industrial software according to claim 1, characterized in that, It also includes a visual and convenient management and fault tolerance module. This module is equipped with a visual operation interface, integrating one-click start / stop recognition, voice switch, Q&A query, and parameter configuration function entry points, and has a built-in fault tolerance mechanism. It can identify network anomalies, text recognition anomalies, and knowledge base content matching failures, and automatically execute the retry function. If the retry fails, it records the anomaly log, stops the current task, and then issues a warning message to the user, while ensuring the continuous operation of other auxiliary processes.

7. A cloud-based intelligent assisted learning method for industrial software, applied to the system described in any one of claims 1-6, characterized in that, Includes the following steps: S1. Perform personalized initialization configuration: Upon first launch, the program generates a personalized configuration file for the user; subsequent launches read the local historical configuration file. The program also initializes and loads the interface, virtual avatar, and subtitle parameters. S2, Adaptive Image Acquisition and Change Detection: The background automatically scans the device's running program and, through a software feature recognition engine equipped with a dedicated feature library for English industrial software, automatically identifies and locks the active English industrial software window, enabling silent real-time image acquisition and image change detection; subsequent recognition calculations are triggered only when a substantial image change is detected, and calculations are automatically skipped when the image is still. S3, Local Offline Text Recognition and Text Shaping: When the screen undergoes substantial changes, the local offline lightweight OCR recognition engine is triggered to extract English text information from the software screen across the entire domain. The post-processing logic is then used to complete text deduplication, classification and shaping, and invalid character removal, outputting structured English scene text data. S4. Data transmission and auxiliary content generation through dual-branch logic: After the structured English scenario text data is encrypted and desensitized, it is transmitted to the remote server. Relying on the English industrial software professional knowledge base and the semantic inference capability of the large model, the dual-branch content generation logic is executed to generate learning guidance or Q&A content adapted to the real-time operation scenario. S5, Multimodal Display Output: The generated auxiliary content is displayed synchronously in a multimodal manner through custom floating subtitles, real-time voice synthesis and broadcasting, and dynamic virtual avatar linkage explanation; S6. Distributed Computing Power Scheduling and Data Security Management: Adopting a local-remote distributed computing power allocation mode, low-computing power operations such as image capture, offline recognition, and content display are performed locally; high-computing power tasks, including semantic parsing, knowledge base matching, content generation, and speech synthesis, are handled by a remote server. The remote server is only responsible for receiving non-privacy intermediate data and returning task calculation results; original captured images and user privacy data are not uploaded to the cloud. Highly confidential privacy data, including user personalized configurations, learning records, operation data, and core privacy data, are stored locally only, and temporary computation data is destroyed immediately after use. Simultaneously, the terminal hardware configuration is assessed; low-spec terminals use an ultra-lightweight mode to reduce image capture frequency and simplify display effects, while high-spec terminals use a full-function mode. An intranet offline execution mode is supported, deploying the industrial software professional knowledge base, lightweight OCR recognition model, and lightweight semantic parsing model on the local terminal, completing all recognition and semantic parsing operations locally without uploading data to a remote server. Receive user editing instructions and complete the addition, modification, and retention of knowledge base entries locally; S7. Fault Tolerance: Real-time monitoring of operational anomalies. When network anomalies, text recognition failures, or knowledge base matching failures are detected, automatic retry is performed. If the retry fails, an exception log is recorded, the current task is terminated, and a warning message is sent to the user, while maintaining the continued operation of other business processes.

Images